An open-source AI Agent platform for creating, managing, running, and integrating AI agents into external systems.
+-------------------------------------------+
| X-Agent Platform |
|-------------------------------------------|
| |
| +----------+ +----------+ +----------+ |
| | Agents | | Sessions | | Chat | |
| | CRUD | | CRUD | | (SSE) | |
| +----------+ +----------+ +----------+ |
| |
| +----------+ +-----------+ |
| | Tools & | | Executor | |
| | MCP Reg | | Mgr Cache | |
| +----------+ +-----------+ |
| |
+-------------------------------------------+
- Agent CRUD — Create, list, update, delete agents with configurable models and tools
- Streaming Chat — Real-time SSE streaming with per-token deltas, tool call logs
- Tool Calling — Built-in
web_searchtool + any MCP-compatible tool servers - MCP Protocol — Register STDIO or SSE-based MCP servers, use their tools in agents
- Session Management — Persistent conversation history per agent
- Multiple Models — Per-agent model config; works with any OpenAI-compatible API
- SQLite / PostgreSQL — Zero-config SQLite for dev, PostgreSQL for production
cd agent-platform
pip install -e .
export OPENAI_API_KEY="sk-your-key"
export OPENAI_API_BASE="https://api.deepseek.com/v1"
uvicorn app.main:app --port 8001- Framework: FastAPI
- ORM: SQLAlchemy (async)
- Agent Runtime: LangGraph
create_react_agent - LLM Client: langchain-openai (OpenAI-compatible)
- Database: PostgreSQL (prod) / SQLite (dev)
- Streaming: Server-Sent Events (sse-starlette)
- Tools: Built-in registry + MCP protocol client
agent-platform/
├── app/
│ ├── api/ # REST API routes
│ │ ├── agents.py # Agent CRUD
│ │ ├── sessions.py # Session management
│ │ ├── chat.py # Streaming/non-streaming chat
│ │ └── tools.py # Tool listing & MCP registration
│ ├── db/ # Database engine & session
│ ├── models/ # SQLAlchemy ORM models
│ ├── runtime/ # Core agent engine
│ │ ├── executor.py # LangGraph ReAct agent
│ │ ├── manager.py # Executor cache
│ │ ├── tools.py # Tool registry & web_search
│ │ └── mcp_client.py# MCP protocol client
│ └── schemas/ # Pydantic request/response models
├── test_full.py # Integration test (all 10 endpoints)
├── Dockerfile
└── docker-compose.yml