llm name fix

- chat support markdown
- stop button for chat
- configurable for deepagent

.gitignore

@@ -11,4 +11,5 @@ django.log
 .env
 
 frontend/node_modules/
-frontend/dist/
+frontend/dist/
+frontend/.vite

README.md

@@ -2,6 +2,10 @@
 
 这是一个基于FastAPI的聊天API服务，使用OpenAI格式的请求来调用pipeline.invoke方法进行聊天。
 
+## Docker Installation
+
+For production deployment using Docker, see the [Installation Guide](README_INSTALL.md).
+
 ## 安装依赖
 
 ```bash
@@ -140,15 +144,35 @@ npm install
 
 ### Start the `front_apis` server
 
-The frontend talks to the `front_apis` FastAPI service, which by default listens on `http://127.0.0.1:8001`.
+The frontend talks to the `front_apis` FastAPI service, which by default listens on `http://127.0.0.1:8500`.
 
 From the project root:
 
 ```bash
-uvicorn fastapi_server.front_apis:app --reload --host 0.0.0.0 --port 8001
+uvicorn fastapi_server.front_apis:app --reload --host 0.0.0.0 --port 8500
 ```
 
-You can change the URL by setting `VITE_FRONT_API_BASE_URL` in `frontend/.env` (defaults to `http://127.0.0.1:8001`).
+Or run directly:
+```bash
+python fastapi_server/front_apis.py
+```
+
+### Backend run modes
+
+Run whichever backend mode you need from the project root:
+
+```bash
+# admin/control plane only (/v1/... frontend APIs)
+uvicorn fastapi_server.front_apis:app --reload --host 0.0.0.0 --port 8500
+
+# DashScope chat runtime only (/apps/... and /v1/apps/... APIs)
+uvicorn fastapi_server.server_dashscope:app --reload --host 0.0.0.0 --port 8588
+
+# combined mode: one process serves both front_apis + DashScope endpoints
+uvicorn fastapi_server.combined:app --reload --host 0.0.0.0 --port 8500
+```
+
+You can change the URL by setting `VITE_FRONT_API_BASE_URL` in `frontend/.env` (defaults to `/`, i.e. same-origin).
 
 ### Start the development server
 

README_INSTALL.md

@@ -0,0 +1,267 @@
+# Installation Guide
+
+This guide explains how to install and run the LangChain Agent application using Docker.
+
+## Prerequisites
+
+- Docker (version 20.10 or later)
+- Docker Compose (version 2.0 or later, or use `docker compose` command)
+
+## Quick Start
+
+1. **Run the installation script:**
+   ```bash
+   ./scripts/shell_scripts/install.sh
+   ```
+
+   This script will:
+   - Check for required tools (Docker, docker-compose)
+   - Create a `.env` file with default configuration
+   - Build Docker images (or use pre-loaded images)
+   - Start all services (PostgreSQL, Backend API, Nginx)
+
+2. **Access the application:**
+   - Frontend: http://localhost (or http://localhost:80)
+   - Backend API: http://localhost:8500
+   - Database: localhost:5432
+
+## Installation for China / Offline Use
+
+If Docker Hub is slow or inaccessible in your region:
+
+### Option 1: Use Chinese Docker Mirrors
+
+Configure Docker to use Chinese registry mirrors:
+
+```bash
+sudo tee /etc/docker/daemon.json <<EOF
+{
+  "registry-mirrors": [
+    "https://registry.docker-cn.com",
+    "https://mirror.ccsogou.com",
+    "https://docker.1ms.run"
+  ]
+}
+EOF
+sudo systemctl daemon-reload
+sudo systemctl restart docker
+```
+
+Then run `./scripts/shell_scripts/install.sh`
+
+### configuring '.env' and 'frontend/.env'
+```bash
+ALI_API_KEY="API_KEY_FOR_ALI_QWEN"
+ALI_BASE_URL="https://dashscope.aliyuncs.com/compatible-mode/v1"
+POSTGRES_ROOT_PASSWORD="ROOT_PASSOWRD_FOR_DB_IN_CONN_STR - required for installation"
+POSTGRES_PASSWORD="USER_PASSWORD_FOR_DB_CONN_STR - required for installation" - need to be same in POSTGRES_PASSWORD
+CONN_STR="CONNECTION_STRING_TO_DATABASE" # DOCKER PASSWORD
+FAST_AUTH_KEYS="API_KEY_FOR_OTHER_APPLICATIONS_TO_USE_BUILT_PIPELINE"
+DAYTONA_API_KEY="DAYTONA_CONFIG - NOT REQUIRED"
+```
+
+### Option 2: Pre-load Docker Images Offline
+
+1. On a machine with good Docker Hub access, run:
+   ```bash
+   ./scripts/shell_scripts/download_images.sh
+   ```
+   This creates `images.tar` with all required images.
+
+2. Transfer `images.tar` to your target machine.
+
+3. Load the images:
+   ```bash
+   docker load < images.tar
+   ```
+
+4. Run the install script:
+   ```bash
+   ./scripts/shell_scripts/install.sh
+   ```
+
+## Manual Installation
+
+If you prefer to set up manually:
+
+1. **Create environment file:**
+   ```bash
+   cp .env.example .env  # Edit as needed
+   ```
+
+2. **Build and start services:**
+   ```bash
+   cd docker
+   docker compose -f docker-compose.prod.yml up -d --build
+   ```
+
+3. **Check service status:**
+   ```bash
+   cd docker
+   docker compose -f docker-compose.prod.yml ps
+   ```
+
+## Configuration
+
+Edit the `.env` file to customize:
+
+- `POSTGRES_DB`: Database name (default: `ai_conversations`)
+- `POSTGRES_USER`: Database user (default: `myapp_user`)
+- `POSTGRES_PASSWORD`: Database password (default: `secure_password_123`)
+- `POSTGRES_PORT`: PostgreSQL port (default: `5432`)
+- `BACKEND_PORT`: Backend API port (default: `8500`)
+- `FRONTEND_PORT`: Frontend web server port (default: `80`)
+
+## Database Initialization
+
+The database is automatically initialized when the PostgreSQL container starts for the first time. The following SQL scripts are executed in order:
+
+1. `scripts/init_database/00_init_user.sh` - Creates database user and database
+2. `scripts/init_database/create_conv_store.sql` - Creates conversation storage tables
+3. `scripts/init_database/create_prompt_config.sql` - Creates prompt configuration tables
+
+## Service Management
+
+All commands run from the `docker/` directory:
+
+### View logs:
+```bash
+cd docker
+docker compose -f docker-compose.prod.yml logs -f
+
+# Specific service
+docker compose -f docker-compose.prod.yml logs -f backend
+docker compose -f docker-compose.prod.yml logs -f postgres
+docker compose -f docker-compose.prod.yml logs -f nginx
+```
+
+### Stop services:
+```bash
+cd docker
+docker compose -f docker-compose.prod.yml down
+```
+
+### Restart services:
+```bash
+cd docker
+docker compose -f docker-compose.prod.yml restart
+```
+
+### Rebuild after code changes:
+```bash
+cd docker
+docker compose -f docker-compose.prod.yml up -d --build
+```
+
+### Reset database (delete all data):
+```bash
+cd docker
+docker compose -f docker-compose.prod.yml down -v
+docker compose -f docker-compose.prod.yml up -d
+```
+
+## Architecture
+
+The application consists of three main services:
+
+1. **PostgreSQL** (`postgres`): Database server
+   - Stores conversations and prompt configurations
+   - Automatically initializes schema on first run
+
+2. **Backend** (`backend`): FastAPI application
+   - Serves API endpoints at port 8500
+   - Handles agent management and chat endpoints
+   - Connects to PostgreSQL database
+
+3. **Nginx** (`nginx`): Web server
+   - Serves the React frontend (port 80)
+   - Proxies API requests to the backend
+   - Handles static file serving
+
+## Project Structure
+
+```
+langchain-agent/
+├── docker/
+│   ├── docker-compose.prod.yml   # Production compose file
+│   └── Dockerfile.prod           # Backend Docker image
+├── scripts/
+│   ├── shell_scripts/
+│   │   ├── install.sh            # Main installation script
+│   │   └── download_images.sh    # For offline image download
+│   └── init_database/            # Database initialization scripts
+├── frontend/                     # React frontend
+├── configs/                      # Pipeline configurations
+├── nginx.conf                    # Nginx configuration
+└── .env                         # Environment variables
+```
+
+## Troubleshooting
+
+### Database connection issues
+
+If the backend can't connect to the database:
+
+1. Check that PostgreSQL is running:
+   ```bash
+   docker compose -f docker-compose.prod.yml ps postgres
+   ```
+
+2. Verify the connection string in `.env` matches the database configuration
+
+3. Check backend logs:
+   ```bash
+   docker compose -f docker-compose.prod.yml logs backend
+   ```
+
+### Frontend not loading / NetworkError
+
+1. Check nginx logs:
+   ```bash
+   docker compose -f docker-compose.prod.yml logs nginx
+   ```
+
+2. Ensure frontend is built with correct API base URL. The `frontend/.env` file should contain:
+   ```
+   VITE_FRONT_API_BASE_URL=/
+   ```
+   Then rebuild: `docker compose -f docker-compose.prod.yml build backend`
+
+### Port conflicts
+
+If ports are already in use, update the port mappings in `.env`:
+
+```bash
+# Example: use port 5433 for PostgreSQL
+POSTGRES_PORT=5433
+```
+
+## Development
+
+For development, you may want to run services separately:
+
+1. Start only PostgreSQL:
+   ```bash
+   cd docker
+   docker compose -f docker-compose.prod.yml up -d postgres
+   ```
+
+2. Run backend locally:
+   ```bash
+   export CONN_STR="postgresql://myapp_user:secure_password_123@localhost:5432/ai_conversations"
+   python -m uvicorn lang_agent.fastapi_server.combined:app --reload --host 0.0.0.0 --port 8500
+   ```
+
+3. Run frontend locally:
+   ```bash
+   cd frontend
+   npm install
+   npm run dev
+   ```
+
+Note: For local frontend development, create a `.env` file in `frontend/` with:
+   ```
+   VITE_FRONT_API_BASE_URL=http://localhost:8500
+   ```
+
+

configs/pipeline_registry.json

@@ -1,24 +1,31 @@
 {
-  "pipelines": {
-    "xiaozhan": {
-      "enabled": true,
-      "config_file": "configs/pipelines/xiaozhan.yaml"
+    "pipelines": {
+        "xiaozhan": {
+            "enabled": true,
+            "config_file": "configs/pipelines/xiaozhan.yaml",
+            "graph_id": "routing",
+            "overrides": {
+                "llm_name": "qwen-plus"
+            }
+        },
+        "blueberry": {
+            "enabled": true,
+            "config_file": "configs/pipelines/blueberry.yaml",
+            "graph_id": "react",
+            "overrides": {
+                "llm_name": "qwen-plus"
+            }
+        }
     },
-    "blueberry": {
-      "enabled": true,
-      "config_file": "configs/pipelines/blueberry.yaml"
+    "api_keys": {
+        "sk-6c7091e6a95f404efb2ec30e8f51b897626d670375cdf822d78262f24ab12367": {
+            "example-key-1": {
+                "default_route_id": "default",
+                "allowed_route_ids": [
+                    "xiaozhan",
+                    "blueberry"
+                ]
+            }
+        }
     }
-  },
-  "api_keys": {
-    "sk-6c7091e6a95f404efb2ec30e8f51b897626d670375cdf822d78262f24ab12367": {
-      "example-key-1": {
-        "default_route_id": "default",
-        "allowed_route_ids": [
-          "xiaozhan",
-          "blueberry"
-        ]
-      }
-    }
-  }
-}
-
+}

configs/prompts/deepagent.txt

@@ -1,3 +1,7 @@
-you are a helpful bot enhanced with skills.
+You are a helpful bot enhanced with skills.
+
 To use a skill, read its SKILL.md file using the read_file tool. Skills are NOT tools — they are instructions for using existing tools.
-Skills with available="false" need dependencies installed first - you can try installing them with apt/brew. You can check if the environment the packages you need.
+
+When using a skill, assume required tools (e.g., npx, curl) are available and execute the commands directly. If a command fails because a tool is missing, install the missing dependency using apt/brew and retry.
+
+For shell commands (e.g., npx, curl), use the execute tool to run them.

docker/Dockerfile.frontend

@@ -0,0 +1,19 @@
+FROM node:20-alpine
+
+WORKDIR /app
+
+RUN npm config set registry https://registry.npmmirror.com
+
+# Build-time API base for Vite (must be set before npm run build).
+ARG VITE_FRONT_API_BASE_URL=/
+ENV VITE_FRONT_API_BASE_URL=${VITE_FRONT_API_BASE_URL}
+
+COPY package*.json ./
+RUN npm install
+
+COPY . .
+RUN npm run build && \
+    mkdir -p /opt/frontend_dist && \
+    cp -r dist/. /opt/frontend_dist/
+
+CMD ["sh", "-c", "rm -rf /app/dist/* && cp -r /opt/frontend_dist/. /app/dist && ls /app/dist"]

docker/Dockerfile.prod

@@ -0,0 +1,93 @@
+# Multi-stage Dockerfile for production deployment
+# Stage 1: Build frontend
+FROM node:20-alpine AS frontend-builder
+
+WORKDIR /app/frontend
+
+# Copy frontend files
+COPY frontend/package*.json ./
+RUN npm ci
+
+COPY frontend/ ./
+RUN npm run build
+
+# Stage 2: Python backend
+FROM python:3.12-slim
+
+WORKDIR /app
+
+# Install system dependencies
+RUN set -eux; \
+    for source_file in /etc/apt/sources.list /etc/apt/sources.list.d/*.list /etc/apt/sources.list.d/*.sources; do \
+        if [ -f "$source_file" ]; then \
+            sed -i 's|deb.debian.org|mirrors.aliyun.com|g' "$source_file"; \
+            sed -i 's|security.debian.org|mirrors.aliyun.com|g' "$source_file"; \
+        fi; \
+    done; \
+    apt-get update; \
+    apt-get install -y --no-install-recommends \
+        postgresql-client \
+        curl; \
+    rm -rf /var/lib/apt/lists/*
+
+# Copy Python dependencies
+COPY pyproject.toml ./
+RUN pip install --no-cache-dir --upgrade pip -i https://pypi.tuna.tsinghua.edu.cn/simple && \
+    pip install --no-cache-dir -e . -i https://pypi.tuna.tsinghua.edu.cn/simple
+
+# Copy application code
+COPY lang_agent/ ./lang_agent/
+COPY configs/ ./configs/
+COPY scripts/ ./scripts/
+COPY assets/ ./assets/
+COPY static/ ./static/
+
+# Copy built frontend from stage 1
+COPY --from=frontend-builder /app/frontend/dist ./frontend/dist
+
+# Set environment variables
+ENV PYTHONPATH=/app
+ENV PYTHONUNBUFFERED=1
+
+# Expose port
+EXPOSE 8500
+
+# Health check
+HEALTHCHECK --interval=30s --timeout=10s --start-period=40s --retries=3 \
+    CMD curl -f http://localhost:8500/health || exit 1
+
+# Create entrypoint script to wait for DB
+# Uses Python to check database connection (more reliable than psql)
+RUN echo '#!/bin/bash\n\
+set -e\n\
+echo "Waiting for database to be ready..."\n\
+python3 << EOF\n\
+import sys\n\
+import time\n\
+import psycopg\n\
+\n\
+max_attempts = 30\n\
+conn_str = "${CONN_STR}"\n\
+\n\
+for i in range(max_attempts):\n\
+    try:\n\
+        with psycopg.connect(conn_str, connect_timeout=2) as conn:\n\
+            with conn.cursor() as cur:\n\
+                cur.execute("SELECT 1")\n\
+        print("Database is ready!")\n\
+        sys.exit(0)\n\
+    except Exception as e:\n\
+        if i == max_attempts - 1:\n\
+            print(f"Warning: Database not ready after {max_attempts * 2} seconds, continuing anyway...")\n\
+            print(f"Error: {e}")\n\
+            sys.exit(0)\n\
+        print(f"Database is unavailable - sleeping (attempt {i+1}/{max_attempts})")\n\
+        time.sleep(2)\n\
+EOF\n\
+exec "$@"' > /entrypoint.sh && chmod +x /entrypoint.sh
+
+ENTRYPOINT ["/entrypoint.sh"]
+
+# Run the combined server
+CMD ["python", "-m", "uvicorn", "lang_agent.fastapi_server.combined:app", "--host", "0.0.0.0", "--port", "8500"]
+

docker/docker-compose.prod.yml

@@ -0,0 +1,100 @@
+version: '3.8'
+
+services:
+  # PostgreSQL database
+  postgres:
+    image: postgres:16-alpine
+    container_name: langchain-agent-db
+    networks:
+      - app-network
+    environment:
+      POSTGRES_DB: postgres
+      POSTGRES_USER: postgres
+      POSTGRES_PASSWORD: ${POSTGRES_ROOT_PASSWORD:-postgres_root_password}
+      # These are used by init scripts to create the app database
+      APP_DB_NAME: ${POSTGRES_DB:-ai_conversations}
+      APP_DB_USER: ${POSTGRES_USER:-myapp_user}
+      APP_DB_PASSWORD: ${POSTGRES_PASSWORD:-secure_password_123}
+    volumes:
+      - postgres_data:/var/lib/postgresql/data
+      - ../scripts/init_database:/docker-entrypoint-initdb.d
+    ports:
+      - "${POSTGRES_PORT:-5434}:5432"
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U postgres"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+    restart: no #unless-stopped
+
+  # Backend API server
+  backend:
+    build:
+      context: ..
