CloudChef

1.8 public betaEffective

Privacy

CloudChef’s own account stores only the records needed to authenticate you, connect Cloudflare, locate your user-owned runtime, and operate the service, plus narrow product telemetry if you opt in. Project and conversation data stays in the Cloudflare account you build in: your own, or your team’s; the browser keeps only rebuildable in-memory views.

On this page

What CloudChef stores in its own account

The operator control plane stores:

  • your sign-in identity: Cloudflare, Google, or a verified email address
  • your Cloudflare account details, if you connect one
  • when you joined the waitlist for a space CloudChef runs, if you did, until you leave it
  • teams you belong to: their names, your role, whether you may build new apps of your own there, and invites sent to your address (with the role and projects they give)
  • per-day counts of model turns and tokens used in a team's account
  • projects: each one's name, which Artifacts repository in the Cloudflare account holds it, the GitHub repository it was imported from (if it was), its live address, and who may view or edit it
  • GitHub connections, yours or a team's: which GitHub accounts CloudChef's GitHub App is installed on, used only to import repositories
  • team project requests (asking for a draft to become a team project): its title, note, preview URL, and which saved version of it was sent; once an admin decides, only its title, that version and the decision are kept
  • when you sent each team invite and created each team, kept for 30 days because they count toward the limits on invites and teams, even after the invite is revoked or the team deleted
  • each account's model settings (the model new chats start on, and which models the picker offers) and your choice whether AI Gateway logs your model requests
  • your page and posts: the line about you, a post for each public project you publish (its caption, live address, a screenshot of it, and any clip of it you approved, kept in CloudChef's own storage), who you follow, what you liked, your comments, your notifications, copies you asked for before you had a space, and reports you sent (a visitor who reports is kept only as a one-way hash of their internet address, to limit reports)
  • hashed authentication sessions and emailed sign-in links
  • encrypted OAuth credentials
  • granted scopes and connection status
  • user-runtime names and endpoints
  • provisioning status
  • narrow operational records

It also holds sampled service logs and, only if you opt in, allowlisted product telemetry.

The control-plane database does not store:

  • prompt or transcript bodies
  • project source
  • deployment plans
  • generated application data
  • raw AI responses

Your page, your posts, their screenshots and clips, their captions, comments and like and follower counts are public: anyone on the web can see them. Who follows you is shown only to you; who liked a post, only to its author. Your email is never shown. Making a project private takes its post off your page and deletes its screenshot; erasing your account deletes all of it. Screenshots are taken by Cloudflare Browser Rendering in CloudChef's own account, which opens your site's public address as any visitor would, and each posted site's address is checked with Cloudflare's URL Scanner. A clip of your app is made by the builder in your own workspace and Cloudflare account, when you ask for one: pictures of your live site and lines spoken by Workers AI. It stays in your account until you press Approve in the chat, and goes on your post only then; Remove clip on the post takes it off.

Chat catalogs, transcripts, generated files, workspace state, deployment records, and generated Cloudflare resources are stored in the connected Cloudflare account. CloudChef accesses them to perform your requests, but does not copy them into the operator control-plane database.

Operator, scope, and roles

DOUS SOFTWARE INC., an Ontario corporation (Ontario Corporation No. 1001622428), operates CloudChef and is the controller for personal data CloudChef processes to operate the service. Its business correspondence address is 350 Bay Street, Suite 1300B, Toronto, Ontario M5H 2S6, Canada. This includes account, authentication, Cloudflare-connection, requested AI-building workflow, support, security, and optional product-telemetry processing. Persistent server-side project and conversation state remains in the connected Cloudflare account. You control that account and are responsible for processing performed by applications you publish. Cloudflare and GitHub also process data under their own terms and privacy roles.

Teams

A team shares one Cloudflare account among the people its admins invite. Members sign in with Google or an emailed link and do not need Cloudflare logins of their own. Each member builds in a separate workspace in the team's account, so their chats, transcripts, project files, and deployed applications are stored there, under that account's controls. Its administrators, and anyone with access to that Cloudflare account, can read them, including the AI Gateway logs of each member's model requests, unless that member turned that logging off in Settings. Each member decides that for themselves; an admin cannot turn it back on. The team, not CloudChef's operator, decides how that account is used.

