Skip to content
AlumiaDocs

Search docs

Find pages, headings, and concepts. Press ⌘K or Ctrl+K to toggle.

Build your first agent

From sign-up to a working agent reply in the lab in under five minutes.

Open in ChatGPTOpen in Claude

This walkthrough takes you from a fresh account to a chatting agent. You'll touch four surfaces: register, agents, lab, and chat.

1. Sign up

Open /auth/register and finish either email/password, Google, or GitHub. After the email confirmation step, an organization is created automatically — you do not pick a name on a separate screen. The default org slug is derived from your account; you can rename it later from Settings.

The org also ships with Ala, a built-in shared agent. Ala is org-scoped, active, and cannot be deleted. You'll see Ala in every workspace.

2. Open the Agents page

Navigate to /<orgSlug>/agents. This is the agent list for your org. Click Create agent in the top right.

Fill three fields:

  • Name — what you and other agents will call this agent.
  • System prompt — the agent's identity and behavior. Be specific. "You are a research assistant. You cite every claim." beats "Be helpful."
  • Model — pick from the catalog. The default is grok-3. Free visible providers such as Groq and NVIDIA work without a balance; paid models require funded billing or a BYOK key.

Then choose Tools. For your first agent, the defaults are fine — you can edit them at any time.

Save. The agent now appears in the list.

3. Open the lab

Go to /<orgSlug>/lab. Click New project and give it a name. The visibility toggle on the project card switches between Private (only you) and Team (everyone in the org). Pick whatever fits.

Open the project. The canvas loads on the right; the prompt bar is fixed at the bottom.

4. Pick the agent and chat

In the prompt bar, the agent picker dropdown sits next to the model selector. Click it and choose the agent you created. The chat panel binds to that agent — every message you send routes to its session.

Type a message and submit. The reply streams in token-by-token. As the agent runs, the canvas reflects what's happening: tool calls, generated blocks, and any peer activity all show up live.

5. Iterate on the prompt

You will tune the system prompt at least once. Open the agent in /<orgSlug>/agents, edit the System prompt field, save, and continue the same lab session — the next message uses the updated prompt. Older messages keep the prompt they ran under.

Next steps