Connect the Claude and ChatGPT plans you already have. Give everyone a personal key. AI Pool spreads work across your plans, fails over when one hits its limit, and shows what every seat is actually worth.
Sign in to each Claude Pro/Max or ChatGPT Plus/Pro/Business plan once. The plans stay in your name, and you keep paying each provider directly.
Create a key per person, team, project or client. Each key has its own budget, and you can revoke it instantly.
One setup script points Claude Code and Codex at the pool. Other tools need only a base URL and a key.
Most seats sit idle most of the day. Pool them, and a few heavy plans cover a whole team. Try your own numbers.
Capacity multiples follow Anthropic's plan names (Max 5x = 5× Pro usage). ChatGPT limits aren't published as multiples, so they aren't added here.
Requests go to plans with headroom. When one hits its 5-hour or weekly limit, traffic moves to the next one automatically.
Already have Claude or ChatGPT? Connect it as private to use it first yourself and fall back to the team pool. Or share it with everyone.
Keys for people, teams, projects or clients, each with an optional monthly dollar or token budget. Revoke any key in one click.
See each plan's 5-hour and weekly usage, when it resets, and how much capacity will expire unused.
Every request is valued at API list prices. Compare that with what you pay, per person, key, model and plan.
Add an Anthropic or OpenAI API key as a last resort. It's only used when every plan is at its limit.
Consumer plans (Claude Pro/Max, ChatGPT Plus/Pro) are licensed to one person, and sharing one person's plan with others can break the provider's terms. The safe pattern is private mode: each person connects their own company-paid plan and uses it themselves, while AI Pool handles keys, budgets and reporting. Business plans and API keys (as overflow) are the sanctioned routes for shared capacity. Check your plans' terms before sharing.
On the machine running AI Pool. In this build that's your own computer or server; requests go only to Anthropic and OpenAI. Upstream API keys are encrypted at rest. Plan sign-ins are stored by the routing engine on the same machine.
Requests fail with a rate-limit error until a window resets, unless you added an overflow API key. Then AI Pool switches to it and bills per token at the provider's price.
No. Claude Code, Codex, OpenCode, Cursor and any OpenAI- or Anthropic-compatible client work by changing the base URL and key. The setup script does it for Claude Code and Codex.
AI Pool's dashboards are built from metadata only: time, model, token counts, which key and plan, and latency. Prompt and response text isn't stored, except the routing engine keeps the last few failed requests on the server for debugging.
This is a concept preview. Click through the demo, then tell us what you think.
Create your workspace