Admins see each member's name, email, role, and this month's count of model turns and tokens in the team's account. Those counts contain no prompt, response, or project content. An admin can pause a member, who then cannot open their workspace or use the team's Cloudflare authorization, or remove them, which also deletes their workspace from the team's account and declines their open team project requests; applications they deployed stay there. An admin can rename the team or delete it. Deleting a team deletes every member's workspace from its Cloudflare account, revokes CloudChef's authorization there unless anyone, you included, still builds in that account on their own or in another team, and removes the team's records from CloudChef: its members, invites, usage counts, GitHub connections, projects, and team project requests. The projects' code and history in the account's Artifacts, and applications deployed from it with their databases and storage, stay in the Cloudflare account; repositories imported from GitHub and CloudChef's GitHub App stay on GitHub until removed there.

Every project's code and history is a git repository in Cloudflare Artifacts, in the Cloudflare account it belongs to. An existing project can be imported from GitHub once: a public repository by its address, a private one through CloudChef's GitHub App, which you or a team admin install on the GitHub accounts that hold them, choosing which repositories it may read. CloudChef stores those installations and where each project was imported from, not the repositories' contents or any GitHub token: an import copies the repository's history into Artifacts with a short-lived, read-only token limited to that repository, and nothing is sent back to GitHub. When a member asks for a draft to become a team project, the draft's own repository, already in the team's account, becomes the project if an admin approves it. GitHub's own terms and privacy statement govern repositories on GitHub.

AI processing

The user-owned workspace runtime sends the prompt, conversation context, and project files needed for a request to the selected builder model through Cloudflare AI in the connected account. The in-chat model picker offers the Workers AI models in CloudChef’s model catalog that the account’s current Cloudflare catalog also serves and that meet the builder’s function-calling and context requirements. When the account’s default AI Gateway holds the account’s own key for an AI provider (Anthropic, OpenAI, xAI, or OpenRouter), whether an admin of the account (or, for your own account, you) added it in Cloudflare or pasted it in Settings, the picker also offers that provider’s models: those requests go from the same runtime through the account’s AI Gateway to that provider, which bills that key and handles the content under the account’s agreement with it. The key is kept in the account’s Cloudflare Secrets Store and attached to its AI Gateway; CloudChef passes a pasted key to Cloudflare once and never stores, logs, or reads it, and removing a key CloudChef added deletes it from the Secrets Store. A team’s admins, or you in Personal, also choose the model new chats start on and which models the picker offers. Every request goes through the account’s default AI Gateway. Unless you turn it off, the gateway logs each request in that account: what was sent (the prompt, conversation context, and project files) and the model’s reply, labelled with the conversation and the person who sent it, so the account can review what a conversation did. The logs follow AI Gateway’s retention settings in that account, and anyone with access to the account can read them, which for a team includes its admins; CloudChef’s operator receives no copy. You can turn this logging off under AI request logs in Settings. It is your own choice, wherever you build, and it applies to your later requests: they still go through the gateway but are not logged and carry no conversation or person label. Logs already recorded stay in the account until its AI Gateway retention removes them or someone with access deletes them in the Cloudflare dashboard. AI Gateway caching is skipped for every request. Cloudflare describes its handling of Workers AI content in its Workers AI data-usage notice. CloudChef does not use AI to make decisions that produce legal or similarly significant effects about you.

Browser storage and telemetry choice

Essential browser storage includes a 30-day authentication cookie, short-lived OAuth and recovery state, theme and builder-model preferences, and a pending prompt in tab-scoped session storage. Chat, transcript, and workspace presentation caches remain only in memory and are rebuilt from your Cloudflare account after a reload. Preferences remain until replaced or cleared in your browser.

Optional product telemetry is off until you allow it. Your choice is stored locally. CloudChef also honors Global Privacy Control and Do Not Track. Disabling telemetry stops future telemetry but does not remove events already received. You may withdraw telemetry consent at any time without affecting processing that occurred before withdrawal.

