Connect a CLI subscription

Already pay for Claude Code, Antigravity, or OpenAI Codex? Use it as the AI behind the @bridge AI Chat — no second subscription, no API key. atbridge.ai runs the CLI you already have installed as a local subprocess on your own account, so there’s nothing to paste and nothing metered per message: it’s your existing plan’s own compute.

Note

Install atbridge.ai and the Thunderbird add-on first — see Install atbridge.ai. New to the in-Thunderbird side? Start with Set up the @bridge extension.

CLI connectors need a paid plan

Claude Code CLI, Antigravity CLI, and Codex CLI appear in the Provider dropdown once you have a paid plan (Personal or Pro) — and they’re included in the 14-day free trial. On the Free plan they’re hidden; the API-key and Ollama providers stay available on every plan.

Which CLI can I connect?

CLI	Command	Sign-in	Acts on your mailbox?
Claude Code	claude	your Claude account	Yes — reads + gated writes via atbridge.ai’s MCP tools
OpenAI Codex	codex	your ChatGPT account	Yes — reads + gated writes via atbridge.ai’s MCP tools
Antigravity	agy	your Google / Antigravity account	Chat & reasoning only (no MCP tools)

All run entirely on your machine. The difference is tools: Claude Code and OpenAI Codex can open, search, draft, and (with your permission) send mail through atbridge.ai’s own MCP tools, sandboxed to a per-conversation workspace. Antigravity answers and reasons but doesn’t call atbridge.ai’s mailbox tools. Every write — regardless of CLI — is still gated by your Connectors & Permissions matrix, so Codex and Claude Code can only do what you’ve allowed.

Codex model choice

On a ChatGPT account, Codex runs its own Codex model — you can’t switch to an arbitrary model, but you can pick the reasoning effort (low / medium / high) from the model dropdown, trading speed for depth.

1. Install the CLI and sign in

1. Install the CLI you want and sign in once with your own account, using that tool’s own instructions:

   • Claude Code — the claude command (claude.com/product/claude-code).
   • Antigravity — the agy command.
   • OpenAI Codex — the codex command, then codex login with your ChatGPT account.

2. Confirm it works from a terminal (e.g. claude --version, agy --version, or codex --version). That’s all the setup the CLI itself needs.

atbridge.ai finds the CLI for you

Thunderbird is often launched with a minimal PATH, so atbridge.ai also looks in the usual install locations (~/.local/bin, Homebrew, npm/bun global bins, Windows .cmd/.exe). If your CLI lives somewhere unusual and isn’t found, set an environment variable pointing at it — ATBRIDGE_CLI_CLAUDE for Claude Code, ATBRIDGE_CLI_ANTIGRAVITY for Antigravity, or ATBRIDGE_CLI_CODEX for Codex — then restart Thunderbird.

2. Connect it in atbridge.ai

1. In Thunderbird, open the @bridge Settings — the gear, or Tools → Add-ons and Themes → atbridge.ai → Options (see Set up the @bridge extension). Go to the AI providers tab.

2. Pick Claude Code CLI, Antigravity CLI, or OpenAI Codex CLI from the Provider dropdown. (Don’t see them? They need a paid plan — see the note above.)

3. There’s no API key needed — you already signed in to the CLI itself.

4. Click Connect. The status turns to a green ✓.

3. Pick the model and chat

Open the @bridge AI Chat (the @b button in Thunderbird), pick the model from the dropdown, and ask away:

“Summarise the unread thread from Maya and draft a reply.”

The request is handled locally by the CLI on your machine, on your own subscription — atbridge.ai never sees an API key and nothing is metered per message.

Good to know

• Session memory. Follow-up questions in the same conversation keep their context — atbridge.ai resumes the CLI’s own session per conversation.
• Web search. The “Allow web search” toggle has no effect for CLI connectors: the CLI’s non-interactive mode doesn’t expose a web-search tool. Use an API-key provider if you need in-chat web search.
• Budget meter. There’s no token-budget donut for CLI connectors — the CLI manages its own context window.

Related

• Connect your AI — overview
• Connect a local model (Ollama)
• Use the @bridge AI Chat