+      dockerfile: docker/Dockerfile.prod
+    container_name: langchain-agent-backend
+    environment:
+      - PYTHONPATH=/app
+      - PYTHONUNBUFFERED=1
+      - CONN_STR=postgresql://${POSTGRES_USER:-myapp_user}:${POSTGRES_PASSWORD:-secure_password_123}@postgres:5432/${POSTGRES_DB:-ai_conversations}
+      - POSTGRES_USER=${POSTGRES_USER:-myapp_user}
+      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD:-secure_password_123}
+      - POSTGRES_DB=${POSTGRES_DB:-ai_conversations}
+    ports:
+      - "${BACKEND_PORT:-8500}:8500"
+    volumes:
+      - ../configs:/app/configs
+      - ../scripts:/app/scripts
+      - ../assets:/app/assets
+      - ../static:/app/static
+    networks:
+      - app-network
+    depends_on:
+      postgres:
+        condition: service_healthy
+    restart: no #unless-stopped
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8500/health"]
+      interval: 30s
+      timeout: 10s
+      retries: 3
+      start_period: 40s
+
+  # Frontend build service
+  frontend:
+    build:
+      context: ../frontend
+      dockerfile: ../docker/Dockerfile.frontend
+      args:
+        VITE_FRONT_API_BASE_URL: ${VITE_FRONT_API_BASE_URL:-/}
+    volumes:
+      - frontend_dist:/app/dist
+    networks:
+      - app-network
+
+  # Nginx for serving frontend (optional - can also serve via FastAPI)
+  nginx:
+    image: nginx:alpine
+    container_name: langchain-agent-nginx
+    networks:
+      - app-network
+    ports:
+      - "${FRONTEND_PORT:-8080}:80"
+    volumes:
+      - ../nginx.conf:/etc/nginx/nginx.conf:ro
+      - frontend_dist:/usr/share/nginx/html:ro
+    depends_on:
+      frontend:
+        condition: service_completed_successfully
+      backend:
+        condition: service_started
+    restart: no #unless-stopped
+
+volumes:
+  postgres_data:
+  frontend_dist:
+
+networks:
+  app-network:
+    driver: bridge
+

frontend/package.json

@@ -11,7 +11,9 @@
   },
   "dependencies": {
     "react": "^18.3.1",
-    "react-dom": "^18.3.1"
+    "react-dom": "^18.3.1",
+    "react-markdown": "^10.1.0",
+    "remark-gfm": "^4.0.1"
   },
   "devDependencies": {
     "@types/react": "^18.3.20",

frontend/src/api/frontApis.test.ts

@@ -0,0 +1,22 @@
+import { describe, expect, it } from "vitest";
+
+import { joinApiUrl } from "./frontApis";
+
+describe("joinApiUrl", () => {
+  it("keeps same-origin paths when base url is slash", () => {
+    expect(joinApiUrl("/", "/v1/pipelines")).toBe("/v1/pipelines");
+  });
+
+  it("joins absolute host and trims trailing slash", () => {
+    expect(joinApiUrl("http://127.0.0.1:8500/", "/v1/pipelines")).toBe(
+      "http://127.0.0.1:8500/v1/pipelines"
+    );
+  });
+
+  it("accepts path without leading slash", () => {
+    expect(joinApiUrl("http://127.0.0.1:8500", "v1/pipelines")).toBe(
+      "http://127.0.0.1:8500/v1/pipelines"
+    );
+  });
+});
+

frontend/src/api/frontApis.ts

@@ -1,23 +1,48 @@
 import type {
   AvailableGraphsResponse,
+  ConversationListItem,
+  ConversationMessageItem,
   GraphConfigListResponse,
   GraphConfigReadResponse,
   GraphConfigUpsertRequest,
   GraphConfigUpsertResponse,
+  McpAvailableToolsResponse,
   McpToolConfigResponse,
   McpToolConfigUpdateRequest,
   McpToolConfigUpdateResponse,
   PipelineCreateRequest,
   PipelineCreateResponse,
+  PipelineConversationListResponse,
+  PipelineConversationMessagesResponse,
   PipelineListResponse,
   PipelineStopResponse,
+  RuntimeAuthInfoResponse,
 } from "../types";
 
-const API_BASE_URL =
-  import.meta.env.VITE_FRONT_API_BASE_URL?.trim() || "http://127.0.0.1:8001";
+const API_BASE_URL = import.meta.env.VITE_FRONT_API_BASE_URL?.trim() || "/";
+
+export function joinApiUrl(baseUrl: string, path: string): string {
+  const normalizedPath = path.startsWith("/") ? path : `/${path}`;
+  const normalizedBase = baseUrl.trim();
+
+  // "/" is commonly used in Docker+nginx builds and should resolve as same-origin.
+  if (!normalizedBase || normalizedBase === "/") {
+    return normalizedPath;
+  }
+
+  return `${normalizedBase.replace(/\/+$/, "")}${normalizedPath}`;
+}
+
+// Log which backend the frontend is targeting on startup, with file + line hint.
+// This runs once when the module is loaded.
+// eslint-disable-next-line no-console
+console.info(
+  `[frontend] Using FRONT_API_BASE_URL=${API_BASE_URL} (src/api/frontApis.ts:16)`
+);
 
 async function fetchJson<T>(path: string, init?: RequestInit): Promise<T> {
-  const response = await fetch(`${API_BASE_URL}${path}`, {
+  const url = joinApiUrl(API_BASE_URL, path);
+  const response = await fetch(url, {
     headers: {
       "Content-Type": "application/json",
       ...(init?.headers || {}),
@@ -36,7 +61,24 @@ async function fetchJson<T>(path: string, init?: RequestInit): Promise<T> {
     }
     throw new Error(message);
   }
-  return (await response.json()) as T;
+
+  if (response.status === 204) {
+    return undefined as T;
+  }
+
+  const bodyText = await response.text();
+  if (!bodyText.trim()) {
+    return undefined as T;
+  }
+
+  try {
+    return JSON.parse(bodyText) as T;
+  } catch {
+    const preview = bodyText.slice(0, 160).replace(/\s+/g, " ").trim();
+    throw new Error(
+      `Expected JSON response from ${url}, but received non-JSON content: ${preview || "<empty>"}`
+    );
+  }
 }
 
 export function listAvailableGraphs(): Promise<AvailableGraphsResponse> {
@@ -107,6 +149,10 @@ export function updateMcpToolConfig(
   });
 }
 
+export function listMcpAvailableTools(): Promise<McpAvailableToolsResponse> {
+  return fetchJson("/v1/tool-configs/mcp/tools");
+}
+
 export function createPipeline(
   payload: PipelineCreateRequest
 ): Promise<PipelineCreateResponse> {
@@ -126,3 +172,138 @@ export function stopPipeline(pipelineId: string): Promise<PipelineStopResponse>
   });
 }
 
+export function getRuntimeAuthInfo(): Promise<RuntimeAuthInfoResponse> {
+  return fetchJson("/v1/runtime-auth");
+}
+
+export async function listPipelineConversations(
+  pipelineId: string,
+  limit = 100
+): Promise<ConversationListItem[]> {
+  const response = await fetchJson<PipelineConversationListResponse>(
+    `/v1/pipelines/${encodeURIComponent(pipelineId)}/conversations?limit=${limit}`
+  );
+  return response.items || [];
+}
+
+export async function getPipelineConversationMessages(
+  pipelineId: string,
+  conversationId: string
+): Promise<ConversationMessageItem[]> {
+  const response = await fetchJson<PipelineConversationMessagesResponse>(
+    `/v1/pipelines/${encodeURIComponent(pipelineId)}/conversations/${encodeURIComponent(conversationId)}/messages`
+  );
+  return response.items || [];
+}
+
+type StreamAgentChatOptions = {
+  appId: string;
+  sessionId: string;
+  apiKey: string;
+  message: string;
+  onText: (text: string) => void;
+  signal?: AbortSignal;
+};
+
+function parseErrorDetail(payload: unknown): string | null {
+  if (!payload || typeof payload !== "object") {
+    return null;
+  }
+  const detail = (payload as { detail?: unknown }).detail;
+  return typeof detail === "string" && detail.trim() ? detail : null;
+}
+
+export async function streamAgentChatResponse(
+  options: StreamAgentChatOptions
+): Promise<string> {
+  const { appId, sessionId, apiKey, message, onText, signal } = options;
+  const response = await fetch(
+    joinApiUrl(
+      API_BASE_URL,
+      `/v1/apps/${encodeURIComponent(appId)}/sessions/${encodeURIComponent(sessionId)}/responses`
+    ),
+    {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/json",
+        Authorization: `Bearer ${apiKey}`,
+      },
+      body: JSON.stringify({
+        messages: [{ role: "user", content: message }],
+        stream: true,
+      }),
+      signal,
+    }
+  );
+
+  if (!response.ok) {
+    let messageText = `Request failed (${response.status})`;
+    try {
+      const payload = (await response.json()) as unknown;
+      const detail = parseErrorDetail(payload);
+      if (detail) {
+        messageText = detail;
+      }
+    } catch {
+      // Keep fallback status-based message.
+    }
+    throw new Error(messageText);
+  }
+
+  if (!response.body) {
+    throw new Error("Streaming response is not available.");
+  }
+
+  const reader = response.body.getReader();
+  const decoder = new TextDecoder();
+  let buffered = "";
+  let latestText = "";
+
+  while (true) {
+    if (signal?.aborted) {
+      reader.cancel();
+      throw new Error("Request cancelled");
+    }
+    const { value, done } = await reader.read();
+    if (done) {
+      break;
+    }
+    buffered += decoder.decode(value, { stream: true });
+
+    let splitIndex = buffered.indexOf("\n\n");
+    while (splitIndex >= 0) {
+      const eventBlock = buffered.slice(0, splitIndex);
+      buffered = buffered.slice(splitIndex + 2);
+      splitIndex = buffered.indexOf("\n\n");
+
+      const lines = eventBlock.split("\n");
+      for (const rawLine of lines) {
+        const line = rawLine.trim();
+        if (!line.startsWith("data:")) {
+          continue;
+        }
+        const payloadRaw = line.slice(5).trim();
+        if (!payloadRaw) {
+          continue;
+        }
+        let payload: unknown;
+        try {
+          payload = JSON.parse(payloadRaw);
+        } catch {
+          continue;
+        }
+        const nextText =
+          typeof (payload as { output?: { text?: unknown } })?.output?.text === "string"
+            ? ((payload as { output: { text: string } }).output.text as string)
+            : "";
+        if (nextText !== latestText) {
+          latestText = nextText;
+          onText(latestText);
+        }
+      }
+    }
+  }
+
+  return latestText;
+}
+

frontend/src/styles.css

@@ -65,6 +65,12 @@ button:disabled {
   gap: 8px;
 }
 
+.agent-item-row {
+  display: grid;
+  gap: 6px;
+  grid-template-columns: 1fr auto;
+}
+
 .agent-item {
   align-items: flex-start;
   display: flex;
@@ -73,6 +79,33 @@ button:disabled {
   width: 100%;
 }
 
+.agent-item-title {
+  align-items: center;
+  display: flex;
+  gap: 8px;
+  justify-content: space-between;
+  width: 100%;
+}
+
+.agent-status-pill {
+  border-radius: 999px;
+  font-size: 11px;
+  font-weight: 600;
+  padding: 2px 8px;
+}
+
+.agent-status-pill.running {
+  background: #dff7e7;
+  border: 1px solid #8cd3a1;
+  color: #1a6b35;
+}
+
+.agent-status-pill.stopped {
+  background: #f2f4f7;
+  border: 1px solid #d1d8e0;
+  color: #4a5565;
+}
+
 .agent-item.selected {
   border-color: #4d7ef3;
   background: #edf3ff;
@@ -82,6 +115,11 @@ button:disabled {
   color: #5f6f82;
 }
 
+.agent-chat-button {
+  align-self: stretch;
+  min-width: 64px;
+}
+
 .content {
   padding: 20px;
 }
@@ -186,6 +224,7 @@ button:disabled {
 }
 
 .prompt-section,
+.file-backend-config,
 .run-info {
   border: 1px solid #dbe2ea;
   border-radius: 10px;
@@ -194,10 +233,37 @@ button:disabled {
 }
 
 .prompt-section h3,
+.file-backend-config h3,
 .run-info h3 {
   margin-top: 0;
 }
 
+.run-info-header {
+  align-items: center;
+  display: flex;
+  justify-content: space-between;
+  gap: 8px;
+}
+
+.runtime-badge {
+  border-radius: 999px;
+  font-size: 12px;
+  font-weight: 600;
+  padding: 4px 10px;
+}
+
+.runtime-badge.running {
+  background: #dff7e7;
+  border: 1px solid #8cd3a1;
+  color: #1a6b35;
+}
+
+.runtime-badge.stopped {
+  background: #f2f4f7;
+  border: 1px solid #d1d8e0;
+  color: #4a5565;
+}
+
 .graph-arch-section {
   border: 1px solid #dbe2ea;
   border-radius: 10px;
@@ -229,6 +295,166 @@ button:disabled {
   padding: 10px;
 }
 
+.run-card-columns {
+  display: grid;
+  gap: 12px;
+  grid-template-columns: minmax(280px, 1fr) minmax(420px, 1.3fr);
+}
+
+.run-card-left,
+.run-card-right {
+  display: flex;
+  flex-direction: column;
+  gap: 6px;
+}
+
+.run-card-right {
+  border-left: 1px solid #dbe2ea;
+  min-width: 0;
+  padding-left: 12px;
+}
+
+.run-card-right code {
+  display: inline-block;
+  max-width: 100%;
+  overflow-x: auto;
+  vertical-align: middle;
+  white-space: nowrap;
+}
+
+.discussion-section {
+  background: #f7fbff;
+  border: 1px solid #d7e6f6;
+  border-radius: 10px;
+  padding: 12px;
+}
+
+.discussion-header {
+  align-items: center;
+  display: flex;
+  justify-content: space-between;
+  gap: 12px;
+}
+
+.discussion-header h3 {
+  margin: 0;
+}
+
+.discussion-layout {
+  display: grid;
+  gap: 12px;
+  grid-template-columns: minmax(260px, 320px) 1fr;
+  margin-top: 10px;
+}
+
+.discussion-list {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  max-height: 65vh;
+  overflow-y: auto;
+}
+
+.discussion-item {
+  align-items: flex-start;
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+  text-align: left;
+  width: 100%;
+}
+
+.discussion-item.selected {
+  background: #edf3ff;
+  border-color: #4d7ef3;
+}
+
+.discussion-item small {
+  color: #687788;
+}
+
+.discussion-thread {
+  border: 1px solid #d7e6f6;
+  border-radius: 10px;
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  max-height: 65vh;
+  overflow-y: auto;
+  padding: 10px;
+}
+
+.discussion-message {
+  background: #fff;
+  border: 1px solid #dbe2ea;
+  border-radius: 8px;
+  padding: 8px;
+}
+
+.discussion-message.human {
+  border-left: 3px solid #4d7ef3;
+}
+
+.discussion-message.ai {
+  border-left: 3px solid #26a269;
+}
+
+.discussion-message.tool {
+  border-left: 3px solid #8e6bd8;
+}
+
+.discussion-message-meta {
+  align-items: baseline;
+  display: flex;
+  gap: 8px;
+}
+
+.discussion-message pre {
+  font-family: ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, monospace;
+  margin: 8px 0 0;
+  overflow-x: auto;
+  white-space: pre;
+}
+
+.discussion-message-markdown > :first-child {
+  margin-top: 0;
+}
+
+.discussion-message-markdown > :last-child {
+  margin-bottom: 0;
+}
+
+.discussion-message-markdown code {
+  background: #f3f5f8;
+  border-radius: 4px;
+  padding: 1px 4px;
+}
+
+.discussion-message-markdown pre code {
+  background: transparent;
+  padding: 0;
+}
+
+.discussion-message-markdown a {
+  color: #1a4fc5;
+  text-decoration: underline;
+}
+
+.discussion-message-markdown p,
+.discussion-message-markdown ul,
+.discussion-message-markdown ol,
+.discussion-message-markdown blockquote,
+.discussion-message-markdown table {
+  margin: 8px 0;
+}
+
+.discussion-message-markdown blockquote {
+  border-left: 3px solid #d0d8e2;
+  color: #4f5f73;
+  margin-left: 0;
+  padding-left: 10px;
+}
+
 .mcp-config-section {
   background: #f7fbff;
   border: 1px solid #d7e6f6;
@@ -258,8 +484,174 @@ button:disabled {
   width: 100%;
 }
 
+.mcp-entry-list {
+  display: grid;
+  gap: 12px;
+  margin-top: 10px;
+}
+
+.mcp-tools-error {
+  color: #a33434;
+  margin: 6px 0 0 0;
+}
+
+.mcp-tools-inline {
+  background: #f8fbff;
+  border: 1px solid #d7e6f6;
+  border-radius: 8px;
+  margin: 0 0 10px 0;
+  padding: 8px;
+}
+
+.mcp-entry-card {
+  background: #fff;
+  border: 1px solid #d7e6f6;
+  border-radius: 10px;
+  padding: 10px;
+}
+
+.mcp-entry-header {
+  align-items: center;
+  display: flex;
+  justify-content: space-between;
+  gap: 10px;
+  margin-bottom: 10px;
+}
+
+.mcp-entry-grid {
+  display: grid;
+  gap: 10px;
+  grid-template-columns: repeat(2, minmax(200px, 1fr));
+}
+
+.mcp-entry-grid label {
+  display: flex;
+  flex-direction: column;
+  font-size: 14px;
+  gap: 6px;
+}
+
+.mcp-entry-grid input,
+.mcp-entry-grid select {
+  border: 1px solid #c9d4e2;
+  border-radius: 8px;
+  font-size: 14px;
+  padding: 8px;
+}
+
+.mcp-entry-wide {
+  grid-column: 1 / -1;
+}
+
 .empty {
   color: #687788;
   margin: 6px 0;
 }
 
+.chat-modal-overlay {
+  align-items: center;
+  background: rgba(16, 24, 40, 0.45);
+  display: flex;
+  inset: 0;
+  justify-content: center;
+  position: fixed;
+  z-index: 20;
+}
+
+.chat-modal {
+  background: #fff;
+  border: 1px solid #d7e6f6;
+  border-radius: 12px;
+  display: grid;
+  gap: 10px;
+  max-height: 86vh;
+  max-width: 820px;
+  padding: 12px;
+  width: min(92vw, 820px);
+}
+
+.chat-modal-header {
+  align-items: center;
+  border-bottom: 1px solid #dbe2ea;
+  display: flex;
+  justify-content: space-between;
+  padding-bottom: 8px;
+}
+
+.chat-modal-header small {
+  color: #687788;
+  display: block;
+  margin-top: 2px;
+}
+
+.chat-modal-messages {
+  background: #f8fbff;
+  border: 1px solid #d7e6f6;
+  border-radius: 10px;
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  max-height: 56vh;
+  overflow-y: auto;
+  padding: 10px;
+}
+
+.chat-modal-message {
+  background: #fff;
+  border: 1px solid #dbe2ea;
+  border-radius: 8px;
+  padding: 8px;
+}
+
+.chat-modal-message.user {
+  border-left: 3px solid #4d7ef3;
+}
+
+.chat-modal-message.assistant {
+  border-left: 3px solid #26a269;
+}
+
+.chat-modal-message.tool {
+  border-left: 3px solid #8e6bd8;
+}
+
+.chat-modal-message p {
+  margin: 6px 0 0 0;
+  white-space: pre-wrap;
+}
+
+.chat-modal-input {
+  display: grid;
+  gap: 8px;
+  grid-template-columns: 1fr auto;
+  align-items: start;
+}
+
+.chat-modal-input textarea {
+  border: 1px solid #c9d4e2;
+  border-radius: 8px;
+  font-size: 14px;
+  padding: 8px;
+  resize: vertical;
+}
+
+.chat-modal-actions {
+  display: flex;
+  gap: 8px;
+  height: 100%;
+}
+
+.chat-modal-actions button {
+  height: auto;
+  white-space: nowrap;
+}
+
+.chat-stop-button {
+  background-color: #dc3545;
+  color: white;
+}
+
+.chat-stop-button:hover {
+  background-color: #bb2d3b;
+}
+

frontend/src/types.ts

@@ -23,6 +23,7 @@ export type GraphConfigReadResponse = {
   tool_keys: string[];
   prompt_dict: Record<string, string>;
   api_key: string;
+  graph_params?: Record<string, unknown>;
 };
 
 export type GraphConfigUpsertRequest = {
@@ -55,6 +56,7 @@ export type PipelineCreateRequest = {
   api_key?: string;
   llm_name: string;
   enabled?: boolean;
+  graph_params?: Record<string, unknown>;
 };
 
 export type PipelineSpec = {
@@ -63,7 +65,6 @@ export type PipelineSpec = {
   enabled: boolean;
   config_file: string;
   llm_name: string;
-  overrides: Record<string, unknown>;
 };
 
 export type PipelineCreateResponse = {
@@ -89,6 +90,38 @@ export type PipelineStopResponse = {
   reload_required: boolean;
 };
 
+export type ConversationListItem = {
+  conversation_id: string;
+  pipeline_id: string;
+  message_count: number;
+  last_updated?: string | null;
+};
+
+export type PipelineConversationListResponse = {
+  pipeline_id: string;
+  items: ConversationListItem[];
+  count: number;
+};
+
+export type ConversationMessageItem = {
+  message_type: string;
+  content: string;
+  sequence_number: number;
+  created_at: string;
+};
+
+export type PipelineConversationMessagesResponse = {
+  pipeline_id: string;
+  conversation_id: string;
+  items: ConversationMessageItem[];
+  count: number;
+};
+
+export type RuntimeAuthInfoResponse = {
+  fast_api_key: string;
+  source: string;
+};
+
 export type McpToolConfigResponse = {
   path: string;
   raw_content: string;
@@ -105,3 +138,15 @@ export type McpToolConfigUpdateResponse = {
   tool_keys: string[];
 };
 
+export type McpAvailableToolsResponse = {
+  available_tools: string[];
+  errors: string[];
+  servers: Record<
+    string,
+    {
+      tools: string[];
+      error?: string | null;
+    }
+  >;
+};
+

frontend/tsconfig.tsbuildinfo

@@ -1 +1 @@
-{"root":["./src/App.tsx","./src/activeConfigSelection.test.ts","./src/activeConfigSelection.ts","./src/main.tsx","./src/types.ts","./src/vite-env.d.ts","./src/api/frontApis.ts"],"version":"5.9.3"}
+{"root":["./src/App.tsx","./src/activeConfigSelection.test.ts","./src/activeConfigSelection.ts","./src/main.tsx","./src/types.ts","./src/vite-env.d.ts","./src/api/frontApis.test.ts","./src/api/frontApis.ts"],"version":"5.9.3"}

frontend/vite.config.js

@@ -4,5 +4,15 @@ export default defineConfig({
     plugins: [react()],
     server: {
         port: 5173,
+        proxy: {
+            "/v1": {
+                target: "http://127.0.0.1:8500",
+                changeOrigin: true,
+            },
+            "/apps": {
+                target: "http://127.0.0.1:8500",
+                changeOrigin: true,
+            },
+        },
     },
 });

frontend/vite.config.ts

@@ -5,6 +5,16 @@ export default defineConfig({
   plugins: [react()],
   server: {
     port: 5173,
+    proxy: {
+      "/v1": {
+        target: "http://127.0.0.1:8500",
+        changeOrigin: true,
+      },
+      "/apps": {
+        target: "http://127.0.0.1:8500",
+        changeOrigin: true,
+      },
+    },
   },
 });
 

lang_agent/components/client_tool_manager.py

@@ -251,32 +251,30 @@ class ClientToolManager:
     def populate_module(self):
         with open(self.config.mcp_config_f, "r") as f:
             self.mcp_configs:dict = commentjson.load(f)
+
+    def _get_to_load_configs(self) -> dict:
+        if self.config.tool_keys is None:
+            return self.mcp_configs
+
+        if len(self.config.tool_keys) == 0:
+            logger.info("no tools will be loaded")
+            return {}
+
+        to_load_config = {}
+        for key in self.config.tool_keys:
+            val = self.mcp_configs.get(key)
+            if val is None:
+                logger.warning(f"{key} is not in mcp tools")
+            else:
+                to_load_config[key] = val
+        return to_load_config
     
     async def aget_tools(self):
         """
         Get tools from all configured MCP servers.