Where data is held and disclosed

Workspace metadata, Agent transcripts, project files, Computer state, and generated infrastructure remain in the connected Cloudflare account. Data is disclosed only as needed to operate the service, follow your instructions, protect the service, or comply with law.

Cloudflare provides authentication integration, Workers, D1, R2, Durable Objects, Containers, Computer, Workers AI, AI Gateway, Secrets Store, the Cloudflare AI model catalog, observability, and related infrastructure. An AI provider (Anthropic, OpenAI, xAI, or OpenRouter) processes requests to its models only for an account whose AI Gateway holds that account’s own key for it, as that account’s provider. GitHub processes information submitted through public support issues or private security reports. Control-plane observability is held in the operator’s Cloudflare account; workspace, Computer, and generated-application observability is held in your connected Cloudflare account. CloudChef does not sell personal data, share it for cross-context behavioral advertising, or use it for targeted advertising.

International processing

Cloudflare and GitHub operate globally, so data may be processed outside your country. CloudChef does not currently offer a selectable residency region. Cloudflare describes its transfer safeguards, including Standard Contractual Clauses where applicable, in its Data Processing Addendum. Review Cloudflare’s Privacy Policy and GitHub’s privacy statement, which describes GitHub’s international data-transfer practices.

Retention in the operator control plane

In the operator control plane:

  • OAuth authorization and Google sign-in state expires after 10 minutes, GitHub installation state after 15
  • emailed sign-in links after 15 minutes, and team invites after 7 days
  • authentication sessions after 30 days
  • the records of when you sent each team invite and created each team after 30 days
  • unreferenced encrypted credential records become eligible for removal after 24 hours

Maintenance runs every 15 minutes in bounded batches, so backlog or retries can delay physical removal. Account, connection, and runtime-locator records remain while the account and service are active or until they are no longer needed or a verified request is fulfilled. There is no automatic inactive-account purge.

The operator’s current Cloudflare plan retains sampled control-plane Workers Logs and traces for seven days. CloudChef does not copy them to another log or trace store. Cloudflare may retain aggregate control-plane Worker metrics for up to three months; those metrics are not a user-addressable event ledger.

Cloudflare D1 Time Travel keeps control-plane database changes recoverable for up to 30 days under the current plan. User-owned D1 recovery windows depend on the user’s Cloudflare plan. CloudChef will not intentionally restore erased records from recovery history. If broader disaster recovery reintroduces them, the erasure must be reapplied unless a lawful retention exception governs.

Erasing and exporting operator-held records

Settings contains a self-service control that erases every record the operator holds for your account:

  • identity and profile
  • authentication sessions and emailed sign-in links
  • encrypted Cloudflare credentials
  • connection metadata, granted scopes, and model settings
  • your runtime locators and team memberships
  • your team project requests, and GitHub connections you started
  • the records of when you sent team invites and created teams, so a team's daily invite limit stops counting the invites you sent

It also deletes your workspace in each team's Cloudflare account, as leaving the team would, and says which it could not reach. If you connected a Cloudflare account of your own, it asks Cloudflare to revoke CloudChef’s authorization unless someone else still builds in that account with it, in a team or on their own (it then names the teams that still use it, or says that another person working on their own does), and tells you when Cloudflare did not confirm a revocation so you can remove it yourself. Because erasure is irreversible it requires a sign-in completed in the last ten minutes, an exact typed confirmation, and an explicit acknowledgement of what is retained. Erasure leaves a team's Cloudflare authorization alone while others still build in it; if you are its only remaining member, the team is removed, and its authorization is revoked unless someone else still builds in that account. If you are the only admin of a team others belong to, make someone else an admin first. Repeating it is harmless. If you sign in with Cloudflare and have already revoked CloudChef’s authorization, you can no longer sign in that way, so use the request path below instead.

