Idun Agent Platform

Productionize and govern generative-AI agents on your own infrastructure.

Idun Agent Platform is an open source control plane for generative AI agents. It turns LangGraph, ADK or Haystack agents into production-ready services with

• Unified deployment across frameworks
• Observability and tracing
• Memory and session persistence
• Guardrails and MCP integration
• Multi-environment and access control

Who is this for

• GenAI developers, who want to ship agents without rebuilding infra each time
• AI and data platform teams, who need governance, auditability and sovereignty

pip install idun-agent-engine

Links

• Quickstart (platform in Docker Compose)
• Documentation
• Discord
• Contributing and Code of Conduct

If you find this project useful, please star the repository and join our Discord community.

Built by Idun Group (France).

Table of contents

• Should you use Idun Agent Platform
• Quickstart (Engine only, 10 minutes)
• Quickstart (Platform, Docker Compose)
• Key capabilities at a glance
• High level architecture
• Enterprise and regulated environments
• Community and support
• Commercial support
• Project status and roadmap
• Contributing

Should you use Idun Agent Platform

You probably should if:

• You have or plan multiple agents built with LangGraph, ADK, Haystack or similar
• You care about observability, guardrails, security, and AI regulation
• You want to self host or run on your own cloud, not depend on a vendor black box

You probably should not if:

• You are just experimenting with a single toy chatbot
• You do not need observability, governance or multi environment setups yet

For GenAI developers

You want to spend time on agent logic, not boilerplate infra.

With Idun you can:

• Wrap your LangGraph, ADK or Haystack agent as a FastAPI service in minutes
• Get tracing, feedback and metrics without rewriting your code
• Run the same agent locally, on staging and in production with the same config
• Plug tools and memory through configuration instead of hard coding everything

For AI and data platform teams

You want to standardize how agents run in production and stay compliant.

With Idun you can:

• Maintain a catalog of approved agents with clear ownership and environments
• Enforce SSO, RBAC and per tenant isolation, integrated with your IdP
• Control which models, tools and data sources each agent can use with MCP
• Enforce guardrails for safety and compliance, with full audit and monitoring

Why Idun exists

Today, each agent framework comes with its own way to deploy, observe and govern agents. The result is a zoo of one off POCs, custom servers and ad hoc dashboards.

Idun Agent Platform gives you:

• One configuration model, define agents configurations in one central hub that works across frameworks
• Production features by default, memory, observability, guardrails, MCP, SSO access
• Flexible deployment, run locally, self host on your own cloud or integrate in your platform
• Centralized control, manage agents, environments and access from one dashboard or CLI

👉 Read the technical whitepaper for a deeper architecture overview.

Key capabilities at a glance

• Observability Plug Langfuse, Phoenix, LangSmith or GCP, get tracing and metrics for every call.

• Guardrails Add content safety, PII detection and prompt injection protection in front of any agent.

• MCP integration Extend agents with Model Context Protocol servers, Idun manages server lifecycle and tool registration.

• Memory and session persistence Persist conversations and state across calls with backends like SQLite or Postgres.

Quickstart (Engine only, 10 minutes)

If you just want to run an agent API (without the full platform UI/Manager), you can run the Idun Agent Engine standalone.

1. Install:

pip install idun-agent-engine

2. Create a minimal LangGraph agent (example_agent.py):

import operator
from typing import Annotated, TypedDict

from langgraph.graph import END, StateGraph


class AgentState(TypedDict):
    messages: Annotated[list, operator.add]


def greet_node(state: AgentState):
    user_message = state["messages"][-1] if state.get("messages") else ""
    return {"messages": [("ai", f"Hello! You said: '{user_message}'")]}


graph = StateGraph(AgentState)
graph.add_node("greet", greet_node)
graph.set_entry_point("greet")
graph.add_edge("greet", END)

app = graph

3. Point the engine to it (config.yaml) and run:

server:
  api:
    port: 8000

agent:
  type: "langgraph"
  config:
    name: "Hello World Agent"
    graph_definition: "./example_agent.py:app"

python -c "from idun_agent_engine.core.server_runner import run_server_from_config; run_server_from_config('config.yaml')"

Then open http://localhost:8000/docs.

High level architecture

Idun Agent Platform is structured in four layers:

• Web dashboard UI to create, configure and monitor agents.

• Manager API Control plane that stores configurations, handles auth, observability and guardrails settings.

• Engine runtime Executes agents via adapters for LangGraph, ADK, Haystack and others, exposes AG-UI FastAPI endpoints.

• Data layer PostgreSQL for checkpointing and configuration, MCP servers for external tools and data.

Quickstart

You need Python 3.12, Docker and Git.

1. Clone the repo

git clone https://github.com/Idun-Group/idun-agent-platform.git
cd idun-agent-platform

2. Start the platform locally

cp .env.example .env

docker compose -f docker-compose.dev.yml up --build

3. Open the dashboard at http://localhost:3000 and create your first agent.

👉 For a complete step by step tutorial, including ADK example code, see the Quickstart guide.

Enterprise and regulated environments

Idun is built for teams that need production governance and data control:

• Data sovereignty: run on-prem or in your own cloud (including EU-only deployments).
• Security: SSO/OIDC, multi-tenant RBAC, API keys, and guardrails.
• Auditability: end-to-end observability across prompts, tools, and model calls.
• Platform standardization: consistent runtime and policy enforcement across frameworks.

If you’re evaluating Idun for enterprise use, contact us at [email protected].