         Handles connection failures gracefully by logging warnings and continuing.
         """
-        
-        def get_to_load_configs() -> dict:
-            if self.config.tool_keys is None:
-                to_load_config = self.mcp_configs
-            else:
-                if len(self.config.tool_keys) == 0:
-                    logger.info("no tools will be loaded")
-                    return {}
-
-                to_load_config = {}
-                for key in self.config.tool_keys:
-                    val = self.mcp_configs.get(key)
-                    if val is None:
-                        logger.warning(f"{key} is not in mcp tools")
-                    else:
-                        to_load_config[key] = val
-            
-            return to_load_config
-        
-        to_load_config = get_to_load_configs()
+        to_load_config = self._get_to_load_configs()
         all_tools = []
         for server_name, server_config in to_load_config.items():
             try:
@@ -298,6 +296,78 @@ class ClientToolManager:
         
         return all_tools
 
+    async def aget_tools_with_errors(self):
+        """
+        Get tools and collect human-readable per-server errors.
+        Returns:
+            (all_tools, errors)
+        """
+        to_load_config = self._get_to_load_configs()
+        all_tools = []
+        errors = []
+        for server_name, server_config in to_load_config.items():
+            try:
+                single_server_config = {server_name: server_config}
+                client = MultiServerMCPClient(single_server_config)
+                tools = await client.get_tools()
+                all_tools.extend(tools)
+                logger.info(
+                    f"Successfully connected to MCP server '{server_name}', retrieved {len(tools)} tools"
+                )
+            except Exception as e:
+                url = (
+                    server_config.get("url", "unknown URL")
+                    if isinstance(server_config, dict)
+                    else "unknown URL"
+                )
+                err_msg = (
+                    f"{server_name} ({url}): {type(e).__name__}: {e}"
+                )
+                errors.append(err_msg)
+                logger.exception(
+                    f"Failed to connect to MCP server '{server_name}' at {url}"
+                )
+                if hasattr(e, "exceptions"):
+                    for nested_exc in e.exceptions:
+                        errors.append(
+                            f"{server_name} nested: {type(nested_exc).__name__}: {nested_exc}"
+                        )
+                continue
+        return all_tools, errors
+
+    async def aget_tools_by_server(self) -> dict:
+        """
+        Get MCP tools grouped by server name, including per-server error (if any).
+        Returns:
+            {
+                "server_name": {
+                    "tools": ["tool_a", "tool_b"],
+                    "error": "ExceptionType: message" | None,
+                },
+                ...
+            }
+        """
+        to_load_config = self._get_to_load_configs()
+        grouped = {}
+        for server_name, server_config in to_load_config.items():
+            grouped[server_name] = {"tools": [], "error": None}
+            try:
+                single_server_config = {server_name: server_config}
+                client = MultiServerMCPClient(single_server_config)
+                tools = await client.get_tools()
+                tool_names = sorted(
+                    {
+                        str(getattr(tool, "name", "")).strip()
+                        for tool in tools
+                        if str(getattr(tool, "name", "")).strip()
+                    }
+                )
+                grouped[server_name] = {"tools": tool_names, "error": None}
+            except Exception as e:
+                grouped[server_name]["error"] = f"{type(e).__name__}: {e}"
+                logger.exception(f"Failed to connect to MCP server '{server_name}'")
+        return grouped
+
     def get_tools(self):
         try:
             loop = asyncio.get_running_loop()

lang_agent/components/server_pipeline_manager.py

@@ -4,6 +4,7 @@ from pathlib import Path as FsPath
 import os.path as osp
 import json
 import copy
+from threading import RLock
 from loguru import logger
 
 from lang_agent.pipeline import Pipeline, PipelineConfig
@@ -20,6 +21,9 @@ class ServerPipelineManager:
         self._api_key_policy: Dict[str, Dict[str, Any]] = {}
         self._pipelines: Dict[str, Pipeline] = {}
         self._pipeline_llm: Dict[str, str] = {}
+        self._registry_path: Optional[str] = None
+        self._registry_mtime_ns: Optional[int] = None
+        self._lock = RLock()
 
     def _resolve_registry_path(self, registry_path: str) -> str:
         path = FsPath(registry_path)
@@ -30,39 +34,102 @@ class ServerPipelineManager:
         root = FsPath(__file__).resolve().parents[2]
         return str((root / path).resolve())
 
-    def load_registry(self, registry_path: str) -> None:
-        abs_path = self._resolve_registry_path(registry_path)
-        if not osp.exists(abs_path):
-            raise ValueError(f"pipeline registry file not found: {abs_path}")
+    def _stat_registry_mtime_ns(self, abs_path: str) -> int:
+        return FsPath(abs_path).stat().st_mtime_ns
 
+    def _read_registry(self, abs_path: str) -> Dict[str, Any]:
         with open(abs_path, "r", encoding="utf-8") as f:
-            registry: dict = json.load(f)
+            return json.load(f)
 
+    def _apply_registry(self, abs_path: str, registry: Dict[str, Any], mtime_ns: int) -> bool:
         pipelines = registry.get("pipelines")
-        if pipelines is None:
+        if pipelines is None or not isinstance(pipelines, dict):
             raise ValueError("`pipelines` in pipeline registry must be an object.")
 
-        self._pipeline_specs = {}
+        parsed_specs: Dict[str, Dict[str, Any]] = {}
         for pipeline_id, spec in pipelines.items():
             if not isinstance(spec, dict):
                 raise ValueError(
                     f"pipeline spec for `{pipeline_id}` must be an object."
                 )
-            self._pipeline_specs[pipeline_id] = {
+            parsed_specs[pipeline_id] = {
                 "enabled": bool(spec.get("enabled", True)),
                 "config_file": spec.get("config_file"),
-                "overrides": spec.get("overrides", {}),
+                "llm_name": spec.get("llm_name"),
             }
-        if not self._pipeline_specs:
+        if not parsed_specs:
             raise ValueError("pipeline registry must define at least one pipeline.")
 
         api_key_policy = registry.get("api_keys", {})
         if api_key_policy and not isinstance(api_key_policy, dict):
             raise ValueError("`api_keys` in pipeline registry must be an object.")
-        self._api_key_policy = api_key_policy
-        logger.info(
-            f"loaded pipeline registry: {abs_path}, pipelines={list(self._pipeline_specs.keys())}"
+
+        with self._lock:
+            old_specs = self._pipeline_specs
+            old_policy = self._api_key_policy
+            old_mtime = self._registry_mtime_ns
+
+            removed = set(old_specs.keys()) - set(parsed_specs.keys())
+            added = set(parsed_specs.keys()) - set(old_specs.keys())
+            modified = {
+                pipeline_id
+                for pipeline_id in (set(old_specs.keys()) & set(parsed_specs.keys()))
+                if old_specs[pipeline_id] != parsed_specs[pipeline_id]
+            }
+            changed = bool(added or removed or modified or old_policy != api_key_policy)
+
+            # Drop stale cache entries for deleted/changed pipelines so future requests
+            # lazily rebuild from the refreshed registry spec.
+            for pipeline_id in (removed | modified):
+                self._pipelines.pop(pipeline_id, None)
+                self._pipeline_llm.pop(pipeline_id, None)
+
+            self._pipeline_specs = parsed_specs
+            self._api_key_policy = api_key_policy
+            self._registry_path = abs_path
+            self._registry_mtime_ns = mtime_ns
+
+        if changed:
+            logger.info(
+                "refreshed pipeline registry: {} | added={} modified={} removed={} mtime={}",
+                abs_path,
+                sorted(added),
+                sorted(modified),
+                sorted(removed),
+                mtime_ns,
+            )
+        elif old_mtime != mtime_ns:
+            logger.debug("pipeline registry mtime changed but specs were unchanged: {}", abs_path)
+        return changed
+
+    def load_registry(self, registry_path: str) -> None:
+        abs_path = self._resolve_registry_path(registry_path)
+        if not osp.exists(abs_path):
+            raise ValueError(f"pipeline registry file not found: {abs_path}")
+        registry = self._read_registry(abs_path)
+        mtime_ns = self._stat_registry_mtime_ns(abs_path)
+        self._apply_registry(abs_path=abs_path, registry=registry, mtime_ns=mtime_ns)
+
+    def refresh_registry_if_needed(
+        self, registry_path: Optional[str] = None, force: bool = False
+    ) -> bool:
+        abs_path = (
+            self._resolve_registry_path(registry_path)
+            if registry_path
+            else self._registry_path
         )
+        if not abs_path:
+            raise ValueError("registry path is not initialized")
+        if not osp.exists(abs_path):
+            raise ValueError(f"pipeline registry file not found: {abs_path}")
+
+        mtime_ns = self._stat_registry_mtime_ns(abs_path)
+        with self._lock:
+            if not force and self._registry_mtime_ns == mtime_ns:
+                return False
+
+        registry = self._read_registry(abs_path)
+        return self._apply_registry(abs_path=abs_path, registry=registry, mtime_ns=mtime_ns)
 
     def _resolve_config_path(self, config_file: str) -> str:
         path = FsPath(config_file)
@@ -85,32 +152,25 @@ class ServerPipelineManager:
             )
 
         config_file = spec.get("config_file")
-        overrides = spec.get("overrides", {})
+        registry_llm_name = spec.get("llm_name")
         if config_file:
             loaded_cfg = load_tyro_conf(self._resolve_config_path(config_file))
             if hasattr(loaded_cfg, "setup"):
                 cfg = loaded_cfg
             else:
-                logger.warning(
-                    f"config_file for pipeline `{pipeline_id}` did not deserialize to config object; "
-                    "falling back to default config and applying pipeline-level overrides."
+                raise ValueError(
+                    "config_file for pipeline "
+                    f"`{pipeline_id}` did not deserialize to a config object. "
+                    "Rebuild the pipeline via /v1/pipelines to regenerate a "
+                    "valid serialized PipelineConfig file."
                 )
-                cfg = copy.deepcopy(self.default_config)
         else:
             cfg = copy.deepcopy(self.default_config)
-        if not isinstance(overrides, dict):
-            raise ValueError(
-                f"pipeline `overrides` for `{pipeline_id}` must be an object."
-            )
-        for key, value in overrides.items():
-            if not hasattr(cfg, key):
-                raise ValueError(
-                    f"unknown override field `{key}` for pipeline `{pipeline_id}`"
-                )
-            setattr(cfg, key, value)
+            if registry_llm_name is not None and hasattr(cfg, "llm_name"):
+                setattr(cfg, "llm_name", registry_llm_name)
 
         p = cfg.setup()
-        llm_name = getattr(cfg, "llm_name", "unknown-model")
+        llm_name = str(getattr(cfg, "llm_name", registry_llm_name or "unknown-model"))
         return p, llm_name
 
     def _authorize(self, api_key: str, pipeline_id: str) -> None:
@@ -138,29 +198,33 @@ class ServerPipelineManager:
             or app_id
         )
 
-        if not pipeline_id:
-            key_policy = (
-                self._api_key_policy.get(api_key, {}) if self._api_key_policy else {}
-            )
-            pipeline_id = key_policy.get(
-                "default_pipeline_id", self.default_pipeline_id
-            )
+        with self._lock:
+            if not pipeline_id:
+                key_policy = (
+                    self._api_key_policy.get(api_key, {}) if self._api_key_policy else {}
+                )
+                pipeline_id = key_policy.get(
+                    "default_pipeline_id", self.default_pipeline_id
+                )
 
-        if pipeline_id not in self._pipeline_specs:
-            raise HTTPException(
-                status_code=404, detail=f"Unknown pipeline_id: {pipeline_id}"
-            )
+            if pipeline_id not in self._pipeline_specs:
+                raise HTTPException(
+                    status_code=404, detail=f"Unknown pipeline_id: {pipeline_id}"
+                )
 
-        self._authorize(api_key, pipeline_id)
+            self._authorize(api_key, pipeline_id)
         return pipeline_id
 
     def get_pipeline(self, pipeline_id: str) -> Tuple[Pipeline, str]:
-        cached = self._pipelines.get(pipeline_id)
-        if cached is not None:
-            return cached, self._pipeline_llm[pipeline_id]
+        with self._lock:
+            cached = self._pipelines.get(pipeline_id)
+            if cached is not None:
+                return cached, self._pipeline_llm[pipeline_id]
 
-        pipeline_obj, llm_name = self._build_pipeline(pipeline_id)
-        self._pipelines[pipeline_id] = pipeline_obj
-        self._pipeline_llm[pipeline_id] = llm_name
+            # Build while holding the lock to avoid duplicate construction for
+            # the same pipeline on concurrent first requests.
+            pipeline_obj, llm_name = self._build_pipeline(pipeline_id)
+            self._pipelines[pipeline_id] = pipeline_obj
+            self._pipeline_llm[pipeline_id] = llm_name
         logger.info(f"lazy-loaded pipeline_id={pipeline_id} model={llm_name}")
         return pipeline_obj, llm_name

lang_agent/config/__init__.py

@@ -4,6 +4,7 @@ from lang_agent.config.core_config import (
     LLMKeyConfig,
     LLMNodeConfig,
     load_tyro_conf,
+    resolve_llm_api_key,
 )
 
 from lang_agent.config.constants import (
@@ -12,5 +13,7 @@ from lang_agent.config.constants import (
     PIPELINE_REGISTRY_PATH,
     VALID_API_KEYS,
     API_KEY_HEADER,
-    API_KEY_HEADER_NO_ERROR
+    API_KEY_HEADER_NO_ERROR,
+    _PROJECT_ROOT,
+    TY_BUILD_SCRIPT,
 )

lang_agent/config/constants.py

@@ -15,3 +15,5 @@ API_KEY_HEADER = APIKeyHeader(name="Authorization", auto_error=True)
 API_KEY_HEADER_NO_ERROR = APIKeyHeader(name="Authorization", auto_error=False)
 
 VALID_API_KEYS = set(filter(None, os.environ.get("FAST_AUTH_KEYS", "").split(",")))
+
+TY_BUILD_SCRIPT = osp.join(_PROJECT_ROOT, "lang_agent", "config", "ty_build_config.py")

lang_agent/config/core_config.py

@@ -10,6 +10,20 @@ from dotenv import load_dotenv
 
 load_dotenv()
 
+
+def resolve_llm_api_key(api_key: Optional[str]) -> Optional[str]:
+    """Resolve the API key for OpenAI-compatible providers."""
+    if api_key not in (None, "", "wrong-key"):
+        resolved_key = api_key
+    else:
+        resolved_key = os.environ.get("ALI_API_KEY") or os.environ.get("OPENAI_API_KEY")
+
+    # Some OpenAI-compatible integrations still read OPENAI_API_KEY from env.
+    if resolved_key and not os.environ.get("OPENAI_API_KEY"):
+        os.environ["OPENAI_API_KEY"] = resolved_key
+
+    return resolved_key
+
 ## NOTE: base classes taken from nerfstudio
 class PrintableConfig:
     """
@@ -99,12 +113,12 @@ class LLMKeyConfig(InstantiateConfig):
     """api key for llm"""
 
     def __post_init__(self):
-        if self.api_key == "wrong-key" or self.api_key is None:
-            self.api_key = os.environ.get("ALI_API_KEY")
-            if self.api_key is None:
-                logger.error(f"no ALI_API_KEY provided for embedding")
-            else:
-                logger.info("ALI_API_KEY loaded from environ")
+        original_api_key = self.api_key
+        self.api_key = resolve_llm_api_key(self.api_key)
+        if self.api_key is None:
+            logger.error("no ALI_API_KEY or OPENAI_API_KEY provided for embedding")
+        elif original_api_key in (None, "", "wrong-key"):
+            logger.info("LLM API key loaded from environment")
 
 
 @dataclass

lang_agent/eval/evaluator.py

@@ -96,7 +96,7 @@ class Evaluator:
         
         df_m.to_csv(metric_f)
 
-        self.config.save_config(f"{head_path}-{n_exp}.yml")
+        self.config.save_config(f"{head_path}-{n_exp}.yaml")
     
     def format_result_df(self, df:pd.DataFrame):
 

lang_agent/fastapi_server/combined.py

@@ -0,0 +1,33 @@
+from fastapi import FastAPI
+from fastapi.middleware.cors import CORSMiddleware
+import uvicorn
+
+from lang_agent.fastapi_server.front_apis import app as front_app
+from lang_agent.fastapi_server.server_dashscope import create_dashscope_router
+
+
+app = FastAPI(
+    title="Combined Front + DashScope APIs",
+    description=(
+        "Single-process app exposing front_apis control endpoints and "
+        "DashScope-compatible chat endpoints."
+    ),
+)
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# Keep existing /v1/... admin APIs unchanged.
+app.include_router(front_app.router)
+
+# Add DashScope endpoints at their existing URLs. We intentionally skip
+# DashScope's root/health routes to avoid clashing with front_apis.
