Quickstart: build your first agent

This walks you through building and running your first agent, end to end. Each step is something you do right now, in order.

1. Sign up

Create your account and organization. Every agent, run, and connector you create belongs to this organization, scoped by role.

2. Add a model key, or use credits

Before an agent can run, it needs a model to call. You have two options:

• Bring your own key (BYOK). Paste your OpenAI or Anthropic API key. It's encrypted (AES-256-GCM) and never shown back to you. Runs on your key bypass platform metering — you pay the provider directly. See bring-your-own-keys.
• Use platform credits. On a paid plan with a positive credit balance, runs use the platform key and decrement your balance automatically. If you're on the free plan with no key added, you won't be able to run an agent yet — add a key or upgrade first.

3. Open the conversational builder and describe the agent

Open the conversational builder and describe what you want the agent to do in plain language — for example, "check my inbox for support requests and draft a reply using our knowledge base." The builder configures the nodes and tools for you based on the description. See the conversational builder.

If you'd rather wire things by hand, switch to the canvas builder — a node graph you build yourself (desktop-only).

4. Review the tools it wired

Look over what the builder configured: which tools it attached (code execution, API calls, knowledge search, a connector action, MCP tools, subagent delegation), and their arguments. Adjust anything that isn't right before running.

5. Run it once and watch the execution trace

Trigger a run. The agent runs on a bounded, multi-turn tool-calling loop: it picks a tool, the arguments are validated, the tool executes, and the result feeds back in for the next step — up to a fixed step limit. Any destructive tool call pauses for your approval before it executes.

Watch the run live, or review it afterward — every tool call, its arguments, and its output are recorded in the execution trace. See execution traces.

6. Release and (optionally) deploy

Once the agent behaves the way you want, release a version. From there you can deploy that version to your own cloud (BYOC) — this activates it on the managed runtime, re-enabling its schedules and triggers — or leave it running as-is and iterate on the next version.