The third path: Idun Engine v0.6

L'essentiel: Idun Engine v0.6 is the third path between LangSmith Deployment ($39/seat + per-node billing, vendor-locked) and rolling your own FastAPI wrapper (three weeks of glue work). One pip install ships a self-hosted, GPL 3.0 runtime around your LangGraph or Google ADK agent with chat UI, admin, traces, guardrails, observability, MCP, and OIDC bundled in. Production-complete on day one.

You are building a production agent. Today, your options force you to compromise.

If you adopt LangSmith Deployment (the product LangChain renamed LangGraph Cloud and then LangGraph Platform into, as of October 2025), you get a polished managed runtime. You also get $39 per seat per month, $0.001 per node execution, and per-minute uptime billing. The free self-hosted tier exists but lacks custom OIDC. Self-hosting on your own infrastructure is gated to the Enterprise plan. Your data and your scaling shape live inside someone else's product roadmap.

If you stay with raw LangGraph (or Google ADK, or LangChain) and roll your own FastAPI wrapper, you spend three weeks of engineering effort before you ship. AG-UI streaming. OpenTelemetry tracing. Multi-provider observability exporters. Guardrails. Memory persistence. MCP tool governance. OIDC auth. An admin panel. A trace viewer. A chat UI to demo to your team. Every one of those is a real engineering task. None of them is your product.

Most teams pick one path and live with the trade-off. Idun Engine is the third one.

The third path

Idun Engine is a self-hosted, open-source runtime that wraps your LangGraph or Google ADK agent into a production-ready FastAPI service. One pip install, one process, your agent live.

``bash pip install idun-agent-engine idun setup idun serve ``

Open http://localhost:8000. Chat with your agent. The admin panel is at /admin, the traces viewer at /admin/traces. Everything is bundled inside the wheel — the Next.js UI, the admin REST surface, the trace store, the migration scripts. There is nothing else to install.

This works because Idun Engine composes open standards instead of inventing new ones:

- LangGraph and Google ADK for agent orchestration - AG-UI for streaming — the protocol any CopilotKit client speaks - OpenTelemetry for tracing, exported in parallel to Langfuse, Phoenix, LangSmith, or GCP Trace - MCP for tool servers, under Linux Foundation governance since 2025 - OIDC for auth, working with Google, Microsoft, or any compliant provider - FastAPI and Pydantic for the HTTP layer and config

We wrote the glue. We did not invent another protocol. If you ever leave Idun Engine, your agent code, your traces, your MCP servers, your auth, your config — all of it moves with you.

How it compares

| | Idun Engine | LangSmith Deployment | Letta | DIY (FastAPI + glue) | |---|---|---|---|---| | Self-hosted | ✅ free | ✅ Enterprise tier only | ✅ free | ✅ | | Open source | ✅ on GitHub | ❌ closed | ✅ on GitHub | n/a | | Agent frameworks | LangGraph + Google ADK | Any (framework-agnostic since 2025) | Letta's own runtime | Any | | Streaming via AG-UI | ✅ CopilotKit-compatible | Proprietary streaming | ❌ | Build yourself | | Tracing via OpenTelemetry | ✅ multi-provider (Langfuse / Phoenix / LangSmith / GCP) | LangSmith-only | Limited | Build yourself | | MCP tool servers | ✅ per-agent governance | ❌ | Limited | Build yourself | | Bundled chat UI + admin + traces viewer | ✅ all three | ✅ their hosted UI | Partial | Build yourself | | Built-in guardrails (15+) | ✅ via Guardrails AI Hub | ❌ | ❌ | Build yourself | | Memory backends | 5 (Postgres / SQLite / in-memory / Vertex AI / ADK DB) | ✅ limited | ✅ Letta-managed | Build yourself | | OIDC / SSO | ✅ Google + Microsoft included | ✅ paid tier only | ❌ | Build yourself |