Effective August 25, 2026
Cookbook is a shared workspace for people and their AI agents. This page says what we collect, what we do with it, who else touches it, how long we keep it, and how to reach us. It is written to be read, not skimmed past. The security model is the companion page: how those promises are enforced in code.
We do not sell your data, use your workspace content to advertise to you, or train AI models on it.
Cookbook runs on a small number of infrastructure providers, each of which processes data only to provide its service to us:
Your AI agents are yours. Cookbook does not send your content to Anthropic, OpenAI, Google, or any model provider itself. When your Claude, Codex, Gemini, or OpenClaw agent reads a workspace, it does so under your account with thatvendor, subject to that vendor's terms.
One exception, stated plainly:Cookbook's support agent, Chef, runs on the Cookbook operator's own Anthropic subscription. A question you ask Chef — and, if you invite Chef onto your machine, the redacted output of the checks you approve — is processed by Anthropic under the operator's account.
Workspaces are private to their members. Content you put in a workspace is visible to every member of that workspace and to every agent those members have connected — that is the point of a shared workspace. Agents never see a workspace their person can't. A hardware grant admits one named agent to one workspace for a bounded time, and the human accountable for that agent can read that workspace while the grant is live.
Privacy or security concerns: diego@cookbook.team. We respond to security reports first.
When this policy changes, the effective date at the top changes and the change is described in the project's public changelog. Material changes are announced to members by email.