+app.include_router(create_dashscope_router(include_meta_routes=False))
+
+if __name__ == "__main__":
+    uvicorn.run(app, host="0.0.0.0", port=8500)

lang_agent/fastapi_server/front_apis.py

@@ -4,6 +4,7 @@ import os
 import os.path as osp
 import sys
 import json
+import psycopg
 
 from fastapi import FastAPI, HTTPException
 from fastapi.middleware.cors import CORSMiddleware
@@ -19,10 +20,15 @@ from lang_agent.config.constants import (
     MCP_CONFIG_DEFAULT_CONTENT,
     PIPELINE_REGISTRY_PATH,
 )
+from lang_agent.config.core_config import load_tyro_conf
 from lang_agent.front_api.build_server_utils import (
     GRAPH_BUILD_FNCS,
     update_pipeline_registry,
 )
+from lang_agent.components.client_tool_manager import (
+    ClientToolManager,
+    ClientToolManagerConfig,
+)
 
 
 class GraphConfigUpsertRequest(BaseModel):
@@ -50,6 +56,7 @@ class GraphConfigReadResponse(BaseModel):
     tool_keys: List[str]
     prompt_dict: Dict[str, str]
     api_key: str = Field(default="")
+    graph_params: Dict[str, Any] = Field(default_factory=dict)
 
 
 class GraphConfigListItem(BaseModel):
@@ -80,6 +87,8 @@ class PipelineCreateRequest(BaseModel):
     api_key: Optional[str] = Field(default=None)
     llm_name: str = Field(default="qwen-plus")
     enabled: bool = Field(default=True)
+    # Arbitrary per-graph options forwarded to the graph build function.
+    graph_params: Dict[str, Any] = Field(default_factory=dict)
 
 
 class PipelineSpec(BaseModel):
@@ -88,7 +97,6 @@ class PipelineSpec(BaseModel):
     enabled: bool
     config_file: str
     llm_name: str
-    overrides: Dict[str, Any] = Field(default_factory=dict)
 
 
 class PipelineCreateResponse(BaseModel):
@@ -114,6 +122,38 @@ class PipelineStopResponse(BaseModel):
     reload_required: bool
 
 
+class ConversationListItem(BaseModel):
+    conversation_id: str
+    pipeline_id: str
+    message_count: int
+    last_updated: Optional[str] = Field(default=None)
+
+
+class PipelineConversationListResponse(BaseModel):
+    pipeline_id: str
+    items: List[ConversationListItem]
+    count: int
+
+
+class ConversationMessageItem(BaseModel):
+    message_type: str
+    content: str
+    sequence_number: int
+    created_at: str
+
+
+class PipelineConversationMessagesResponse(BaseModel):
+    pipeline_id: str
+    conversation_id: str
+    items: List[ConversationMessageItem]
+    count: int
+
+
+class RuntimeAuthInfoResponse(BaseModel):
+    fast_api_key: str
+    source: str
+
+
 class ApiKeyPolicyItem(BaseModel):
     api_key: str
     default_pipeline_id: Optional[str] = Field(default=None)
@@ -154,6 +194,12 @@ class McpConfigUpdateResponse(BaseModel):
     tool_keys: List[str]
 
 
+class McpAvailableToolsResponse(BaseModel):
+    available_tools: List[str] = Field(default_factory=list)
+    errors: List[str] = Field(default_factory=list)
+    servers: Dict[str, Dict[str, Any]] = Field(default_factory=dict)
+
+
 app = FastAPI(
     title="Front APIs",
     description="Manage graph configs and launch graph pipelines.",
@@ -190,11 +236,15 @@ async def root():
             "/v1/pipelines (POST) - build config + upsert pipeline registry entry",
             "/v1/pipelines (GET) - list registry pipeline specs",
             "/v1/pipelines/{pipeline_id} (DELETE) - disable pipeline in registry",
+            "/v1/runtime-auth (GET) - show runtime FAST API key info",
+            "/v1/pipelines/{pipeline_id}/conversations (GET) - list pipeline conversations",
+            "/v1/pipelines/{pipeline_id}/conversations/{conversation_id}/messages (GET) - list messages in a conversation",
             "/v1/pipelines/api-keys (GET) - list API key routing policies",
             "/v1/pipelines/api-keys/{api_key} (PUT) - upsert API key routing policy",
             "/v1/pipelines/api-keys/{api_key} (DELETE) - delete API key routing policy",
             "/v1/tool-configs/mcp (GET)",
             "/v1/tool-configs/mcp (PUT)",
+            "/v1/tool-configs/mcp/tools (GET)",
         ],
     }
 
@@ -240,25 +290,118 @@ def _write_pipeline_registry(registry: Dict[str, Any]) -> None:
         f.write("\n")
 
 
+def _resolve_runtime_fast_api_key() -> RuntimeAuthInfoResponse:
+    """Pick a runtime auth key from pipeline registry first, then FAST_AUTH_KEYS env."""
+    try:
+        registry = _read_pipeline_registry()
+        api_keys = registry.get("api_keys", {})
+        if isinstance(api_keys, dict):
+            for key in api_keys.keys():
+                candidate = str(key).strip()
+                if candidate:
+                    return RuntimeAuthInfoResponse(
+                        fast_api_key=candidate, source="pipeline_registry"
+                    )
+    except Exception:
+        # fall back to env parsing below
+        pass
+
+    raw_env = os.environ.get("FAST_AUTH_KEYS", "")
+    for token in raw_env.split(","):
+        candidate = token.strip()
+        if candidate:
+            return RuntimeAuthInfoResponse(fast_api_key=candidate, source="env")
+    return RuntimeAuthInfoResponse(fast_api_key="", source="none")
+
+
 def _normalize_pipeline_spec(pipeline_id: str, spec: Dict[str, Any]) -> PipelineSpec:
     if not isinstance(spec, dict):
         raise ValueError(f"pipeline spec for '{pipeline_id}' must be an object")
-    overrides = spec.get("overrides", {})
-    if overrides is None:
-        overrides = {}
-    if not isinstance(overrides, dict):
-        raise ValueError(f"`overrides` for pipeline '{pipeline_id}' must be an object")
-    llm_name = str(overrides.get("llm_name") or "unknown")
+    llm_name = str(spec.get("llm_name") or "unknown")
     return PipelineSpec(
         pipeline_id=pipeline_id,
         graph_id=str(spec.get("graph_id") or pipeline_id),
         enabled=bool(spec.get("enabled", True)),
         config_file=str(spec.get("config_file") or ""),
         llm_name=llm_name,
-        overrides=overrides,
     )
 
 
+def _resolve_config_path(config_file: str) -> str:
+    if osp.isabs(config_file):
+        return config_file
+    return osp.join(_PROJECT_ROOT, config_file)
+
+
+def _normalize_deepagent_backend_name(file_backend_config: Any) -> Optional[str]:
+    if file_backend_config is None:
+        return None
+
+    type_names = {
+        type(file_backend_config).__name__.lower(),
+        getattr(getattr(file_backend_config, "_target", None), "__name__", "").lower(),
+    }
+    if any("statebk" in name for name in type_names):
+        return "state_bk"
+    if any("localshell" in name for name in type_names):
+        return "local_shell"
+    if any("daytona" in name for name in type_names):
+        return "daytona_sandbox"
+    return None
+
+
+def _extract_graph_params_from_config(graph_id: Optional[str], loaded_cfg: Any) -> Dict[str, Any]:
+    if graph_id != "deepagent":
+        return {}
+
+    graph_config = getattr(loaded_cfg, "graph_config", None)
+    file_backend_config = getattr(graph_config, "file_backend_config", None)
+    if file_backend_config is None:
+        return {}
+
+    graph_params: Dict[str, Any] = {}
+    act_bkend = _normalize_deepagent_backend_name(file_backend_config)
+    if act_bkend:
+        graph_params["act_bkend"] = act_bkend
+
+    serialized_backend_config: Dict[str, Any] = {}
+    for key in ("skills_dir", "rt_skills_dir", "workspace_dir", "api_key"):
+        value = getattr(file_backend_config, key, None)
+        if value is not None:
+            serialized_backend_config[key] = value
+
+    if serialized_backend_config:
+        graph_params["file_backend_config"] = serialized_backend_config
+
+    return graph_params
+
+
+def _load_graph_params_for_pipeline(
+    pipeline_id: str, graph_id: Optional[str]
+) -> Dict[str, Any]:
+    try:
+        registry = _read_pipeline_registry()
+        pipeline_spec = registry.get("pipelines", {}).get(pipeline_id, {})
+        config_file = ""
+        if isinstance(pipeline_spec, dict):
+            config_file = str(pipeline_spec.get("config_file") or "").strip()
+        if not config_file:
+            fallback = osp.join(_PROJECT_ROOT, "configs", "pipelines", f"{pipeline_id}.yaml")
+            if osp.exists(fallback):
+                config_file = fallback
+        if not config_file:
+            return {}
+
+        config_path = _resolve_config_path(config_file)
+        if not osp.exists(config_path):
+            return {}
+
+        loaded_cfg = load_tyro_conf(config_path)
+        return _extract_graph_params_from_config(graph_id, loaded_cfg)
+    except Exception:
+        return {}
+
+
 def _normalize_api_key_policy(api_key: str, policy: Dict[str, Any]) -> ApiKeyPolicyItem:
     if not isinstance(policy, dict):
         raise ValueError(f"api key policy for '{api_key}' must be an object")
@@ -362,6 +505,9 @@ async def get_default_graph_config(pipeline_id: str):
         tool_keys=tool_keys,
         prompt_dict=prompt_dict,
         api_key=(active.get("api_key") or ""),
+        graph_params=_load_graph_params_for_pipeline(
+            pipeline_id, active.get("graph_id")
+        ),
     )
 
 
@@ -400,6 +546,9 @@ async def get_graph_config(pipeline_id: str, prompt_set_id: str):
         tool_keys=tool_keys,
         prompt_dict=prompt_dict,
         api_key=(meta.get("api_key") or ""),
+        graph_params=_load_graph_params_for_pipeline(
+            pipeline_id, meta.get("graph_id")
+        ),
     )
 
 
@@ -459,6 +608,38 @@ async def update_mcp_tool_config(body: McpConfigUpdateRequest):
     )
 
 
+@app.get("/v1/tool-configs/mcp/tools", response_model=McpAvailableToolsResponse)
+async def list_mcp_available_tools():
+    try:
+        _read_mcp_config_raw()
+        manager = ClientToolManager(
+            ClientToolManagerConfig(mcp_config_f=MCP_CONFIG_PATH)
+        )
+        servers = await manager.aget_tools_by_server()
+        available_tools = sorted(
+            {
+                tool_name
+                for server_info in servers.values()
+                for tool_name in server_info.get("tools", [])
+            }
+        )
+        errors = [
+            f"{server_name}: {server_info.get('error')}"
+            for server_name, server_info in servers.items()
+            if server_info.get("error")
+        ]
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+    return McpAvailableToolsResponse(
+        available_tools=available_tools,
+        errors=errors,
+        servers=servers,
+    )
+
+
 @app.get("/v1/pipelines", response_model=PipelineListResponse)
 async def list_running_pipelines():
     try:
@@ -511,8 +692,9 @@ async def create_pipeline(body: PipelineCreateRequest):
             ),
         )
 
-    config_file = f"configs/pipelines/{pipeline_id}.yml"
+    config_file = f"configs/pipelines/{pipeline_id}.yaml"
     config_abs_dir = osp.join(_PROJECT_ROOT, "configs", "pipelines")
+    extra_params = dict(body.graph_params or {})
     try:
         build_fn(
             pipeline_id=pipeline_id,
@@ -521,6 +703,7 @@ async def create_pipeline(body: PipelineCreateRequest):
             api_key=resolved_api_key,
             llm_name=body.llm_name,
             pipeline_config_dir=config_abs_dir,
+            **extra_params,
         )
 
         update_pipeline_registry(
@@ -555,7 +738,7 @@ async def create_pipeline(body: PipelineCreateRequest):
         config_file=normalized.config_file,
         llm_name=normalized.llm_name,
         enabled=normalized.enabled,
-        reload_required=True,
+        reload_required=False,
         registry_path=PIPELINE_REGISTRY_PATH,
     )
 
@@ -583,7 +766,129 @@ async def stop_pipeline(pipeline_id: str):
         pipeline_id=pipeline_id,
         status="disabled",
         enabled=False,
-        reload_required=True,
+        reload_required=False,
+    )
+
+
+@app.get("/v1/runtime-auth", response_model=RuntimeAuthInfoResponse)
+async def get_runtime_auth_info():
+    return _resolve_runtime_fast_api_key()
+
+
+@app.get(
+    "/v1/pipelines/{pipeline_id}/conversations",
+    response_model=PipelineConversationListResponse,
+)
+async def list_pipeline_conversations(pipeline_id: str, limit: int = 100):
+    if limit < 1 or limit > 500:
+        raise HTTPException(status_code=400, detail="limit must be between 1 and 500")
+
+    conn_str = os.environ.get("CONN_STR")
+    if not conn_str:
+        raise HTTPException(status_code=500, detail="CONN_STR not set")
+
+    try:
+        with psycopg.connect(conn_str) as conn:
+            with conn.cursor(row_factory=psycopg.rows.dict_row) as cur:
+                cur.execute(
+                    """
+                    SELECT
+                        conversation_id,
+                        pipeline_id,
+                        COUNT(*) AS message_count,
+                        MAX(created_at) AS last_updated
+                    FROM messages
+                    WHERE pipeline_id = %s
+                    GROUP BY conversation_id, pipeline_id
+                    ORDER BY last_updated DESC
+                    LIMIT %s
+                    """,
+                    (pipeline_id, limit),
+                )
+                rows = cur.fetchall()
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+    items = [
+        ConversationListItem(
+            conversation_id=str(row["conversation_id"]),
+            pipeline_id=str(row["pipeline_id"]),
+            message_count=int(row["message_count"]),
+            last_updated=(
+                row["last_updated"].isoformat() if row.get("last_updated") else None
+            ),
+        )
+        for row in rows
+    ]
+    return PipelineConversationListResponse(
+        pipeline_id=pipeline_id, items=items, count=len(items)
+    )
+
+
+@app.get(
+    "/v1/pipelines/{pipeline_id}/conversations/{conversation_id}/messages",
+    response_model=PipelineConversationMessagesResponse,
+)
+async def get_pipeline_conversation_messages(pipeline_id: str, conversation_id: str):
+    conn_str = os.environ.get("CONN_STR")
+    if not conn_str:
+        raise HTTPException(status_code=500, detail="CONN_STR not set")
+
+    try:
+        with psycopg.connect(conn_str) as conn:
+            with conn.cursor(row_factory=psycopg.rows.dict_row) as cur:
+                cur.execute(
+                    """
+                    SELECT 1
+                    FROM messages
+                    WHERE pipeline_id = %s AND conversation_id = %s
+                    LIMIT 1
+                    """,
+                    (pipeline_id, conversation_id),
+                )
+                exists = cur.fetchone()
+                if exists is None:
+                    raise HTTPException(
+                        status_code=404,
+                        detail=(
+                            f"conversation_id '{conversation_id}' not found for "
+                            f"pipeline '{pipeline_id}'"
+                        ),
+                    )
+
+                cur.execute(
+                    """
+                    SELECT
+                        message_type,
+                        content,
+                        sequence_number,
+                        created_at
+                    FROM messages
+                    WHERE pipeline_id = %s AND conversation_id = %s
+                    ORDER BY sequence_number ASC
+                    """,
+                    (pipeline_id, conversation_id),
+                )
+                rows = cur.fetchall()
+    except HTTPException:
+        raise
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+    items = [
+        ConversationMessageItem(
+            message_type=str(row["message_type"]),
+            content=str(row["content"]),
+            sequence_number=int(row["sequence_number"]),
+            created_at=row["created_at"].isoformat() if row.get("created_at") else "",
+        )
+        for row in rows
+    ]
+    return PipelineConversationMessagesResponse(
+        pipeline_id=pipeline_id,
+        conversation_id=conversation_id,
+        items=items,
+        count=len(items),
     )
 
 
@@ -688,5 +993,15 @@ async def delete_pipeline_api_key_policy(api_key: str):
     return ApiKeyPolicyDeleteResponse(
         api_key=normalized_key,
         status="deleted",
-        reload_required=True,
+        reload_required=False,
+    )
+
+
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run(
+        "front_apis:app",
+        host="0.0.0.0",
+        port=8500,
+        reload=True,
     )

lang_agent/fastapi_server/server_dashscope.py

@@ -1,9 +1,8 @@
-from fastapi import FastAPI, HTTPException, Path, Request, Depends, Security
+from fastapi import APIRouter, Depends, FastAPI, HTTPException, Path, Request, Security
 from fastapi.middleware.cors import CORSMiddleware
-from fastapi.responses import StreamingResponse, JSONResponse
-from fastapi.security import APIKeyHeader
+from fastapi.responses import JSONResponse, StreamingResponse
 from pydantic import BaseModel, Field
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any, Dict, List, Optional
 import os
 import os.path as osp
 import sys
@@ -20,16 +19,28 @@ from lang_agent.pipeline import PipelineConfig
 from lang_agent.components.server_pipeline_manager import ServerPipelineManager
 from lang_agent.config.constants import PIPELINE_REGISTRY_PATH, API_KEY_HEADER, VALID_API_KEYS
 
-# Load base config for route-level overrides (pipelines are lazy-loaded from registry)
-pipeline_config = tyro.cli(PipelineConfig)
-logger.info(f"starting agent with base pipeline config: \n{pipeline_config}")
+def _build_default_pipeline_config() -> PipelineConfig:
+    """
+    Build import-time defaults without parsing CLI args.
+
+    This keeps module import safe for reuse by combined apps and tests.
