Skip to content

Connect an AI credential

Every agent needs a way to think: a model to call. CommandChain gives you three ways to provide one, and you can mix them. This page explains each and where you enter it.

  • Trial credits. In the cloud beta, the platform funds your team’s thinking up to a granted balance. Nothing to enter, and nothing to install. This is how a new account works on day one. See trial credits for how the balance works and what happens when it runs out.
  • Your own AI credential (Direct-LLM). Bring an Anthropic, OpenAI, or Gemini key and your agents call that provider directly. Usage bills to your provider account, not to trial credits.
  • A runner. With a runner connected, your agents get their completions from a coding-agent tool running on your own machine, authenticated with a coding-agent login you capture once. This is the default backend.

In the app, open Settings, then AI Backend, choose Direct LLM, and enter your provider key. From then on your agents call that provider directly, and usage bills to your account.

The backend is a per-tenant setting with an optional per-agent override, so you can mix: most of your team on a runner, one agent pinned to Direct-LLM, or the reverse.

When agents execute on a runner, they use a coding-agent login rather than a raw API key. You capture that login once from a machine you are already signed in on, and it is stored for your tenant and released to a single run at a time, never kept on disk by the runner.

The runner installer captures it for you during pairing. You can also push it explicitly from a paired runner:

Terminal window
commandchain-runner creds push

This detects the coding tools you have signed into (Claude Code, Codex, Gemini) and uploads what it finds. Stored credentials are listed as metadata only, never as token material.