Skip to main content
This tutorial assumes the full Idun stack is running via Docker Compose. If you have not set it up yet, follow the quickstart first.
docker compose -f docker-compose.dev.yml up --build
This walkthrough covers the full path from login to a working, tested agent. By the end you’ll have an agent enrolled in the platform, responding to messages through the built-in chat, and wired up for guardrails, observability, and memory.
1

Log in

Open http://localhost:3000 in your browser.Login pageEnter your email and password. If your workspace has Google OIDC configured, you can sign in with Google instead.
2

Explore the agent dashboard

After login, you land on the agent dashboard.Agent dashboardEach agent appears as a card showing its framework (LangGraph or ADK), status (Draft or Active), and resource indicators. The badges (OBS, MEM, GUARD, MCP, SSO, INT) tell you at a glance which platform features are attached.Right now the dashboard is empty. Time to change that.
3

Create an agent

Click Create an agent to open the creation wizard. It has three steps.Step 1: Basics. Enter a name for your agent and select a framework. LangGraph and ADK each appear as a radio card.Agent creation wizard, step 1Step 2: Framework config. Provide the graph definition path (for example, ./agent/graph.py:graph), select the host type (Localhost or Remote), and set the port.Agent creation wizard, step 2Step 3: Enrollment. The wizard shows instructions for connecting the agent from your code. This is the step that matters most, and the next section walks through it in detail.
4

Enroll the agent

Now you go back to your own codebase and connect the agent to the Manager.Install the engine SDK:
pip install idun-agent-engine
Set the environment variables that tell the engine where to find the Manager and how to authenticate:
export IDUN_MANAGER_HOST=http://localhost:8000
export IDUN_AGENT_API_KEY=<your-agent-api-key>
The API key is available on the agent’s API Integration tab (covered in step 6). Copy it from there.When the agent starts, it calls the Manager, pulls its materialized config, and boots with every attached resource already wired in. No local config file needed.
The agent phones home to the Manager at startup. All configuration lives in the platform, so when you change a guardrail or add an observability provider in the UI, the agent picks up the new config on restart.
5

Test via the built-in chat

Go back to the agent detail page in the browser and click the Chat tab.Built-in AG-UI chat playgroundThe playground supports multiple protocol modes (AG-UI, AG-UI Stream, Custom Stream) and lets you manage threads and inspect events. Pick a mode, type a message, send it.If the agent responds, enrollment worked. If it doesn’t, check that the engine process is running and that IDUN_MANAGER_HOST and IDUN_AGENT_API_KEY are set correctly.
6

Check API integration

Click the API Integration tab on the agent detail page.API integration tabThis tab shows the base URL, your authorization key (masked by default, with a show/hide toggle), and the available endpoints. Use these to wire the agent into your own applications or test with curl.
7

Add resources

From the agent’s Overview tab, you’ll see buttons to add guardrails, observability, memory, and MCP servers. Each links to the corresponding resource catalog page where you create a config and attach it to the agent.The workflow:
  1. Click Add guardrails (or observability, memory, etc.)
  2. Create or select a resource config on the catalog page
  3. Attach it to your agent
  4. Return to the agent detail page and click Restart in the header
The restart pushes the updated materialized config. The agent picks up the changes on its next boot.
You can attach multiple resources of the same type. For example, an agent can have both a PII detection guardrail and a topic restriction guardrail running simultaneously.

​
Next steps

Your agent is enrolled and responding to messages. Pick what to configure next.

Guardrails

Add PII detection, topic restriction, toxicity filtering, and ban lists.

Observability

Send traces to Langfuse, Phoenix, LangSmith, or GCP.

Memory

Persist conversation state across sessions with checkpoint backends.

Tool governance

Manage MCP servers and control which tools your agents can access.
Last modified on March 22, 2026