Deploy your familiar

Ship it. Your agent goes live against your own API key — the site wraps every run in the guardrail gate, but the inference runs on your provider account, not ours.

Documentation

What "deploy" means here

There's no server to deploy to — that's the point. Your agent is a configuration (system prompt + guardrail rules) that runs against your own API key. "Deployed" means: your agent is live in the console below, wrapped in the guardrail gate, spending your provider credit — not ours.

The live loop

1. Your task is input-filtered.
2. Your browser calls your provider's API with your agent's system prompt.
3. The response is output-checked.
4. Anything the agent proposes that looks irreversible pauses for your approval.
5. You get the guardrail report for the run.

Taking it elsewhere

The export bundle contains everything needed to run your agent outside this site: the system prompt, the guardrail rules as data, and a Python starter script. Your agent, your infrastructure.

Costs

Live runs bill your provider account directly, per your provider's pricing. The sandbox is always free — rehearse there first.

FAQ

Do I need the sandbox before deploying?

The checklist requires at least one sandbox run ending in "Allowed". It's how you prove the agent behaves before it touches anything real.

Where do live runs execute?

In your browser, against your provider's API, with your key. Build Familiar has no servers in the path.

What if my provider blocks browser calls?

Some providers restrict cross-origin browser requests. If a live run fails with a network error, use the export bundle to run your agent from your own machine instead — same configuration, same guardrails.

Can I deploy without an API key?

No — live runs need your key. Set it at Your API key. The sandbox and export don't need one.