Idun Engine — Open-source runtime for AI agents

Idun Engine is the open-source runtime that takes a LangGraph, Google ADK or LangChain agent prototype to a production-grade FastAPI service. One pip install idun-agent-engine wraps your graph with the pieces you would have rebuilt anyway: demo chat UI, admin panel, OpenTelemetry trace viewer, PII and budget guardrails, persistent memory, MCP tool governance, OIDC authentication. GPL 3.0 licence, full source on github.com/Idun-Group/idun-agent-platform.

For whom

CTOs, Heads of Platform, Principal Engineers evaluating alternatives to LangGraph Platform (formerly LangGraph Cloud, formerly LangSmith Deployment) who reject per-node billing and vendor lock-in. Self-hosted: your infra, your code, your data.

Quick comparison

vs LangGraph Platform: self-hosted, GPL 3.0, no per-execution billing, custom OIDC included from first install. vs CrewAI: runtime-agnostic, first-class LangGraph and Google ADK support, LangChain via the LangGraph adapter. vs "rolling your own FastAPI": 3 weeks of effort saved (AG-UI streaming, OTel, multi-LLM exporters, guardrails, memory, MCP, OIDC, admin, traces, chat UI — all bundled).

Technical stack

Python 3.12+, FastAPI, OpenTelemetry, multi-LLM support (Anthropic Claude, OpenAI, Google Gemini, Mistral, Cohere, self-hosted models via vLLM), Postgres or Redis persistence for memory, native integration with Langfuse / Phoenix / Arize for observability.

Install

pip install idun-agent-engine then idun-standalone init my-agent. Three files: your LangGraph/ADK graph, an agent.yaml config, a .env. One command to serve: idun-standalone run. The agent runs locally on port 8000 with chat UI, admin, traces and health endpoint.

Backing

Maintained by Idun Group, Paris. Backed by Wilco, IMT Starter, Polytechnique X-UP, La French Tech.