+    """
+    pipeline_config = PipelineConfig()
+    logger.info(f"starting agent with base pipeline config: \n{pipeline_config}")
+    return pipeline_config
 
 
-PIPELINE_MANAGER = ServerPipelineManager(
-    default_pipeline_id=os.environ.get("FAST_DEFAULT_PIPELINE_ID", "default"),
-    default_config=pipeline_config,
-)
-PIPELINE_MANAGER.load_registry(PIPELINE_REGISTRY_PATH)
+def _build_pipeline_manager(base_config: PipelineConfig) -> ServerPipelineManager:
+    pipeline_manager = ServerPipelineManager(
+        default_pipeline_id=os.environ.get("FAST_DEFAULT_PIPELINE_ID", "default"),
+        default_config=base_config,
+    )
+    pipeline_manager.load_registry(PIPELINE_REGISTRY_PATH)
+    return pipeline_manager
+
+
+pipeline_config = _build_default_pipeline_config()
+PIPELINE_MANAGER = _build_pipeline_manager(pipeline_config)
 
 
 async def verify_api_key(api_key: str = Security(API_KEY_HEADER)):
@@ -55,20 +66,6 @@ class DSApplicationCallRequest(BaseModel):
     thread_id: Optional[str] = Field(default="3")
 
 
-app = FastAPI(
-    title="DashScope-Compatible Application API",
-    description="DashScope Application.call compatible endpoint backed by pipeline.chat",
-)
-
-app.add_middleware(
-    CORSMiddleware,
-    allow_origins=["*"],
-    allow_credentials=True,
-    allow_methods=["*"],
-    allow_headers=["*"],
-)
-
-
 def sse_chunks_from_stream(
     chunk_generator, response_id: str, model: str = "qwen-flash"
 ):
@@ -188,7 +185,14 @@ async def _process_dashscope_request(
     app_id: Optional[str],
     session_id: Optional[str],
     api_key: str,
+    pipeline_manager: ServerPipelineManager,
 ):
+    try:
+        pipeline_manager.refresh_registry_if_needed()
+    except Exception as e:
+        logger.error(f"failed to refresh pipeline registry: {e}")
+        raise HTTPException(status_code=500, detail=f"Failed to refresh pipeline registry: {e}")
+
     req_app_id = app_id or body.get("app_id")
     body_input = body.get("input", {}) if isinstance(body.get("input"), dict) else {}
     req_session_id = session_id or body_input.get("session_id")
@@ -201,10 +205,10 @@ async def _process_dashscope_request(
     thread_id = body_input.get("session_id") or req_session_id or "3"
     user_msg = _extract_user_message(messages)
 
-    pipeline_id = PIPELINE_MANAGER.resolve_pipeline_id(
+    pipeline_id = pipeline_manager.resolve_pipeline_id(
         body=body, app_id=req_app_id, api_key=api_key
     )
-    selected_pipeline, selected_model = PIPELINE_MANAGER.get_pipeline(pipeline_id)
+    selected_pipeline, selected_model = pipeline_manager.get_pipeline(pipeline_id)
 
     # Namespace thread ids to prevent memory collisions across pipelines.
     thread_id = f"{pipeline_id}:{thread_id}"
@@ -245,76 +249,117 @@ async def _process_dashscope_request(
     return JSONResponse(content=data)
 
 
-@app.post("/v1/apps/{app_id}/sessions/{session_id}/responses")
-@app.post("/api/v1/apps/{app_id}/sessions/{session_id}/responses")
-async def application_responses(
-    request: Request,
-    app_id: str = Path(...),
-    session_id: str = Path(...),
-    api_key: str = Depends(verify_api_key),
-):
-    try:
-        body = await request.json()
-        return await _process_dashscope_request(
-            body=body,
-            app_id=app_id,
-            session_id=session_id,
-            api_key=api_key,
+def create_dashscope_router(
+    pipeline_manager: Optional[ServerPipelineManager] = None,
+    include_meta_routes: bool = True,
+) -> APIRouter:
+    manager = pipeline_manager or PIPELINE_MANAGER
+    router = APIRouter()
+
+    @router.post("/v1/apps/{app_id}/sessions/{session_id}/responses")
+    @router.post("/api/v1/apps/{app_id}/sessions/{session_id}/responses")
+    async def application_responses(
+        request: Request,
+        app_id: str = Path(...),
+        session_id: str = Path(...),
+        api_key: str = Depends(verify_api_key),
+    ):
+        try:
+            body = await request.json()
+            return await _process_dashscope_request(
+                body=body,
+                app_id=app_id,
+                session_id=session_id,
+                api_key=api_key,
+                pipeline_manager=manager,
+            )
+
+        except HTTPException:
+            raise
+        except Exception as e:
+            logger.error(f"DashScope-compatible endpoint error: {e}")
+            raise HTTPException(status_code=500, detail=str(e))
+
+    # Compatibility: some SDKs call /apps/{app_id}/completion without /v1 and
+    # without session in path.
+    @router.post("/apps/{app_id}/completion")
+    @router.post("/v1/apps/{app_id}/completion")
+    @router.post("/api/apps/{app_id}/completion")
+    @router.post("/api/v1/apps/{app_id}/completion")
+    async def application_completion(
+        request: Request,
+        app_id: str = Path(...),
+        api_key: str = Depends(verify_api_key),
+    ):
+        try:
+            body = await request.json()
+            return await _process_dashscope_request(
+                body=body,
+                app_id=app_id,
+                session_id=None,
+                api_key=api_key,
+                pipeline_manager=manager,
+            )
+
+        except HTTPException:
+            raise
+        except Exception as e:
+            logger.error(f"DashScope-compatible completion error: {e}")
+            raise HTTPException(status_code=500, detail=str(e))
+
+    if include_meta_routes:
+        @router.get("/")
+        async def root():
+            return {
+                "message": "DashScope Application-compatible API",
+                "endpoints": [
+                    "/v1/apps/{app_id}/sessions/{session_id}/responses",
+                    "/health",
+                ],
+            }
+
+        @router.get("/health")
+        async def health():
+            return {"status": "healthy"}
+
+    return router
+
+
+def create_dashscope_app(
+    pipeline_manager: Optional[ServerPipelineManager] = None,
+) -> FastAPI:
+    dashscope_app = FastAPI(
+        title="DashScope-Compatible Application API",
+        description="DashScope Application.call compatible endpoint backed by pipeline.chat",
+    )
+    dashscope_app.add_middleware(
+        CORSMiddleware,
+        allow_origins=["*"],
+        allow_credentials=True,
+        allow_methods=["*"],
+        allow_headers=["*"],
+    )
+    dashscope_app.include_router(
+        create_dashscope_router(
+            pipeline_manager=pipeline_manager,
+            include_meta_routes=True,
         )
-
-    except HTTPException:
-        raise
-    except Exception as e:
-        logger.error(f"DashScope-compatible endpoint error: {e}")
-        raise HTTPException(status_code=500, detail=str(e))
+    )
+    return dashscope_app
 
 
-# Compatibility: some SDKs call /apps/{app_id}/completion without /v1 and without session in path
-@app.post("/apps/{app_id}/completion")
-@app.post("/v1/apps/{app_id}/completion")
-@app.post("/api/apps/{app_id}/completion")
-@app.post("/api/v1/apps/{app_id}/completion")
-async def application_completion(
-    request: Request,
-    app_id: str = Path(...),
-    api_key: str = Depends(verify_api_key),
-):
-    try:
-        body = await request.json()
-        return await _process_dashscope_request(
-            body=body,
-            app_id=app_id,
-            session_id=None,
-            api_key=api_key,
-        )
-
-    except HTTPException:
-        raise
-    except Exception as e:
-        logger.error(f"DashScope-compatible completion error: {e}")
-        raise HTTPException(status_code=500, detail=str(e))
-
-
-@app.get("/")
-async def root():
-    return {
-        "message": "DashScope Application-compatible API",
-        "endpoints": [
-            "/v1/apps/{app_id}/sessions/{session_id}/responses",
-            "/health",
-        ],
-    }
-
-
-@app.get("/health")
-async def health():
-    return {"status": "healthy"}
+dashscope_router = create_dashscope_router(include_meta_routes=False)
+app = create_dashscope_app()
 
 
 if __name__ == "__main__":
+    # CLI parsing is intentionally only in script mode to keep module import safe.
+    cli_pipeline_config = tyro.cli(PipelineConfig)
+    logger.info(f"starting agent with CLI pipeline config: \n{cli_pipeline_config}")
+    cli_pipeline_manager = _build_pipeline_manager(cli_pipeline_config)
     uvicorn.run(
-        "server_dashscope:app",
-        host="0.0.0.0",
-        port=pipeline_config.port,
-        reload=True,
+        create_dashscope_app(pipeline_manager=cli_pipeline_manager),
+        host=cli_pipeline_config.host,
+        port=cli_pipeline_config.port,
+        reload=False,
     )

lang_agent/front_api/build_server_utils.py

@@ -1,19 +1,26 @@
-from typing import Any, Dict, List
+from typing import Any, Dict, List, Literal, Optional
 import os
 import os.path as osp
 import subprocess
 import json
 
 from lang_agent.config.core_config import load_tyro_conf
+from lang_agent.config.constants import TY_BUILD_SCRIPT, _PROJECT_ROOT
 
-_PROJECT_ROOT = osp.dirname(osp.dirname(osp.dirname(osp.abspath(__file__))))
-_TY_BUILD_SCRIPT = osp.join(_PROJECT_ROOT, "lang_agent", "config", "ty_build_config.py")
+_DEEP_AGENT_BACKEND_ALIASES = {
+    "state_bk": "statebk",
+    "statebk": "statebk",
+    "local_shell": "localshell",
+    "localshell": "localshell",
+    "daytona_sandbox": "daytonasandbox",
+    "daytonasandbox": "daytonasandbox",
+}
 
 
 def opt_to_config(save_path: str, *nargs):
     os.makedirs(osp.dirname(save_path), exist_ok=True)
     subprocess.run(
-        ["python", _TY_BUILD_SCRIPT, "--save-path", save_path, *nargs],
+        ["python", TY_BUILD_SCRIPT, "--save-path", save_path, *nargs],
         check=True,
         cwd=_PROJECT_ROOT,
     )
@@ -22,7 +29,7 @@ def opt_to_config(save_path: str, *nargs):
 def _build_and_load_pipeline_config(
     pipeline_id: str, pipeline_config_dir: str, cmd: List[str]
 ):
-    save_config_f = osp.join(pipeline_config_dir, f"{pipeline_id}.yml")
+    save_config_f = osp.join(pipeline_config_dir, f"{pipeline_id}.yaml")
     opt_to_config(save_config_f, *cmd)
 
     # TODO: think if returning the built pipeline is better or just the config obj for front_api
@@ -52,7 +59,7 @@ def update_pipeline_registry(
     pipeline["enabled"] = bool(enabled)
     pipeline["config_file"] = config_file
     pipeline["graph_id"] = graph_id
-    pipeline["overrides"] = {"llm_name": llm_name}
+    pipeline["llm_name"] = llm_name
 
     with open(registry_f, "w", encoding="utf-8") as f:
         json.dump(registry, f, indent=4)
@@ -64,25 +71,46 @@ def build_route(
     tool_keys: List[str],
     api_key: str,
     llm_name: str = "qwen-plus",
-    pipeline_config_dir="configs/pipelines",
+    pipeline_config_dir: str = "configs/pipelines",
+    **_: Any,
 ):
     cmd_opt = [
+        "--pipeline.pipeline-id",
+        pipeline_id,
+        "--pipeline.llm-name", 
+        llm_name,
         "route",  # ------------
-        "--llm-name", llm_name,
-        "--api-key", api_key,
-        "--pipeline-id", pipeline_id,
-        "--prompt-set-id", prompt_set,
-        "tool_node",  # ------------
-        "--llm-name", llm_name,
-        "--api-key", api_key,
-        "--pipeline-id", pipeline_id,
-        "--prompt-set-id", prompt_set,
+        "--llm-name",
+        llm_name,
+        "--api-key",
+        api_key,
+        "--pipeline-id",
+        pipeline_id,
+        "--prompt-set-id",
+        prompt_set,
     ]
 
     if tool_keys:
         cmd_opt.extend(
             ["--tool-manager-config.client-tool-manager.tool-keys", *tool_keys]
         )
+        # Tyro parses list options greedily across positional subcommands; repeat a
+        # parent-level option to terminate list parsing before `tool_node`.
+        cmd_opt.extend(["--pipeline-id", pipeline_id])
+
+    cmd_opt.extend(
+        [
+            "tool_node",  # ------------
+            "--llm-name",
+            llm_name,
+            "--api-key",
+            api_key,
+            "--pipeline-id",
+            pipeline_id,
+            "--prompt-set-id",
+            prompt_set,
+        ]
+    )
 
     return _build_and_load_pipeline_config(pipeline_id, pipeline_config_dir, cmd_opt)
 
@@ -93,15 +121,24 @@ def build_react(
     tool_keys: List[str],
     api_key: str,
     llm_name: str = "qwen-plus",
-    pipeline_config_dir="configs/pipelines",
+    pipeline_config_dir: str = "configs/pipelines",
+    **_: Any,
 ):
     cmd_opt = [
+        "--pipeline.pipeline-id",
+        pipeline_id,
+        "--pipeline.llm-name", 
+        llm_name,
         "react",  # ------------
-        "--llm-name", llm_name,
-        "--api-key", api_key,
-        "--pipeline-id", pipeline_id,
-        "--prompt-set-id", prompt_set,
-        ]
+        "--llm-name",
+        llm_name,
+        "--api-key",
+        api_key,
+        "--pipeline-id",
+        pipeline_id,
+        "--prompt-set-id",
+        prompt_set,
+    ]
     if tool_keys:
         cmd_opt.extend(
             ["--tool-manager-config.client-tool-manager.tool-keys", *tool_keys]
@@ -110,8 +147,77 @@ def build_react(
     return _build_and_load_pipeline_config(pipeline_id, pipeline_config_dir, cmd_opt)
 
 
+def build_deep_agent(
+    pipeline_id: str,
+    prompt_set: str,
+    tool_keys: List[str],
+    api_key: str,
+    llm_name: str = "qwen-plus",
+    pipeline_config_dir: str = "configs/pipelines",
+    act_bkend: Literal[
+        "local_shell",
+        "localshell",
+        "state_bk",
+        "statebk",
+        "daytona_sandbox",
+        "daytonasandbox",
+    ] = "state_bk",
+    file_backend_config: Optional[Dict[str, Any]] = None,
+    **_: Any,
+):
+    backend_subcommand = _DEEP_AGENT_BACKEND_ALIASES.get(act_bkend)
+    if backend_subcommand is None:
+        raise ValueError(
+            "Unsupported deepagent backend "
+            f"'{act_bkend}'. Expected one of {sorted(_DEEP_AGENT_BACKEND_ALIASES.keys())}"
+        )
+
+    cmd_opt = [
+        "--pipeline.pipeline-id",
+        pipeline_id,
+        "--pipeline.llm-name", 
+        llm_name,
+        "deepagent",
+        "--llm-name",
+        llm_name,
+        "--api-key",
+        api_key,
+        "--pipeline-id",
+        pipeline_id,
+        "--prompt-set-id",
+        prompt_set,
+    ]
+
+    if tool_keys:
+        cmd_opt.extend(
+            ["--tool-manager-config.client-tool-manager.tool-keys", *tool_keys]
+        )
+        cmd_opt.extend(["--pipeline-id", pipeline_id])
+
+    cmd_opt.append(backend_subcommand)
+
+    if file_backend_config:
+        if "skills_dir" in file_backend_config and file_backend_config["skills_dir"]:
+            cmd_opt.extend(["--skills-dir", file_backend_config["skills_dir"]])
+        if (
+            "rt_skills_dir" in file_backend_config
+            and file_backend_config["rt_skills_dir"]
+        ):
+            cmd_opt.extend(["--rt-skills-dir", file_backend_config["rt_skills_dir"]])
+        if (
+            "workspace_dir" in file_backend_config
+            and file_backend_config["workspace_dir"]
+        ):
+            cmd_opt.extend(["--workspace-dir", file_backend_config["workspace_dir"]])
+        if "api_key" in file_backend_config and file_backend_config["api_key"]:
+            cmd_opt.extend(["--api-key", file_backend_config["api_key"]])
+
+    return _build_and_load_pipeline_config(pipeline_id, pipeline_config_dir, cmd_opt)
+
+
 # {pipeline_id: build_function}
 GRAPH_BUILD_FNCS = {
     "routing": build_route,
     "react": build_react,
+    "deepagent": build_deep_agent,
 }

lang_agent/fs_bkends/base.py

@@ -1,8 +1,10 @@
-from dataclasses import dataclass, field, is_dataclass
+import os
+from dataclasses import dataclass
 from typing import Any
-import tyro
-import os.path as osp
 from abc import ABC, abstractmethod
+from loguru import logger
+
+from lang_agent.config import InstantiateConfig
 
 
 class BaseFilesystemBackend(ABC):
@@ -25,4 +27,25 @@ class BaseFilesystemBackend(ABC):
         if hasattr(self.config, "rt_skills_dir"):
             return {"skills" : [self.config.rt_skills_dir]}
         else:
-            return {}
+            return {}
+
+
+@dataclass
+class FilesystemBackendConfig(InstantiateConfig):
+    """
+    Shared filesystem backend config behavior.
+    If subclasses define these fields, this hook ensures they exist:
+      - skills_dir
+      - workspace_dir
+    """
+
+    def _ensure_dir_if_present(self, attr_name: str) -> None:
+        path = getattr(self, attr_name, None)
+        if not isinstance(path, str) or not path.strip():
+            return
+        os.makedirs(path, exist_ok=True)
+        logger.info(f"Ensured {attr_name} exists: {path}")
+
+    def __post_init__(self) -> None:
+        self._ensure_dir_if_present("skills_dir")
+        self._ensure_dir_if_present("workspace_dir")

lang_agent/fs_bkends/daytona_sandbox.py

@@ -8,13 +8,12 @@ from loguru import logger
 from daytona import Daytona, DaytonaConfig, FileUpload
 from langchain_daytona import DaytonaSandbox
 
-from lang_agent.config import InstantiateConfig
-from lang_agent.fs_bkends import BaseFilesystemBackend
+from lang_agent.fs_bkends.base import BaseFilesystemBackend, FilesystemBackendConfig
 
 
 @tyro.conf.configure(tyro.conf.SuppressFixed)
 @dataclass
-class DaytonaSandboxConfig(InstantiateConfig):
+class DaytonaSandboxConfig(FilesystemBackendConfig):
     _target: Type = field(default_factory=lambda: DaytonaSandboxBk)
 
     api_key: Optional[str] = None
@@ -27,6 +26,7 @@ class DaytonaSandboxConfig(InstantiateConfig):
     """runtime skills path inside the sandbox (auto-set from sandbox workdir)"""
 
     def __post_init__(self):
+        super().__post_init__()
         if self.api_key is None:
             self.api_key = os.environ.get("DAYTONA_API_KEY")
             if self.api_key is None:

lang_agent/fs_bkends/localshell.py

@@ -1,21 +1,16 @@
-from dataclasses import dataclass, field, is_dataclass
-from typing import Type, TypedDict, Literal, Dict, List, Tuple, Optional
+from dataclasses import dataclass, field
+from typing import Type
 import tyro
-import os.path as osp
-from abc import ABC, abstractmethod
-import glob
 from loguru import logger
 
-from deepagents.backends.utils import create_file_data
 from deepagents.backends import LocalShellBackend
 
-from lang_agent.config import InstantiateConfig
-from lang_agent.fs_bkends import BaseFilesystemBackend
+from lang_agent.fs_bkends.base import BaseFilesystemBackend, FilesystemBackendConfig
 
 
 @tyro.conf.configure(tyro.conf.SuppressFixed)
 @dataclass
-class LocalShellConfig(InstantiateConfig):
+class LocalShellConfig(FilesystemBackendConfig):
     _target:Type = field(default_factory=lambda:LocalShell)
 
     workspace_dir:str = "./workspace"
@@ -38,4 +33,24 @@ class LocalShell(BaseFilesystemBackend):
         self.backend = LocalShellBackend(root_dir=self.config.workspace_dir,
                                          virtual_mode=True,
                                         #  env={"PATH": "/usr/bin:/bin"}
-                                         inherit_env=True)
+                                         inherit_env=True)
+
+
+
+if __name__ == "__main__":
+    import sys
+
+    # Instantiate a LocalShell instance with the default config
+    config = LocalShellConfig()
+    shell = LocalShell(config)
+
+    # Try checking access to 'npx'
+    try:
+        result = shell.backend.execute("npx --version")
+        if result.exit_code == 0:
+            print("npx is available, version:", result.output.strip())
+        else:
+            print("npx returned non-zero exit code:", result.exit_code, file=sys.stderr)
+            print("output:", result.output, file=sys.stderr)
+    except Exception as e:
+        print("Could not access 'npx':", str(e), file=sys.stderr)

lang_agent/fs_bkends/statebk.py

@@ -1,16 +1,14 @@
-from dataclasses import dataclass, field, is_dataclass
-from typing import Type, TypedDict, Literal, Dict, List, Tuple, Optional
+from dataclasses import dataclass, field
+from typing import Type
 import tyro
 import os.path as osp
-from abc import ABC, abstractmethod
 import glob
 from loguru import logger
 
 from deepagents.backends.utils import create_file_data
 from deepagents.backends import StateBackend
 
-from lang_agent.config import InstantiateConfig
-from lang_agent.fs_bkends import BaseFilesystemBackend
+from lang_agent.fs_bkends.base import BaseFilesystemBackend, FilesystemBackendConfig
 
 def read_as_utf8(file_path:str):
     with open(file_path, "r", encoding="utf-8") as f:
@@ -31,7 +29,7 @@ def build_skill_fs_dict(skill_dir:str, virt_path:str="/skills"):
 
 @tyro.conf.configure(tyro.conf.SuppressFixed)
 @dataclass
-class StateBkConfig(InstantiateConfig):
+class StateBkConfig(FilesystemBackendConfig):
     _target:Type = field(default_factory=lambda:StateBk)
 
     skills_dir:str = "./assets/skills"
@@ -40,10 +38,6 @@ class StateBkConfig(InstantiateConfig):
     rt_skills_dir:str = "/skills"
     """path to directory with skills in runtime directory"""
 
-    def __post_init__(self):
-        err_msg = f"{self.skills_dir} does not exist"
-        assert osp.exists(self.skills_dir), err_msg
-
 
 class StateBk(BaseFilesystemBackend):
     def __init__(self, config:StateBkConfig):

lang_agent/graphs/deepagents_qt.py

@@ -61,7 +61,12 @@ class DeepAgent(GraphBase):
                                             checkpointer=self.mem,
                                             **bkend_agent_params)
         
-        self.prompt_store = build_prompt_store(file_path=self.config.sys_prompt_f, default_key="sys_prompt")
+        self.prompt_store = build_prompt_store(
+            pipeline_id=self.config.pipeline_id,
+            prompt_set_id=self.config.prompt_set_id,
+            file_path=self.config.sys_prompt_f,
+            default_key="sys_prompt",
+        )
         self.sys_prompt = self.prompt_store.get("sys_prompt")
 
     def _agent_call(self, state:State):

lang_agent/pipeline.py

@@ -13,7 +13,7 @@ from langchain_core.messages import SystemMessage, HumanMessage, BaseMessage
 from langchain.agents import create_agent
 from langgraph.checkpoint.memory import MemorySaver
 
-from lang_agent.config import LLMNodeConfig, load_tyro_conf
+from lang_agent.config import LLMNodeConfig, load_tyro_conf, resolve_llm_api_key
 from lang_agent.graphs import AnnotatedGraph, ReactGraphConfig, RoutingConfig
 from lang_agent.base import GraphBase
 from lang_agent.components import conv_store
@@ -62,7 +62,7 @@ class PipelineConfig(LLMNodeConfig):
     host: str = "0.0.0.0"
     """where am I hosted"""
 
-    port: int = 8588
+    port: int = 8500
     """what is my port"""
 
     # graph_config: AnnotatedGraph = field(default_factory=ReactGraphConfig)
@@ -104,7 +104,13 @@ class Pipeline:
                 if self.config.base_url is not None
                 else self.config.graph_config.base_url
             )
-            self.config.graph_config.api_key = self.config.api_key
+            pipeline_api_key = resolve_llm_api_key(self.config.api_key)
+            graph_api_key = resolve_llm_api_key(
+                getattr(self.config.graph_config, "api_key", None)
+            )
+            resolved_api_key = pipeline_api_key or graph_api_key
+            self.config.api_key = resolved_api_key
+            self.config.graph_config.api_key = resolved_api_key
 
         self.graph: GraphBase = self.config.graph_config.setup()
 

nginx.conf

@@ -0,0 +1,85 @@
+events {
+    worker_connections 1024;
+}
+
+http {
+    include       /etc/nginx/mime.types;
+    default_type  application/octet-stream;
+
+    sendfile        on;
+    keepalive_timeout  65;
+
+    # Upstream backend
+    upstream backend {
+        server backend:8500;
+    }
+
+    server {
+        listen 80;
+        server_name localhost;
+
+        root /usr/share/nginx/html;
+        index index.html;
+
+        # Always revalidate the SPA entrypoint so clients pick up the latest
+        # hashed JS bundle after redeploys.
