Overview

AI Pool gives your workspace one endpoint, , that speaks the Anthropic and OpenAI APIs. Requests made with an AI Pool key are routed to the Claude or ChatGPT plans connected to your workspace, never to another workspace's plans.

For admins

  1. Sign up and name your workspace. The 7-day trial starts now.
  2. Plans → Connect Claude plan / Connect ChatGPT plan. Sign in to the provider in the window that opens. Set the plan type (Pro, Max 5x…) so savings are calculated correctly.
  3. Keys → Create key. Pick a scope (person, team, project or client) and an optional monthly budget. The key is shown once. Use Test key to send a real request.
  4. Members → Invite. Copy the invite link to each teammate. Links expire after 14 days.
  5. Billing → Activate before the trial ends (simulated in this build).

For members

  1. Open the invite link, create your login (or log in), and you're in the workspace.
  2. Optional: if you already pay for Claude or ChatGPT, connect it under Plans. Choose Private to keep it for yourself or Shared to add it to the team pool.
  3. Keys → Create key, copy it, then run the setup script below.

Configure your tools

One-line setup (Claude Code, Codex)


      

The script asks for your key (it's never put in a URL), checks it works, backs up your existing config, and configures the tools it finds.

Claude Code (manual)

Add to ~/.claude/settings.json:


      

If Claude Code was signed in to a personal account, run /logout once so it uses your AI Pool key.

Codex CLI (manual)

Add to ~/.codex/config.toml, then export AIPOOL_API_KEY="aip_…" in your shell profile:


      

Anything else

OpenAI-compatible clients: base URL . Anthropic-compatible clients: base URL . Use your AI Pool key as the API key.

Private vs shared plans

Shared plans serve everyone in the workspace. Private plans serve only the person who connected them. Their requests use the private plan first and fall back to the shared pool when it's rate-limited or doesn't offer the model. Admins see every plan; members see shared plans and their own.

Keys & budgets

Keys can belong to a person (usage is attributed to them) or to a team, project or client. Admins can set a monthly dollar budget, valued at API list prices, and/or a token budget. When a key reaches its budget, requests return 429 budget_exceeded until the next calendar month or until an admin raises it. Usage is counted within a few seconds, so a burst can go slightly over.

Capacity & overflow

Each response from Claude and ChatGPT plans reports how much of the 5-hour and weekly allowance is used and when it resets. The Capacity page shows this per plan once traffic has gone through it. An overflow API key (Anthropic or OpenAI) is used only when no plan has capacity; its usage is billed by the provider per token.

Troubleshooting