Quickstart: see Noeta running in 5 minutes

What you'll do: install Noeta, boot the offline coding agent, open the web UI, send a message, and inspect the trace. No API key needed — the default stub provider is a deterministic LLM double.

1. Install

pip install noeta-agent

This pulls in the SDK and runtime. You need Python 3.11+.

2. Boot the agent

python -m noeta.agent

This starts the coding agent server with the offline stub provider and in-memory storage. It prints a URL — something like http://127.0.0.1:54321/. The server blocks until Ctrl-C; there is no daemon mode.

3. Open the web UI

Point your browser at the printed URL, then navigate to /chat. You should see the chat composer. Type a message — for example:

List the Python files in this directory and tell me what each one does.

The stub provider returns a scripted two-turn response, so you will see the agent "think" and then reply with a canned answer. The point is not the quality of the response — it is that every step is recorded.

4. View the trace

From the chat view, click on a session to open its trace. The trace view shows every event in the session's EventLog: the user message, the LLM turn, any tool calls, the tool results, and the final answer. Each row includes token counts and cache statistics.

This trace is not generated from process memory — it is folded from the EventLog, the same way the agent's own state is recovered. See Event sourcing to understand why this matters.

5. Drive it in-process (optional)

If you prefer code over a browser, the same backend boots in a few lines:

from noeta.agent.backend.lifecycle import BackendConfig, serve_backend

# Defaults are fully offline: the stub provider, :memory: storage.
# port=0 binds an OS-assigned port. Workspace is the current directory.
config = BackendConfig(port=0)
server, url, shutdown = serve_backend(config)
try:
    assert url.startswith("http://")
    print(f"Backend running at {url}")
finally:
    shutdown()

Next steps

• Connect a real model — Configure a provider walks through Anthropic and OpenAI-compatible setups.
• Build your own agent — Your first agent is a 20-minute guided SDK walkthrough using @tool, Options, and query().
• Understand the design — start with Event sourcing and the architecture overview.
• Use the coding agent for real work — Use the coding agent covers workspace setup, presets, and skills.