+        location = /index.html {
+            add_header Cache-Control "no-store, no-cache, must-revalidate, proxy-revalidate" always;
+            add_header Pragma "no-cache" always;
+            add_header Expires "0" always;
+            try_files $uri =404;
+        }
+
+        # Serve frontend static files
+        location / {
+            try_files $uri $uri/ /index.html;
+        }
+
+        # Proxy API requests to backend
+        location /v1/ {
+            proxy_pass http://backend;
+            proxy_http_version 1.1;
+            proxy_set_header Upgrade $http_upgrade;
+            proxy_set_header Connection 'upgrade';
+            proxy_set_header Host $host;
+            proxy_set_header X-Real-IP $remote_addr;
+            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+            proxy_set_header X-Forwarded-Proto $scheme;
+            proxy_cache_bypass $http_upgrade;
+        }
+
+        # Proxy DashScope API requests
+        location /apps/ {
+            proxy_pass http://backend;
+            proxy_http_version 1.1;
+            proxy_set_header Upgrade $http_upgrade;
+            proxy_set_header Connection 'upgrade';
+            proxy_set_header Host $host;
+            proxy_set_header X-Real-IP $remote_addr;
+            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+            proxy_set_header X-Forwarded-Proto $scheme;
+            proxy_cache_bypass $http_upgrade;
+        }
+
+        # Proxy v1/apps requests
+        location /v1/apps/ {
+            proxy_pass http://backend;
+            proxy_http_version 1.1;
+            proxy_set_header Upgrade $http_upgrade;
+            proxy_set_header Connection 'upgrade';
+            proxy_set_header Host $host;
+            proxy_set_header X-Real-IP $remote_addr;
+            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+            proxy_set_header X-Forwarded-Proto $scheme;
+            proxy_cache_bypass $http_upgrade;
+        }
+
+        # Health check endpoint
+        location /health {
+            proxy_pass http://backend/health;
+            access_log off;
+        }
+    }
+}
+
+

pyproject.toml

@@ -24,7 +24,10 @@ dependencies = [
     "commentjson",
     "pandas",
     "asgiref",
-    "psycopg[binary]"
+    "psycopg[binary]",
+    "deepagents",
+    "daytona",
+    "langchain_daytona"
 ]
 
 [tool.setuptools.packages.find]

scripts/init_database/00_init_user.sh

@@ -0,0 +1,40 @@
+#!/bin/bash
+# Initialize database user and database
+# This script runs before SQL files in docker-entrypoint-initdb.d
+# It must be named with 00_ prefix to run first
+
+set -e
+
+APP_DB_NAME="${APP_DB_NAME:-ai_conversations}"
+APP_DB_USER="${APP_DB_USER:-myapp_user}"
+APP_DB_PASSWORD="${APP_DB_PASSWORD:-secure_password_123}"
+
+echo "Creating database user: $APP_DB_USER"
+# Create user
+psql -v ON_ERROR_STOP=1 --username "$POSTGRES_USER" --dbname "$POSTGRES_DB" <<-EOSQL
+    DO \$\$
+    BEGIN
+        IF NOT EXISTS (SELECT FROM pg_catalog.pg_user WHERE usename = '$APP_DB_USER') THEN
+            CREATE USER $APP_DB_USER WITH PASSWORD '$APP_DB_PASSWORD';
+        END IF;
+    END
+    \$\$;
+    ALTER USER $APP_DB_USER CREATEDB;
+EOSQL
+
+echo "Creating database: $APP_DB_NAME"
+# Create database
+psql -v ON_ERROR_STOP=1 --username "$POSTGRES_USER" --dbname "$POSTGRES_DB" <<-EOSQL
+    SELECT 'CREATE DATABASE $APP_DB_NAME'
+    WHERE NOT EXISTS (SELECT FROM pg_database WHERE datname = '$APP_DB_NAME')\gexec
+    GRANT ALL PRIVILEGES ON DATABASE $APP_DB_NAME TO $APP_DB_USER;
+EOSQL
+
+echo "Granting schema privileges"
+# Grant schema privileges
+psql -v ON_ERROR_STOP=1 --username "$POSTGRES_USER" --dbname "$APP_DB_NAME" <<-EOSQL
+    GRANT ALL ON SCHEMA public TO $APP_DB_USER;
+EOSQL
+
+echo "Database initialization complete!"
+

scripts/init_database/01_run_sql_files.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+# Run SQL initialization files in the correct database context
+# This script runs after 00_init_user.sh creates the database
+
+set -e
+
+APP_DB_NAME="${APP_DB_NAME:-ai_conversations}"
+
+echo "Running SQL initialization files in database: $APP_DB_NAME"
+
+# Run create_conv_store.sql
+if [ -f /docker-entrypoint-initdb.d/create_conv_store.sql ]; then
+    echo "Executing create_conv_store.sql..."
+    psql -v ON_ERROR_STOP=1 --username "$POSTGRES_USER" --dbname "$APP_DB_NAME" -f /docker-entrypoint-initdb.d/create_conv_store.sql
+fi
+
+# Run create_prompt_config.sql
+if [ -f /docker-entrypoint-initdb.d/create_prompt_config.sql ]; then
+    echo "Executing create_prompt_config.sql..."
+    psql -v ON_ERROR_STOP=1 --username "$POSTGRES_USER" --dbname "$APP_DB_NAME" -f /docker-entrypoint-initdb.d/create_prompt_config.sql
+fi
+
+echo "SQL initialization files completed!"
+
+

scripts/init_database/create_conv_store.sql

@@ -1,4 +1,5 @@
 -- Create the messages table
+-- This script runs in the ai_conversations database context
 CREATE TABLE IF NOT EXISTS messages (
     id BIGSERIAL PRIMARY KEY,
     conversation_id TEXT NOT NULL,
@@ -13,4 +14,8 @@ CREATE TABLE IF NOT EXISTS messages (
 CREATE INDEX IF NOT EXISTS idx_messages_conversation ON messages (conversation_id, sequence_number);
 
 -- Index for fast lookup by pipeline_id
-CREATE INDEX IF NOT EXISTS idx_messages_pipeline ON messages (pipeline_id);
+CREATE INDEX IF NOT EXISTS idx_messages_pipeline ON messages (pipeline_id);
+
+-- Grant permissions to app user
+GRANT ALL PRIVILEGES ON ALL TABLES IN SCHEMA public TO myapp_user;
+GRANT ALL PRIVILEGES ON ALL SEQUENCES IN SCHEMA public TO myapp_user;

scripts/init_database/create_prompt_config.sql

@@ -1,4 +1,5 @@
 -- A prompt_set groups a full collection of prompts together.
+-- This script runs in the ai_conversations database context
 -- Each pipeline can have many sets (versions, A/B variants, etc.);
 -- exactly one should be marked is_active per pipeline.
 CREATE TABLE IF NOT EXISTS prompt_sets (
@@ -41,6 +42,10 @@ CREATE TABLE IF NOT EXISTS prompt_templates (
 CREATE INDEX IF NOT EXISTS idx_prompt_templates_set_id
     ON prompt_templates(prompt_set_id);
 
+-- Grant permissions to app user
+GRANT ALL PRIVILEGES ON ALL TABLES IN SCHEMA public TO myapp_user;
+GRANT ALL PRIVILEGES ON ALL SEQUENCES IN SCHEMA public TO myapp_user;
+
 -- Seed: initial prompt set for lang_agent/graphs/routing.py
 -- The pipeline_id can be used by RoutingConfig.pipeline_id to load these prompts.
 INSERT INTO prompt_sets (pipeline_id, graph_id, name, description, is_active, list)
@@ -160,3 +165,30 @@ ON CONFLICT (prompt_set_id, prompt_key)
 DO UPDATE SET
     content = EXCLUDED.content,
     updated_at = now();
+
+-- Seed: initial prompt set for lang_agent/graphs/deepagents_qt.py
+-- DeepAgent uses prompt key "sys_prompt" with DB-first, file-fallback loading.
+INSERT INTO prompt_sets (pipeline_id, graph_id, name, description, is_active, list)
+SELECT
+    'deepagent',
+    'deepagent',
+    'default',
+    'Initial prompt set for DeepAgent',
+    true,
+    ''
+WHERE NOT EXISTS (
+    SELECT 1
+    FROM prompt_sets
+    WHERE pipeline_id = 'deepagent'
+      AND name = 'default'
+);
+
+INSERT INTO prompt_templates (prompt_set_id, prompt_key, content)
+SELECT ps.id, 'sys_prompt', '你是一个擅长调用工具和处理文件任务的深度代理。'
+FROM prompt_sets ps
+WHERE ps.pipeline_id = 'deepagent'
+  AND ps.name = 'default'
+ON CONFLICT (prompt_set_id, prompt_key)
+DO UPDATE SET
+    content = EXCLUDED.content,
+    updated_at = now();

scripts/init_database/init_all.sh

@@ -0,0 +1,49 @@
+#!/bin/bash
+# Database initialization script
+# This script runs all SQL initialization files in the correct order
+
+set -e
+
+DB_NAME="${POSTGRES_DB:-ai_conversations}"
+DB_USER="${POSTGRES_USER:-myapp_user}"
+DB_PASSWORD="${POSTGRES_PASSWORD:-secure_password_123}"
+DB_HOST="${POSTGRES_HOST:-localhost}"
+DB_PORT="${POSTGRES_PORT:-5432}"
+
+export PGPASSWORD="$DB_PASSWORD"
+
+echo "Initializing database: $DB_NAME on $DB_HOST:$DB_PORT"
+
+# Wait for PostgreSQL to be ready
+until psql -h "$DB_HOST" -p "$DB_PORT" -U "$DB_USER" -d postgres -c '\q' 2>/dev/null; do
+  echo "Waiting for PostgreSQL to be ready..."
+  sleep 2
+done
+
+echo "PostgreSQL is ready!"
+
+# Create database if it doesn't exist
+psql -h "$DB_HOST" -p "$DB_PORT" -U "$DB_USER" -d postgres <<EOF
+SELECT 'CREATE DATABASE $DB_NAME'
+WHERE NOT EXISTS (SELECT FROM pg_database WHERE datname = '$DB_NAME')\gexec
+EOF
+
+# Grant privileges
+psql -h "$DB_HOST" -p "$DB_PORT" -U "$DB_USER" -d postgres <<EOF
+GRANT ALL PRIVILEGES ON DATABASE $DB_NAME TO $DB_USER;
+EOF
+
+# Run initialization scripts in order
+echo "Running database initialization scripts..."
+
+# 1. Create conversation store tables
+echo "Creating conversation store tables..."
+psql -h "$DB_HOST" -p "$DB_PORT" -U "$DB_USER" -d "$DB_NAME" -f /docker-entrypoint-initdb.d/create_conv_store.sql
+
+# 2. Create prompt configuration tables
+echo "Creating prompt configuration tables..."
+psql -h "$DB_HOST" -p "$DB_PORT" -U "$DB_USER" -d "$DB_NAME" -f /docker-entrypoint-initdb.d/create_prompt_config.sql
+
+echo "Database initialization complete!"
+
+

scripts/py_scripts/chat_dashcope.py

@@ -0,0 +1,141 @@
+#!/usr/bin/env python3
+"""
+Simple chat loop to interact with the blueberry pipeline via DashScope-compatible API.
+
+Usage:
+    python scripts/py_scripts/chat_dashcope.py
+
+The script connects to the server running on http://localhost:8500
+and uses the API key from the pipeline registry.
+"""
+
+import requests
+import json
+import sys
+from typing import Optional
+
+
+# Configuration from pipeline_registry.json
+API_KEY = "sk-6c7091e6a95f404efb2ec30e8f51b897626d670375cdf822d78262f24ab12367"
+PIPELINE_ID = "blueberry"
+BASE_URL = "http://localhost:8500"
+SESSION_ID = "chat-session-1"
+
+
+def send_message(
+    message: str,
+    session_id: str = SESSION_ID,
+    stream: bool = False,
+    app_id: str = PIPELINE_ID,
+) -> Optional[str]:
+    """Send a message to the blueberry pipeline and return the response."""
+    url = f"{BASE_URL}/v1/apps/{app_id}/sessions/{session_id}/responses"
+    headers = {
+        "Authorization": f"Bearer {API_KEY}",
+        "Content-Type": "application/json",
+    }
+    
+    payload = {
+        "messages": [
+            {"role": "user", "content": message}
+        ],
+        "stream": stream,
+    }
+    
+    try:
+        if stream:
+            # Handle streaming response
+            response = requests.post(url, headers=headers, json=payload, stream=True)
+            response.raise_for_status()
+            
+            accumulated_text = ""
+            for line in response.iter_lines():
+                if line:
+                    line_str = line.decode('utf-8')
+                    if line_str.startswith('data: '):
+                        data_str = line_str[6:]  # Remove 'data: ' prefix
+                        try:
+                            data = json.loads(data_str)
+                            output = data.get("output", {})
+                            text = output.get("text", "")
+                            if text:
+                                accumulated_text = text
+                                # Print incremental updates (you can modify this behavior)
+                                print(f"\rAssistant: {accumulated_text}", end="", flush=True)
+                            
+                            if data.get("is_end", False):
+                                print()  # New line after streaming completes
+                                return accumulated_text
+                        except json.JSONDecodeError:
+                            continue
+            return accumulated_text
+        else:
+            # Handle non-streaming response
+            response = requests.post(url, headers=headers, json=payload)
+            response.raise_for_status()
+            data = response.json()
+            output = data.get("output", {})
+            return output.get("text", "")
+            
+    except requests.exceptions.RequestException as e:
+        print(f"Error sending message: {e}", file=sys.stderr)
+        if hasattr(e, 'response') and e.response is not None:
+            try:
+                error_detail = e.response.json()
+                print(f"Error details: {error_detail}", file=sys.stderr)
+            except:
+                print(f"Response status: {e.response.status_code}", file=sys.stderr)
+        return None
+
+
+def main():
+    """Main chat loop."""
+    print("=" * 60)
+    print(f"Chat with Blueberry Pipeline")
+    print(f"Pipeline ID: {PIPELINE_ID}")
+    print(f"Server: {BASE_URL}")
+    print(f"Session ID: {SESSION_ID}")
+    print("=" * 60)
+    print("Type your messages (or 'quit'/'exit' to end, 'stream' to toggle streaming)")
+    print("Streaming mode is ON by default")
+    print()
+    
+    stream_mode = True
+    
+    while True:
+        try:
+            user_input = input("You: ").strip()
+            
+            if not user_input:
+                continue
+                
+            if user_input.lower() in ['quit', 'exit', 'q']:
+                print("Goodbye!")
+                break
+                
+            if user_input.lower() == 'stream':
+                stream_mode = not stream_mode
+                print(f"Streaming mode: {'ON' if stream_mode else 'OFF'}")
+                continue
+            
+            print("Assistant: ", end="", flush=True)
+            response = send_message(user_input, stream=stream_mode)
+            
+            if response is None:
+                print("(No response received)")
+            elif not stream_mode:
+                print(response)
+            # For streaming, the response is already printed incrementally
+            
+            print()  # Empty line for readability
+            
+        except KeyboardInterrupt:
+            print("\n\nGoodbye!")
