Skip to the content
Menu

Bring your own CLI

Each agent runs on the coding CLI you pick, like Claude Code, Codex or Gemini, on your own machine with your own logins and accounts.

Pick a CLI per agent

  • Claude Code, Codex, Gemini, or any other coding CLI, whose output GizAi reads as text.
  • A second account of one, with its own environment (CLAUDE_CONFIG_DIR, CODEX_HOME).
  • Settings → Coding CLIs adds them, or finds the ones installed. Each agent picks its CLI under Runs on, in its form.

With only Codex installed, GizAi adds it under Settings → Coding CLIs and runs every agent but the Team Lead on it.

On your own machine

  • Everything runs on your computer, with your own git repositories and your own logins.
  • Agents run on the coding CLIs you installed, with your own logins and accounts.
  • GizAi needs no account of its own.
  • Your data stays in one folder on your computer, and GizAi sends nothing about you: see Your data and No telemetry.

GizAi never sells Claude usage: agents keep running on your own Claude Code login or API key. On Windows, GizAi runs the CLIs installed on Windows itself, never inside WSL (Install).

Your keys stay yours

  • MCP sign-ins and values, and your Bitbucket login, stay in your system’s keychain.
  • GizAi never stores a GitHub token or password: pushes go over SSH with your keys, or over HTTPS with gh’s login.
  • An agent’s folders never include /, your home folder, GizAi’s data or folders with keys (Folders).

The Team Lead needs Claude Code

The Team Lead’s chat runs on Claude Code. Runs on, under the chat’s text box, moves a chat to another Claude Code account: see The chat.

What each CLI gets

  • Claude Code gets everything: the Team Lead’s chat, MCP servers and the browser, and slash commands and skills.
  • Codex has its own web search, and its sandbox blocks what it doesn’t allow.
  • Gemini fetches pages, and searches the web by its own policy.
  • Other CLIs run with their output read as text.

Memory works the same on every CLI. Which tools each CLI can use: Tools per agent.

Usage

Usage (in Company) shows the agents’ input tokens (cache included), output tokens and API cost for today, 7 days, 30 days or this month. It shows them in total with a bar per day, per agent and per project, with the Team Lead’s chat on its own line.

  • The Projects list shows each project’s API cost this month.
  • API cost is what the tokens would cost at API prices, not a bill.
  • A CLI that reports no cost (Codex, Gemini) shows its tokens and an unknown cost.

Subscription limits

Usage → Subscription, the tab the page opens on, has a block per coding CLI in Settings, so two accounts of one CLI show apart.

  • A Claude Code account shows its session, weekly and Fable limits; a Codex account its plan’s two windows (usually 5 hours and a week).
  • Each shows how much is used, when it resets and when the number was read, amber from 80% and red when reached.
  • Each lists the agents on it and the chats that run there, so you see whom to move to another account.
  • Gemini and other CLIs say GizAi can’t read their limits yet.

The numbers are the newest the CLI itself reported in a run or chat turn: Claude Code in its stream, Codex in its session log ($CODEX_HOME/sessions), and a run that hit a limit. GizAi never reads their logins or asks Anthropic or OpenAI.

Read more

Edit on GitHub (opens GitHub)