LightOn x Idun RAG Agent

A document assistant connecting LightOn Console (knowledge layer) to Idun Agent Platform (agent runtime), via a LangGraph agent powered by Gemini.

Full write-up: docs/blog/lighton-x-idun-rag-agent.md.

Structure

sdk/
  client.py       Console v3 API client
  models.py       Pydantic response models
  exceptions.py   Custom exceptions
  tests/          Unit and integration tests
mcp_server.py     MCP server exposing the SDK as tools
agent.py          LangGraph agent (planner + executor)

Setup

Requires Python 3.12 and uv.

uv sync

Create .env:

LIGHTON_API_KEY=your_lighton_key
GOOGLE_API_KEY=your_gemini_key

Run the MCP server

uv run python mcp_server.py

Listens on http://127.0.0.1:8000/mcp. Expose with ngrok if Idun Cloud needs to reach it.

Configure the agent in Idun Cloud

At cloud.idunplatform.com:

1. Create a LangGraph agent, graph definition agent.py:graph
2. Add the MCP server (streamable HTTP, URL + /mcp)
3. Create system-prompt and plan-prompt, assign both to the agent (prompt text in the blog post)

Tests

uv run pytest sdk/tests/test_unit.py          # mocked, fast
uv run pytest sdk/tests/test_integration.py   # live API