+            break
+        except Exception as e:
+            print(f"\nError: {e}", file=sys.stderr)
+
+
+if __name__ == "__main__":
+    main()
+

scripts/py_scripts/migrate_yaml_prompts_to_db.py

@@ -0,0 +1,364 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import datetime as dt
+import glob
+import os
+import os.path as osp
+import sys
+from dataclasses import dataclass
+from typing import Dict, Iterable, List, Optional
+
+import commentjson
+import psycopg
+
+
+PROJECT_ROOT = osp.dirname(osp.dirname(osp.dirname(osp.abspath(__file__))))
+if PROJECT_ROOT not in sys.path:
+    sys.path.append(PROJECT_ROOT)
+
+from lang_agent.config import load_tyro_conf  # noqa: E402
+from lang_agent.config.db_config_manager import DBConfigManager  # noqa: E402
+
+
+@dataclass
+class MigrationPayload:
+    config_path: str
+    pipeline_id: str
+    graph_id: str
+    prompt_dict: Dict[str, str]
+    tool_keys: List[str]
+    api_key: Optional[str]
+
+
+def _infer_pipeline_id(pipeline_conf, config_path: str) -> str:
+    candidates = [
+        getattr(pipeline_conf, "pipeline_id", None),
+        getattr(getattr(pipeline_conf, "graph_config", None), "pipeline_id", None),
+    ]
+    for candidate in candidates:
+        if candidate is None:
+            continue
+        value = str(candidate).strip()
+        if value and value.lower() != "null":
+            return value
+    return osp.splitext(osp.basename(config_path))[0]
+
+
+def _infer_graph_id(graph_conf) -> str:
+    if graph_conf is None:
+        return "unknown"
+    class_name = graph_conf.__class__.__name__.lower()
+    if "routing" in class_name or class_name == "routeconfig":
+        return "routing"
+    if "react" in class_name:
+        return "react"
+
+    target = getattr(graph_conf, "_target", None)
+    if target is not None:
+        target_name = getattr(target, "__name__", str(target)).lower()
+        if "routing" in target_name:
+            return "routing"
+        if "react" in target_name:
+            return "react"
+    return "unknown"
+
+
+def _extract_tool_keys(graph_conf) -> List[str]:
+    if graph_conf is None:
+        return []
+    tool_cfg = getattr(graph_conf, "tool_manager_config", None)
+    client_cfg = getattr(tool_cfg, "client_tool_manager", None)
+    keys = getattr(client_cfg, "tool_keys", None)
+    if not keys:
+        return []
+    out: List[str] = []
+    seen = set()
+    for key in keys:
+        cleaned = str(key).strip()
+        if not cleaned or cleaned in seen:
+            continue
+        seen.add(cleaned)
+        out.append(cleaned)
+    return out
+
+
+def _load_prompt_dict(prompt_path: str, default_key: str = "sys_prompt") -> Dict[str, str]:
+    if not prompt_path:
+        return {}
+    if not osp.exists(prompt_path):
+        return {}
+
+    if osp.isdir(prompt_path):
+        prompt_files = sorted(
+            p for p in glob.glob(osp.join(prompt_path, "*.txt")) if "optional" not in p
+        )
+        out = {}
+        for prompt_f in prompt_files:
+            key = osp.splitext(osp.basename(prompt_f))[0]
+            with open(prompt_f, "r", encoding="utf-8") as f:
+                out[key] = f.read()
+        return out
+
+    if prompt_path.endswith(".json"):
+        with open(prompt_path, "r", encoding="utf-8") as f:
+            obj = commentjson.load(f)
+        if not isinstance(obj, dict):
+            return {}
+        return {str(k): v if isinstance(v, str) else str(v) for k, v in obj.items()}
+
+    if prompt_path.endswith(".txt"):
+        with open(prompt_path, "r", encoding="utf-8") as f:
+            return {default_key: f.read()}
+
+    return {}
+
+
+def _extract_prompt_dict(graph_conf) -> Dict[str, str]:
+    if graph_conf is None:
+        return {}
+    if hasattr(graph_conf, "sys_prompt_f"):
+        return _load_prompt_dict(str(getattr(graph_conf, "sys_prompt_f")), "sys_prompt")
+    if hasattr(graph_conf, "sys_promp_dir"):
+        return _load_prompt_dict(str(getattr(graph_conf, "sys_promp_dir")))
+    return {}
+
+
+def _extract_tool_node_prompt_dict(graph_conf) -> Dict[str, str]:
+    tool_node_conf = getattr(graph_conf, "tool_node_config", None)
+    if tool_node_conf is None:
+        return {}
+
+    out: Dict[str, str] = {}
+    if hasattr(tool_node_conf, "tool_prompt_f"):
+        out.update(
+            _load_prompt_dict(str(getattr(tool_node_conf, "tool_prompt_f")), "tool_prompt")
+        )
+    if hasattr(tool_node_conf, "chatty_sys_prompt_f"):
+        out.update(
+            _load_prompt_dict(
+                str(getattr(tool_node_conf, "chatty_sys_prompt_f")), "chatty_prompt"
+            )
+        )
+    return out
+
+
+def _prompt_key_whitelist(graph_conf, graph_id: str) -> Optional[set]:
+    if graph_id == "react":
+        return {"sys_prompt"}
+    if graph_id != "routing":
+        return None
+
+    allowed = {"route_prompt", "chat_prompt", "tool_prompt"}
+    tool_node_conf = getattr(graph_conf, "tool_node_config", None)
+    if tool_node_conf is None:
+        return allowed
+
+    cls_name = tool_node_conf.__class__.__name__.lower()
+    target = getattr(tool_node_conf, "_target", None)
+    target_name = getattr(target, "__name__", str(target)).lower() if target else ""
+    if "chatty" in cls_name or "chatty" in target_name:
+        allowed.add("chatty_prompt")
+    return allowed
+
+
+def _collect_payload(config_path: str) -> MigrationPayload:
+    conf = load_tyro_conf(config_path)
+    graph_conf = getattr(conf, "graph_config", None)
+    graph_id = _infer_graph_id(graph_conf)
+    prompt_dict = _extract_prompt_dict(graph_conf)
+    prompt_dict.update(_extract_tool_node_prompt_dict(graph_conf))
+    whitelist = _prompt_key_whitelist(graph_conf, graph_id)
+    if whitelist is not None:
+        prompt_dict = {k: v for k, v in prompt_dict.items() if k in whitelist}
+    return MigrationPayload(
+        config_path=config_path,
+        pipeline_id=_infer_pipeline_id(conf, config_path),
+        graph_id=graph_id,
+        prompt_dict=prompt_dict,
+        tool_keys=_extract_tool_keys(graph_conf),
+        api_key=getattr(conf, "api_key", None),
+    )
+
+
+def _resolve_config_paths(config_dir: str, config_paths: Optional[Iterable[str]]) -> List[str]:
+    if config_paths:
+        resolved = [osp.abspath(path) for path in config_paths]
+    else:
+        pattern = osp.join(osp.abspath(config_dir), "*.yaml")
+        resolved = sorted(glob.glob(pattern))
+    return [path for path in resolved if osp.exists(path)]
+
+
+def _ensure_prompt_set(
+    conn: psycopg.Connection,
+    pipeline_id: str,
+    graph_id: str,
+    set_name: str,
+    description: str,
+) -> str:
+    with conn.cursor() as cur:
+        cur.execute(
+            """
+            SELECT id FROM prompt_sets
+            WHERE pipeline_id = %s AND name = %s
+            ORDER BY updated_at DESC, created_at DESC
+            LIMIT 1
+            """,
+            (pipeline_id, set_name),
+        )
+        row = cur.fetchone()
+        if row is not None:
+            return str(row[0])
+
+        cur.execute(
+            """
+            INSERT INTO prompt_sets (pipeline_id, graph_id, name, description, is_active, list)
+            VALUES (%s, %s, %s, %s, false, '')
+            RETURNING id
+            """,
+            (pipeline_id, graph_id, set_name, description),
+        )
+        created = cur.fetchone()
+        return str(created[0])
+
+
+def _activate_prompt_set(conn: psycopg.Connection, pipeline_id: str, prompt_set_id: str) -> None:
+    with conn.cursor() as cur:
+        cur.execute(
+            "UPDATE prompt_sets SET is_active = false, updated_at = now() WHERE pipeline_id = %s",
+            (pipeline_id,),
+        )
+        cur.execute(
+            "UPDATE prompt_sets SET is_active = true, updated_at = now() WHERE id = %s",
+            (prompt_set_id,),
+        )
+
+
+def _run_migration(
+    payloads: List[MigrationPayload],
+    set_name: str,
+    description: str,
+    dry_run: bool,
+    activate: bool,
+) -> None:
+    for payload in payloads:
+        print(
+            f"[PLAN] pipeline={payload.pipeline_id} graph={payload.graph_id} "
+            f"prompts={len(payload.prompt_dict)} tools={len(payload.tool_keys)} "
+            f"config={payload.config_path}"
+        )
+        if dry_run:
+            continue
+
+        manager = DBConfigManager()
+        with psycopg.connect(manager.conn_str) as conn:
+            prompt_set_id = _ensure_prompt_set(
+                conn=conn,
+                pipeline_id=payload.pipeline_id,
+                graph_id=payload.graph_id,
+                set_name=set_name,
+                description=description,
+            )
+            conn.commit()
+
+            manager.set_config(
+                pipeline_id=payload.pipeline_id,
+                graph_id=payload.graph_id,
+                prompt_set_id=prompt_set_id,
+                tool_list=payload.tool_keys,
+                prompt_dict=payload.prompt_dict,
+                api_key=payload.api_key,
+            )
+
+            if activate:
+                _activate_prompt_set(
+                    conn=conn,
+                    pipeline_id=payload.pipeline_id,
+                    prompt_set_id=prompt_set_id,
+                )
+                conn.commit()
+
+        print(
+            f"[DONE] pipeline={payload.pipeline_id} "
+            f"prompt_set={prompt_set_id} activate={activate}"
+        )
+
+
+def main() -> None:
+    date_str = dt.date.today().isoformat()
+    parser = argparse.ArgumentParser(
+        description="Import prompt definitions from pipeline YAML files into DB prompt_sets."
+    )
+    parser.add_argument(
+        "--config-dir",
+        default=osp.join(PROJECT_ROOT, "configs", "pipelines"),
+        help="Directory containing pipeline YAML files.",
+    )
+    parser.add_argument(
+        "--config",
+        action="append",
+        default=[],
+        help="Specific pipeline config yaml path. Can be passed multiple times.",
+    )
+    parser.add_argument(
+        "--pipeline-id",
+        action="append",
+        default=[],
+        help="If provided, only migrate these pipeline IDs (repeatable).",
+    )
+    parser.add_argument(
+        "--set-name",
+        # default=f"migrated-{date_str}",
+        default="default",
+        help="Prompt set name to create/reuse under each pipeline.",
+    )
+    parser.add_argument(
+        "--description",
+        default="Migrated from pipeline YAML prompt files",
+        help="Prompt set description.",
+    )
+    parser.add_argument(
+        "--dry-run",
+        action="store_true",
+        help="Print what would be migrated without writing to DB.",
+    )
+    parser.add_argument(
+        "--activate",
+        action="store_true",
+        help="Mark imported set active for each migrated pipeline.",
+    )
+    args = parser.parse_args()
+
+    config_paths = _resolve_config_paths(args.config_dir, args.config)
+    if not config_paths:
+        raise SystemExit("No config files found. Provide --config or --config-dir.")
+
+    requested_pipelines = {p.strip() for p in args.pipeline_id if p.strip()}
+
+    payloads: List[MigrationPayload] = []
+    for config_path in config_paths:
+        payload = _collect_payload(config_path)
+        if requested_pipelines and payload.pipeline_id not in requested_pipelines:
+            continue
+        if not payload.prompt_dict:
+            print(f"[SKIP] no prompts found for config={config_path}")
+            continue
+        payloads.append(payload)
+
+    if not payloads:
+        raise SystemExit("No pipelines matched with prompt content to migrate.")
+
+    _run_migration(
+        payloads=payloads,
+        set_name=args.set_name,
+        description=args.description,
+        dry_run=args.dry_run,
+        activate=args.activate,
+    )
+
+
+if __name__ == "__main__":
+    main()
+

scripts/shell_scripts/create_config.sh

@@ -1,3 +1,8 @@
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(dirname "$(dirname "$SCRIPT_DIR")")"
+
+cd "$PROJECT_ROOT"
+
 source ~/.bashrc
 conda init
 conda activate lang

scripts/shell_scripts/download_images.sh

@@ -0,0 +1,37 @@
+#!/bin/bash
+# Script to download and package Docker images for offline use
+# Run this on a machine with good Docker Hub access, then transfer images.tar to China
+
+set -e
+
+echo "=== Docker Image Downloader for Offline Use ==="
+echo ""
+
+# Images needed
+IMAGES=(
+    "node:20-alpine"
+    "python:3.12-slim"
+    "postgres:16-alpine"
+    "nginx:alpine"
+)
+
+OUTPUT_FILE="images.tar"
+
+echo "Pulling Docker images..."
+for img in "${IMAGES[@]}"; do
+    echo "  Pulling $img..."
+    docker pull "$img"
+done
+
+echo ""
+echo "Saving to $OUTPUT_FILE..."
+docker save "${IMAGES[@]}" -o "$OUTPUT_FILE"
+
+echo ""
+echo "Done! File size:"
+ls -lh "$OUTPUT_FILE"
+
+echo ""
+echo "To transfer to China machine and load:"
+echo "  scp images.tar user@china-machine:/path/"
+echo "  docker load < images.tar"

scripts/shell_scripts/install.sh

@@ -0,0 +1,167 @@
+#!/bin/bash
+# Installation script for LangChain Agent
+# This script sets up and runs the entire application stack
+
+set -e
+
+# Colors for output
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+NC='\033[0m' # No Color
+
+# Configuration
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(dirname "$(dirname "$SCRIPT_DIR")")"
+ENV_FILE="$PROJECT_ROOT/.env"
+
+echo -e "${GREEN}=== LangChain Agent Installation Script ===${NC}\n"
+
+# Check for required tools
+check_requirements() {
+    echo -e "${YELLOW}Checking requirements...${NC}"
+    
+    if ! command -v docker &> /dev/null; then
+        echo -e "${RED}Error: Docker is not installed. Please install Docker first.${NC}"
+        exit 1
+    fi
+    
+    if ! command -v docker-compose &> /dev/null && ! docker compose version &> /dev/null; then
+        echo -e "${RED}Error: docker-compose is not installed. Please install docker-compose first.${NC}"
+        exit 1
+    fi
+    
+    echo -e "${GREEN}✓ All requirements met${NC}\n"
+}
+
+# Create .env file if it doesn't exist
+create_env_file() {
+    if [ ! -f "$ENV_FILE" ]; then
+        echo -e "${YELLOW}Creating .env file...${NC}"
+        cat > "$ENV_FILE" <<EOF
+# Database Configuration
+POSTGRES_DB=ai_conversations
+POSTGRES_USER=myapp_user
+POSTGRES_PASSWORD=secure_password_123
+POSTGRES_PORT=5432
+
+# Backend Configuration
+BACKEND_PORT=8500
+
+# Frontend Configuration
+FRONTEND_PORT=8080
+
+# Database Connection String (used by backend)
+CONN_STR=postgresql://myapp_user:secure_password_123@postgres:5432/ai_conversations
+EOF
+        echo -e "${GREEN}✓ Created .env file at $ENV_FILE${NC}"
+        echo -e "${YELLOW}  Please review and update the .env file with your preferred settings.${NC}\n"
+    else
+        echo -e "${GREEN}✓ .env file already exists${NC}\n"
+    fi
+}
+
+# Build Docker images
+build_images() {
+    echo -e "${YELLOW}Building Docker images (including frontend)...${NC}"
+    cd "$PROJECT_ROOT"
+    
+    # Check if docker-compose or docker compose
+    if docker compose version &> /dev/null; then
+        COMPOSE_CMD="docker compose"
+    else
+        COMPOSE_CMD="docker-compose"
+    fi
+    
+    $COMPOSE_CMD -f docker/docker-compose.prod.yml build
+    
+    echo -e "${GREEN}✓ Docker images built successfully${NC}\n"
+}
+
+# Initialize database
+init_database() {
+    echo -e "${YELLOW}Initializing database...${NC}"
+    
+    # Wait for PostgreSQL to be ready
+    echo "Waiting for PostgreSQL to start..."
+    sleep 5
+    
+    # The SQL files in scripts/init_database/ will be automatically executed
+    # by PostgreSQL's docker-entrypoint-initdb.d mechanism
+    # We just need to wait a bit for it to complete
+    
+    echo -e "${GREEN}✓ Database initialization will be handled automatically by PostgreSQL container${NC}\n"
+}
+
+# Start services
+start_services() {
+    echo -e "${YELLOW}Starting services...${NC}"
+    cd "$PROJECT_ROOT"
+    
+    # Check if docker-compose or docker compose
+    if docker compose version &> /dev/null; then
+        COMPOSE_CMD="docker compose"
+    else
+        COMPOSE_CMD="docker-compose"
+    fi
+    
+    $COMPOSE_CMD -f docker/docker-compose.prod.yml up -d
+    
+    echo -e "${GREEN}✓ Services started${NC}\n"
+}
+
+# Show status
+show_status() {
+    # Load environment variables from .env if it exists
+    if [ -f "$ENV_FILE" ]; then
+        set -a
+        source "$ENV_FILE"
+        set +a
+    fi
+    
+    echo -e "${GREEN}=== Installation Complete ===${NC}\n"
+    echo -e "Services are starting up. Please wait a moment for them to be ready.\n"
+    echo -e "Access points:"
+    echo -e "  - Frontend:     http://localhost:${FRONTEND_PORT:-80}"
+    echo -e "  - Backend API:  http://localhost:${BACKEND_PORT:-8500}"
+    echo -e "  - Database:     localhost:${POSTGRES_PORT:-5432}\n"
+    echo -e "To view logs:"
+    echo -e "  docker-compose -f docker/docker-compose.prod.yml logs -f\n"
+    echo -e "To stop services:"
+    echo -e "  docker-compose -f docker/docker-compose.prod.yml down\n"
+    echo -e "To restart services:"
+    echo -e "  docker-compose -f docker/docker-compose.prod.yml restart\n"
+}
+
+# Main execution
+main() {
+    check_requirements
+    create_env_file
+    build_images
+    start_services
+    init_database
+    show_status
+    
+    echo -e "${YELLOW}Waiting for services to be healthy...${NC}"
+    sleep 10
+    
+    # Load environment variables for health check
+    if [ -f "$ENV_FILE" ]; then
+        set -a
+        source "$ENV_FILE"
+        set +a
+    fi
+    
+    # Check service health
+    echo -e "\n${YELLOW}Checking service health...${NC}"
+    sleep 5  # Give services a bit more time
+    if curl -f http://localhost:${BACKEND_PORT:-8500}/health &> /dev/null; then
+        echo -e "${GREEN}✓ Backend is healthy${NC}"
+    else
+        echo -e "${YELLOW}⚠ Backend is still starting up. Check logs with: docker-compose -f docker/docker-compose.prod.yml logs backend${NC}"
+    fi
+}
+
+# Run main function
+main
+

tests/test_combined_app.py

@@ -0,0 +1,46 @@
+import importlib
+import os
+import sys
+
+from fastapi.testclient import TestClient
+
+os.environ.setdefault("CONN_STR", "postgresql://dummy:dummy@localhost/dummy")
+
+
+def test_server_dashscope_import_is_cli_safe(monkeypatch):
+    """
+    Importing server_dashscope should not invoke tyro.cli at module import time.
+    """
+    import tyro
+
+    monkeypatch.setattr(
+        tyro,
+        "cli",
+        lambda *_args, **_kwargs: (_ for _ in ()).throw(
+            AssertionError("tyro.cli must not run during module import")
+        ),
+    )
+    sys.modules.pop("fastapi_server.server_dashscope", None)
+
+    module = importlib.import_module("fastapi_server.server_dashscope")
+    assert module.app is not None
+    assert module.dashscope_router is not None
+
+
+def test_combined_app_serves_front_and_dashscope_routes():
+    from fastapi_server.combined import app
+
+    client = TestClient(app)
+
+    # front_apis route should be available.
+    front_resp = client.get("/v1/pipelines/graphs")
+    assert front_resp.status_code == 200, front_resp.text
+    assert "available_graphs" in front_resp.json()
+
+    # DashScope route should exist at the same path (missing auth should not be 404).