Technical whitepaper

1) System Overview

Objective: Provide a unified, production-grade runtime for heterogeneous agent frameworks with centralized configuration, security, and observability. Components:

• Idun Agent Engine — wraps LangGraph/ADK/Haystack agents into a FastAPI service with unified API using AG-UI protocol, memory, guardrails, and tracing. Works standalone (YAML) or managed (fetches config from Manager).
• Idun Agent Manager — FastAPI + PostgreSQL service for CRUD on engine configs; serves signed configs to Engines; enforces SSO/RBAC and tenancy.
• Idun Agent UI — Next.js admin for creating, deploying, observing, and governing agents.
• Idun Agent Schema — shared Pydantic models (EngineConfig, ManagedAgent, API contracts) ensuring type-safe interoperability across services.

Non-functional pillars: Sovereignty (EU/on-prem), Security (guardrails, SSO/RBAC, air-gap), Observability (Langfuse/Phoenix/OpenTelemetry), Scalability (rate limiting, LB), Interoperability (multi-LLM, multi-framework).

2) Reference Architecture

flowchart LR
  subgraph Client_Apps["Client / Apps"]
    UI["Business Apps / Chat UIs"]
    Dev["Dev Tools / CI/CD"]
  end

  subgraph Idun_Platform["Idun Platform"]
    MGR["Agent Manager (Configs, SSO, RBAC, Templates)"]
    ENG["Agent Engines (FastAPI)"]
    OBS["Observability (Langfuse • Phoenix • OTel)"]
    VDB[(Vector DB / Memory)]
    CFGDB[(PostgreSQL Config DB)]
  end

  subgraph Models_Tools["Models & Tools"]
    LLMlocal["LLM Local / vLLM"]
    LLMext["LLM External"]
    TOOLS["Tools: MCP, APIs, DBs, SaaS"]
  end

  UI --> ENG
  Dev --> MGR
  MGR --> CFGDB
  MGR --> ENG
  ENG --> OBS
  ENG --> VDB
  ENG --> LLMlocal
  ENG --> LLMext
  ENG --> TOOLS

Topologies:

• On-prem / EU cloud with air-gapped LLMs and private data planes.
• Hybrid with local Engines and managed UI/Manager.

3) Managed Mode: Control-Plane Flow

sequenceDiagram
  participant Dev as Developer
  participant UI as Idun Agent UI
  participant MGR as Agent Manager
  participant ENG as Agent Engine
  participant OBS as Observability

  Dev->>UI: Create/Update Agent Template
  UI->>MGR: POST /agents (EngineConfig, RBAC, secrets refs)
  ENG->>MGR: GET /agents/{id}/config (API key/OIDC)
  MGR-->>ENG: Signed EngineConfig + secrets bindings
  ENG->>OBS: Stream traces/metrics
  Client->>ENG: /v1/chat request
  ENG-->>Client: Response + trace_id

Benefits: central policy enforcement, versioned configs, and uniform telemetry across heterogeneous frameworks.

4) Security & Compliance

• Identity & Access: SSO/OIDC, RBAC, per-tenant isolation; API keys for service-to-service.
• Data Residency: EU/on-prem deployment; optional air-gap with local LLMs.
• Guardrails: input/output filters, tool-use allowlists, prompt policy; LLM-as-a-judge evaluation paths.
• Auditability: full tracing of prompts, tools, and model calls (OTel + Langfuse/Phoenix).

5) Observability & Evaluation

• Tracing: spans across retrieval, function/tool calls, and model steps.
• Feedback loops: user scores + auto-eval ("LLM as a judge"); connect to regression dashboards.
• Budgeting: per-tenant usage/alerts; latency/error SLIs.

6) Interoperability

• Frameworks: LangGraph, CrewAI, ADK, Haystack (plug-in via adapters).
• LLMs: local (vLLM) and external (OpenAI/Anthropic/Mistral) behind a single gateway; RAG & memory via Vector DB.
• Protocols: MCP/A2A for tool and agent-to-agent collaboration.

7) Monorepo Layout

libs/
  idun_agent_engine/     # runtime + adapters + FastAPI
  idun_agent_schema/     # Pydantic models (single source of truth)
services/
  idun_agent_manager/    # control plane (FastAPI + PostgreSQL)
  idun_agent_ui/         # Next.js admin

Community and support

• Questions and help, join the Discord
• Proposals and ideas, GitHub Discussions
• Bugs and feature requests, GitHub Issues

Commercial support

Idun Agent Platform is maintained by Idun Group. We can help with:

• Design and review of your agent platform architecture
• Secure deployment on your infrastructure
• Integration with your IdP, observability stack and compliance workflows

Contact us at [email protected] for enterprise support.

Project status and roadmap

The platform is under active development and already used in production in real projects.

• ✅ Core runtime on PyPI as idun-agent-engine, with adapters for LangGraph and ADK
• ✅ Local and self-hosted deployment with Docker
• ✅ AG-UI compatible CopilotKit endpoint, MCP server support, Guardrails AI, observability (Langfuse, LangSmith, Phoenix, GCP Trace), SSO access to Manager UI
• 🚧 More agent frameworks and MCP integrations, environment management (DEV/STG/PRD), and expanded observability & evaluation
• 🚧 Deployment templates (Terraform, Helm/Kubernetes), ready-to-use agents & MCP tools, and Idun Cloud managed offering

See ROADMAP.md for details and planned features.

Contributing

Contributions are welcome. Please see CONTRIBUTING.md for guidelines.