Skip to main content

Quickstart

Build, test and publish an AI agent, then call it from your own code.

1. Create an account​

Sign up with Continue with Google, or with an email and password — verify your email when the message arrives. Signing up with Google skips verification, since Google has already confirmed the address. New accounts get 100 free credits, enough for a handful of short runs.

:::tip Already have a password account? Signing in with Google using the same email address links the two, as long as that address is verified on your Google account. Afterwards either method works and it stays one account — your agents, runs and credits are untouched. :::

:::warning Free credits expire The signup grant expires 90 days after you receive it. Credits you buy never expire — only the free grant does. :::

2. Add an LLM provider key​

Agent runs use your provider keys, not ours. Before your first run, go to Dashboard → Settings → LLM Keys and add a key for at least one provider (OpenAI, Anthropic, or Gemini).

Keys are encrypted with AES-256-GCM at rest and decrypted only inside the worker that executes your run. They are never logged and never sent back to the browser after saving.

3. Create an agent​

  1. Go to Dashboard → Agents → New Agent.
  2. Pick a pattern — ReAct, chain-of-thought, self-reflection and others come pre-wired — or start from an empty canvas.
  3. Name it and click Create.

You land in the canvas editor.

4. Build the workflow​

The canvas has three regions: a node palette on the left, the graph in the middle, and a config panel on the right.

  • Drag nodes from the palette onto the canvas.
  • Connect them by dragging from one node's handle to another's.
  • Click any node to configure it in the right-hand panel.
  • Click Review in the toolbar to see what's still incomplete — each item links straight to the node that needs attention.

A minimal agent is Input → Agent → Output. See the node reference for what each type does.

Save with ⌘S or the Save changes button.

5. Test it​

Click Test in the toolbar. The runner streams live progress through every node — status, token counts, latency, and the output of each step — so you can see where a run went wrong rather than guessing from a final error.

Because you bring your own provider keys, LLM tokens are not billed in credits — you pay your provider directly. Credits are charged only for tool calls, so an agent that only calls an LLM costs 0 credits to run. See Billing for the exact rates.

6. Publish​

Open Settings in the agent toolbar and click Publish. This flips the agent from draft to published and unlocks API access.

note

Draft agents can only be tested from the dashboard. You must publish before you can create an API key or call the agent from your code.

7. Call it from your code​

On the same Settings page, create an API key for the agent. Copy it immediately — it is shown once.

curl -X POST https://your-domain.com/api/v1/agents/$AGENT_SLUG/run \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input": {"topic": "Latest AI papers"}}'

The call returns 202 Accepted immediately — execution is asynchronous:

{
  "runId": "0fb7…",
  "status": "pending",
  "eventsUrl": "/api/events/runs/0fb7…",
  "pollUrl": "/api/v1/runs/0fb7…"
}

Subscribe to eventsUrl for a live SSE stream of the run, or poll pollUrl for the final result. Full details in Run an agent via API.

Next steps​