Settings also contains a self-service export of those same operator-held records. It saves a JSON file, carrying a schema version and an export timestamp, containing:

  • your identity and profile, and whether your model requests are logged in AI Gateway
  • your Cloudflare connection metadata, granted scopes, its default model, and which models the picker offers
  • the existence, storage time, and key version of your encrypted credential record
  • your runtime locators
  • your team memberships, invites addressed to you, and your daily usage counts
  • when you sent each team invite and created each team, while those records are kept
  • your own projects and the team projects you were given, and the team project requests you sent
  • GitHub connections you started, and sign-in links emailed to you
  • your authentication-session and OAuth-state records

Encrypted credential material, initialisation vectors, credential handles, session token hashes, and the hashes and states that sign-in links, invites, and authorizations in progress are checked against are never exported. Because a single file discloses your whole account record, the export requires the same sign-in completed in the last ten minutes that erasure requires. Each section is bounded at 200 records and reports the untruncated count beside them, and a section that could not be read is named in the file, which reports itself incomplete rather than looking whole.

What removal does not delete

That control deliberately deletes nothing inside your own Cloudflare account, and nothing in a team's but your workspace there. Workers, D1 databases, R2 buckets, KV namespaces, Containers, Durable Objects, and Agents that CloudChef deployed are retained, keep running, keep billing to your account, and are yours to remove. Chats, transcripts, project files, and deployment records also stay there under that account’s controls, and the account export does not contain them: download individual project source with Download code in the project header, and use your Cloudflare account’s own tools for the rest. The browser holds no persistent chat, transcript, or project replica. Its theme, model, telemetry-choice, and tab-session data are not reachable from the server and are in no export; the authentication cookie is sent with requests so the server can verify sign-in. Settings lists exactly what to clear.

Workspace Worker and Computer container logs, AI Gateway request logs (when you have not turned them off), and generated-application Worker logs and traces, remain in your connected Cloudflare account. Their access and retention follow that account’s permissions, plan, product settings, and Cloudflare controls. CloudChef does not copy them into the operator’s log or trace store, and removing a project does not immediately erase provider-retained observability.

Removing a project hides it from the active project list and makes its Agent and workspace eligible for teardown no earlier than 30 minutes later. CloudChef also schedules deletion of the project’s generated Worker and all of its versions and Durable Objects, D1 databases, KV namespaces, and R2 buckets from your connected Cloudflare account. R2 objects are emptied in bounded batches before the bucket is deleted; provider failures or revoked authorization can delay cleanup and are retried while access remains available. It is not complete erasure: catalog, transcript, deployment, provider-retained observability, and browser records remain under their applicable retention boundaries. Browser data remains until you clear it. Delete projects and allow this cleanup to finish before deleting your CloudChef account data: account deletion revokes the authorization cleanup needs, so a project deleted inside this window keeps running and billing until you remove its resources in your Cloudflare account.

GitHub retains public support issues and private security reports according to its policies and repository controls. Records may also need to remain for security, fraud prevention, legal compliance, disputes, or backups.

Your choices and rights

Depending on applicable law, you may request access, correction, portability, restriction, objection, or erasure and may complain to the data-protection authority responsible where you live or where an alleged infringement occurred. Access to the records the operator holds, a machine-readable copy of them, and their erasure are all self-service in Settings, as described above. For everything else — correction, restriction, objection, data the operator does not hold, or a case the Settings controls cannot reach — start with the public Support form and include only the request type and your GitHub handle. If a private method can be arranged, a maintainer will identify it in the issue; until then, do not provide sensitive information. CloudChef does not yet provide a verified confidential privacy inbox.

CloudChef may request proportionate information to verify control of the relevant sign-in identity (Cloudflare, Google, or an email address) before disclosing or deleting data. Public-beta response targets are not legal deadlines; where privacy law sets a deadline, including the GDPR’s usual one-month period, that deadline governs. Some rights and deletion requests are subject to lawful exceptions, and current technical limits are described above rather than treated as a waiver of those rights.

Security and age limit

Controls include encrypted Cloudflare credentials, hashed session and capability tokens, same-origin checks, short-lived runtime capabilities, tenant binding, bounded request sizes, deployment approval, and security readback. No system is risk-free. CloudChef is for adults and is not directed to anyone under 18, who must not use the service.

Changes and contact

Material changes will be published with an updated version and effective date and, when practical, additional notice in the service. Use Support for privacy questions, sharing only the request type and your GitHub handle in the public form.