+    dash_resp = client.post(
+        "/api/v1/apps/blueberry/sessions/test-session/responses",
+        json={"input": {"prompt": "hello"}, "stream": False},
+    )
+    assert dash_resp.status_code != 404, dash_resp.text
+

tests/test_dashscope_client.py

@@ -30,7 +30,7 @@ except Exception as e:
 
 
 # <<< Paste your running FastAPI base url here >>>
-BASE_URL = os.getenv("DS_BASE_URL", "http://127.0.0.1:8588/api/")
+BASE_URL = os.getenv("DS_BASE_URL", "http://127.0.0.1:8500/api/")
 
 
 # Params

tests/test_front_apis_registry.py

@@ -1,13 +1,18 @@
 import json
 import os
 from pathlib import Path
+from datetime import datetime, timedelta, timezone
+import importlib
 
 from fastapi.testclient import TestClient
 
 
 os.environ.setdefault("CONN_STR", "postgresql://dummy:dummy@localhost/dummy")
 
-import fastapi_server.front_apis as front_apis
+try:
+    front_apis = importlib.import_module("lang_agent.fastapi_server.front_apis")
+except ModuleNotFoundError:
+    front_apis = importlib.import_module("fastapi_server.front_apis")
 
 
 def _fake_build_fn(
@@ -20,7 +25,7 @@ def _fake_build_fn(
 ):
     out_dir = Path(pipeline_config_dir)
     out_dir.mkdir(parents=True, exist_ok=True)
-    out_file = out_dir / f"{pipeline_id}.yml"
+    out_file = out_dir / f"{pipeline_id}.yaml"
     out_file.write_text(
         json.dumps(
             {
@@ -36,9 +41,101 @@ def _fake_build_fn(
     return {"path": str(out_file)}
 
 
+class _FakeCursor:
+    def __init__(self, rows):
+        self._rows = rows
+        self._result = []
+        self._last_sql = ""
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc, tb):
+        return False
+
+    def execute(self, sql, params=None):
+        self._last_sql = sql
+        query = " ".join(sql.split()).lower()
+        params = params or ()
+
+        if "group by conversation_id, pipeline_id" in query:
+            pipeline_id = params[0]
+            limit = int(params[1])
+            grouped = {}
+            for row in self._rows:
+                if row["pipeline_id"] != pipeline_id:
+                    continue
+                conv_id = row["conversation_id"]
+                if conv_id not in grouped:
+                    grouped[conv_id] = {
+                        "conversation_id": conv_id,
+                        "pipeline_id": row["pipeline_id"],
+                        "message_count": 0,
+                        "last_updated": row["created_at"],
+                    }
+                grouped[conv_id]["message_count"] += 1
+                if row["created_at"] > grouped[conv_id]["last_updated"]:
+                    grouped[conv_id]["last_updated"] = row["created_at"]
+            values = sorted(grouped.values(), key=lambda x: x["last_updated"], reverse=True)
+            self._result = values[:limit]
+            return
+
+        if "select 1 from messages" in query:
+            pipeline_id, conversation_id = params
+            found = any(
+                row["pipeline_id"] == pipeline_id
+                and row["conversation_id"] == conversation_id
+                for row in self._rows
+            )
+            self._result = [{"exists": 1}] if found else []
+            return
+
+        if "order by sequence_number asc" in query:
+            pipeline_id, conversation_id = params
+            self._result = sorted(
+                [
+                    {
+                        "message_type": row["message_type"],
+                        "content": row["content"],
+                        "sequence_number": row["sequence_number"],
+                        "created_at": row["created_at"],
+                    }
+                    for row in self._rows
+                    if row["pipeline_id"] == pipeline_id
+                    and row["conversation_id"] == conversation_id
+                ],
+                key=lambda x: x["sequence_number"],
+            )
+            return
+
+        raise AssertionError(f"Unsupported SQL in test fake: {self._last_sql}")
+
+    def fetchall(self):
+        return self._result
+
+    def fetchone(self):
+        if not self._result:
+            return None
+        return self._result[0]
+
+
+class _FakeConnection:
+    def __init__(self, rows):
+        self._rows = rows
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc, tb):
+        return False
+
+    def cursor(self, row_factory=None):
+        return _FakeCursor(self._rows)
+
+
 def test_registry_route_lifecycle(monkeypatch, tmp_path):
     registry_path = tmp_path / "pipeline_registry.json"
-    monkeypatch.setattr(front_apis, "_PIPELINE_REGISTRY_PATH", str(registry_path))
+    monkeypatch.setattr(front_apis, "PIPELINE_REGISTRY_PATH", str(registry_path))
     monkeypatch.setitem(front_apis.GRAPH_BUILD_FNCS, "routing", _fake_build_fn)
 
     client = TestClient(front_apis.app)
@@ -60,7 +157,7 @@ def test_registry_route_lifecycle(monkeypatch, tmp_path):
     assert create_data["pipeline_id"] == "xiaozhan"
     assert create_data["graph_id"] == "routing"
     assert create_data["llm_name"] == "qwen-plus"
-    assert create_data["reload_required"] is True
+    assert create_data["reload_required"] is False
 
     list_resp = client.get("/v1/pipelines")
     assert list_resp.status_code == 200, list_resp.text
@@ -91,7 +188,7 @@ def test_registry_route_lifecycle(monkeypatch, tmp_path):
 
 def test_registry_api_key_policy_lifecycle(monkeypatch, tmp_path):
     registry_path = tmp_path / "pipeline_registry.json"
-    monkeypatch.setattr(front_apis, "_PIPELINE_REGISTRY_PATH", str(registry_path))
+    monkeypatch.setattr(front_apis, "PIPELINE_REGISTRY_PATH", str(registry_path))
     monkeypatch.setitem(front_apis.GRAPH_BUILD_FNCS, "routing", _fake_build_fn)
 
     client = TestClient(front_apis.app)
@@ -136,4 +233,119 @@ def test_registry_api_key_policy_lifecycle(monkeypatch, tmp_path):
     delete_data = delete_resp.json()
     assert delete_data["api_key"] == "sk-test-key"
     assert delete_data["status"] == "deleted"
-    assert delete_data["reload_required"] is True
+    assert delete_data["reload_required"] is False
+
+
+def test_pipeline_conversation_routes(monkeypatch):
+    now = datetime.now(timezone.utc)
+    rows = [
+        {
+            "conversation_id": "agent-a:conv-1",
+            "pipeline_id": "agent-a",
+            "message_type": "human",
+            "content": "hello",
+            "sequence_number": 1,
+            "created_at": now - timedelta(seconds=30),
+        },
+        {
+            "conversation_id": "agent-a:conv-1",
+            "pipeline_id": "agent-a",
+            "message_type": "ai",
+            "content": "hi there",
+            "sequence_number": 2,
+            "created_at": now - timedelta(seconds=20),
+        },
+        {
+            "conversation_id": "agent-a:conv-2",
+            "pipeline_id": "agent-a",
+            "message_type": "human",
+            "content": "second thread",
+            "sequence_number": 1,
+            "created_at": now - timedelta(seconds=10),
+        },
+        {
+            "conversation_id": "agent-b:conv-9",
+            "pipeline_id": "agent-b",
+            "message_type": "human",
+            "content": "other pipeline",
+            "sequence_number": 1,
+            "created_at": now - timedelta(seconds=5),
+        },
+    ]
+
+    monkeypatch.setenv("CONN_STR", "postgresql://dummy:dummy@localhost/dummy")
+    monkeypatch.setattr(
+        front_apis.psycopg,
+        "connect",
+        lambda _conn_str: _FakeConnection(rows),
+    )
+
+    client = TestClient(front_apis.app)
+
+    list_resp = client.get("/v1/pipelines/agent-a/conversations")
+    assert list_resp.status_code == 200, list_resp.text
+    list_data = list_resp.json()
+    assert list_data["pipeline_id"] == "agent-a"
+    assert list_data["count"] == 2
+    assert [item["conversation_id"] for item in list_data["items"]] == [
+        "agent-a:conv-2",
+        "agent-a:conv-1",
+    ]
+    assert all(item["pipeline_id"] == "agent-a" for item in list_data["items"])
+
+    msg_resp = client.get("/v1/pipelines/agent-a/conversations/agent-a:conv-1/messages")
+    assert msg_resp.status_code == 200, msg_resp.text
+    msg_data = msg_resp.json()
+    assert msg_data["pipeline_id"] == "agent-a"
+    assert msg_data["conversation_id"] == "agent-a:conv-1"
+    assert msg_data["count"] == 2
+    assert [item["message_type"] for item in msg_data["items"]] == ["human", "ai"]
+    assert [item["sequence_number"] for item in msg_data["items"]] == [1, 2]
+
+
+def test_pipeline_conversation_messages_404(monkeypatch):
+    rows = [
+        {
+            "conversation_id": "agent-b:conv-9",
+            "pipeline_id": "agent-b",
+            "message_type": "human",
+            "content": "other pipeline",
+            "sequence_number": 1,
+            "created_at": datetime.now(timezone.utc),
+        },
+    ]
+    monkeypatch.setenv("CONN_STR", "postgresql://dummy:dummy@localhost/dummy")
+    monkeypatch.setattr(
+        front_apis.psycopg,
+        "connect",
+        lambda _conn_str: _FakeConnection(rows),
+    )
+
+    client = TestClient(front_apis.app)
+    resp = client.get("/v1/pipelines/agent-a/conversations/agent-b:conv-9/messages")
+    assert resp.status_code == 404, resp.text
+    assert "not found for pipeline 'agent-a'" in resp.json()["detail"]
+
+
+def test_runtime_auth_info_prefers_registry_then_env(monkeypatch, tmp_path):
+    registry_path = tmp_path / "pipeline_registry.json"
+    registry_path.write_text(
+        json.dumps(
+            {
+                "pipelines": {},
+                "api_keys": {
+                    "sk-from-registry": {"default_pipeline_id": "blueberry"},
+                },
+            }
+        ),
+        encoding="utf-8",
+    )
+    monkeypatch.setattr(front_apis, "PIPELINE_REGISTRY_PATH", str(registry_path))
+    monkeypatch.setenv("FAST_AUTH_KEYS", "sk-from-env,other")
+
+    client = TestClient(front_apis.app)
+    resp = client.get("/v1/runtime-auth")
+    assert resp.status_code == 200, resp.text
+    data = resp.json()
+    assert data["fast_api_key"] == "sk-from-registry"
+    assert data["source"] == "pipeline_registry"

tests/test_migrate_yaml_prompts_to_db.py

@@ -0,0 +1,113 @@
+import importlib.util
+import sys
+from pathlib import Path
+from types import SimpleNamespace
+
+
+def _load_module():
+    project_root = Path(__file__).resolve().parents[1]
+    script_path = project_root / "scripts" / "py_scripts" / "migrate_yaml_prompts_to_db.py"
+    spec = importlib.util.spec_from_file_location("migrate_yaml_prompts_to_db", script_path)
+    module = importlib.util.module_from_spec(spec)
+    assert spec.loader is not None
+    sys.modules[spec.name] = module
+    spec.loader.exec_module(module)
+    return module
+
+
+def test_infer_pipeline_id_falls_back_to_filename():
+    module = _load_module()
+    conf = SimpleNamespace(
+        pipeline_id=None,
+        graph_config=SimpleNamespace(pipeline_id=None),
+    )
+    out = module._infer_pipeline_id(conf, "/tmp/blueberry.yaml")
+    assert out == "blueberry"
+
+
+def test_extract_prompt_dict_for_react_txt(tmp_path):
+    module = _load_module()
+    prompt_f = tmp_path / "sys.txt"
+    prompt_f.write_text("hello react", encoding="utf-8")
+    graph_conf = SimpleNamespace(sys_prompt_f=str(prompt_f))
+    prompt_dict = module._extract_prompt_dict(graph_conf)
+    assert prompt_dict == {"sys_prompt": "hello react"}
+
+
+def test_extract_prompt_dict_for_routing_dir(tmp_path):
+    module = _load_module()
+    (tmp_path / "route_prompt.txt").write_text("route", encoding="utf-8")
+    (tmp_path / "chat_prompt.txt").write_text("chat", encoding="utf-8")
+    graph_conf = SimpleNamespace(sys_promp_dir=str(tmp_path))
+    prompt_dict = module._extract_prompt_dict(graph_conf)
+    assert prompt_dict["route_prompt"] == "route"
+    assert prompt_dict["chat_prompt"] == "chat"
+
+
+def test_collect_payload_routing_ignores_chatty_prompt_for_tool_node(tmp_path):
+    module = _load_module()
+    prompt_dir = tmp_path / "prompts"
+    prompt_dir.mkdir()
+    (prompt_dir / "route_prompt.txt").write_text("route", encoding="utf-8")
+    (prompt_dir / "chat_prompt.txt").write_text("chat", encoding="utf-8")
+    (prompt_dir / "tool_prompt.txt").write_text("tool", encoding="utf-8")
+    (prompt_dir / "chatty_prompt.txt").write_text("chatty", encoding="utf-8")
+
+    class RoutingConfig:
+        pass
+
+    class ToolNodeConfig:
+        pass
+
+    graph_conf = RoutingConfig()
+    graph_conf.sys_promp_dir = str(prompt_dir)
+    graph_conf.tool_node_config = ToolNodeConfig()
+    graph_conf.tool_node_config.tool_prompt_f = str(prompt_dir / "tool_prompt.txt")
+
+    conf = SimpleNamespace(
+        pipeline_id=None,
+        api_key="sk",
+        graph_config=graph_conf,
+    )
+
+    module.load_tyro_conf = lambda _: conf
+    payload = module._collect_payload(str(tmp_path / "xiaozhan.yaml"))
+    assert payload.pipeline_id == "xiaozhan"
+    assert set(payload.prompt_dict.keys()) == {"route_prompt", "chat_prompt", "tool_prompt"}
+    assert "chatty_prompt" not in payload.prompt_dict
+
+
+def test_collect_payload_routing_includes_chatty_prompt_for_chatty_node(tmp_path):
+    module = _load_module()
+    prompt_dir = tmp_path / "prompts"
+    prompt_dir.mkdir()
+    (prompt_dir / "route_prompt.txt").write_text("route", encoding="utf-8")
+    (prompt_dir / "chat_prompt.txt").write_text("chat", encoding="utf-8")
+    (prompt_dir / "tool_prompt.txt").write_text("tool", encoding="utf-8")
+    (prompt_dir / "chatty_prompt.txt").write_text("chatty", encoding="utf-8")
+
+    class RoutingConfig:
+        pass
+
+    class ChattyToolNodeConfig:
+        pass
+
+    graph_conf = RoutingConfig()
+    graph_conf.sys_promp_dir = str(prompt_dir)
+    graph_conf.tool_node_config = ChattyToolNodeConfig()
+    graph_conf.tool_node_config.tool_prompt_f = str(prompt_dir / "tool_prompt.txt")
+    graph_conf.tool_node_config.chatty_sys_prompt_f = str(
+        prompt_dir / "chatty_prompt.txt"
+    )
+
+    conf = SimpleNamespace(
+        pipeline_id="xiaozhan",
+        api_key="sk",
+        graph_config=graph_conf,
+    )
+
+    module.load_tyro_conf = lambda _: conf
+    payload = module._collect_payload(str(tmp_path / "xiaozhan.yaml"))
+    assert payload.pipeline_id == "xiaozhan"
+    assert "chatty_prompt" in payload.prompt_dict
+

tests/test_server_pipeline_manager_refresh.py

@@ -0,0 +1,156 @@
+import json
+import time
+
+import pytest
+from fastapi import HTTPException
+
+from lang_agent.components.server_pipeline_manager import ServerPipelineManager
+
+
+class _DummyPipeline:
+    def __init__(self, model: str):
+        self.model = model
+
+
+class _DummyConfig:
+    def __init__(self, llm_name: str = "qwen-plus"):
+        self.llm_name = llm_name
+
+    def setup(self):
+        return _DummyPipeline(model=self.llm_name)
+
+
+def _write_registry(path, pipelines, api_keys=None):
+    content = {"pipelines": pipelines, "api_keys": api_keys or {}}
+    path.write_text(json.dumps(content, indent=2), encoding="utf-8")
+    # Ensure mtime changes reliably on fast CI filesystems.
+    time.sleep(0.01)
+
+
+def test_refresh_registry_picks_up_new_pipeline(tmp_path):
+    registry_path = tmp_path / "pipeline_registry.json"
+    _write_registry(
+        registry_path,
+        pipelines={
+            "default": {
+                "enabled": True,
+                "config_file": None,
+                "llm_name": "qwen-plus",
+            }
+        },
+    )
+    manager = ServerPipelineManager(
+        default_pipeline_id="default",
+        default_config=_DummyConfig(),
+    )
+    manager.load_registry(str(registry_path))
+
+    with pytest.raises(HTTPException) as exc_info:
+        manager.resolve_pipeline_id(
+            body={"pipeline_id": "blueberry"}, app_id=None, api_key="k1"
+        )
+    assert exc_info.value.status_code == 404
+
+    _write_registry(
+        registry_path,
+        pipelines={
+            "default": {
+                "enabled": True,
+                "config_file": None,
+                "llm_name": "qwen-plus",
+            },
+            "blueberry": {
+                "enabled": True,
+                "config_file": None,
+                "llm_name": "qwen-max",
+            },
+        },
+    )
+    changed = manager.refresh_registry_if_needed()
+    assert changed is True
+
+    resolved = manager.resolve_pipeline_id(
+        body={"pipeline_id": "blueberry"}, app_id=None, api_key="k1"
+    )
+    assert resolved == "blueberry"
+
+
+def test_refresh_registry_invalidates_cache_for_changed_pipeline(tmp_path):
+    registry_path = tmp_path / "pipeline_registry.json"
+    _write_registry(
+        registry_path,
+        pipelines={
+            "blueberry": {
+                "enabled": True,
+                "config_file": None,
+                "llm_name": "qwen-plus",
+            }
+        },
+    )
+    manager = ServerPipelineManager(
+        default_pipeline_id="blueberry",
+        default_config=_DummyConfig(),
+    )
+    manager.load_registry(str(registry_path))
+
+    first_pipeline, first_model = manager.get_pipeline("blueberry")
+    assert first_model == "qwen-plus"
+
+    _write_registry(
+        registry_path,
+        pipelines={
+            "blueberry": {
+                "enabled": True,
+                "config_file": None,
+                "llm_name": "qwen-max",
+            }
+        },
+    )
+    changed = manager.refresh_registry_if_needed()
+    assert changed is True
+
+    second_pipeline, second_model = manager.get_pipeline("blueberry")
+    assert second_model == "qwen-max"
+    assert second_pipeline is not first_pipeline
+
+
+def test_refresh_registry_applies_disabled_state_immediately(tmp_path):
+    registry_path = tmp_path / "pipeline_registry.json"
+    _write_registry(
+        registry_path,
+        pipelines={
+            "blueberry": {
+                "enabled": True,
+                "config_file": None,
+                "llm_name": "qwen-plus",
+            }
+        },
+    )
+    manager = ServerPipelineManager(
+        default_pipeline_id="blueberry",
+        default_config=_DummyConfig(),
+    )
+    manager.load_registry(str(registry_path))
+    manager.get_pipeline("blueberry")
+
+    _write_registry(
+        registry_path,
+        pipelines={
+            "blueberry": {
+                "enabled": False,
+                "config_file": None,
+                "llm_name": "qwen-plus",
+            }
+        },
+    )
+    changed = manager.refresh_registry_if_needed()
+    assert changed is True
+
+    with pytest.raises(HTTPException) as exc_info:
+        manager.get_pipeline("blueberry")
+    assert exc_info.value.status_code == 403
+
+
+
+
+