Privacy

Crited turns a specification into an application and deploys it into your own Vercel and Supabase accounts. We do not host what you build, and we never see the data your application collects from its users. Most of what a Crited-built product knows about anyone never touches us.

Crited is a product of Menuo Inc., a Delaware C corporation, operating from Chicago, Illinois, United States.

For account and workspace data we are the controller. For what you put into Crited, including your specifications, your project content and anything you connect, we act as a processor on your instructions. That includes anything other people contribute through a share link you created.

Questions: info@crited.com

• Your email, name and avatar, from Google or GitHub when you sign in. We never receive your password for those services.
• The organisations you belong to and your role in each.
• Projects and every version of their specification. Versions are append-only: editing does not delete earlier ones, because being able to see and revert what changed is the point of them.
• Conversations with the agents, including tool calls and their results.
• Generated artifacts: source archives, documents, diagram snapshots.
• Comments left on artifacts. When you share a link publicly, anyone who comments supplies a name that is stored with it: you decide who gets that link, so you decide whose name we end up holding.
• Deleting your account removes your name from comments and specification versions rather than removing the content, so the conversations other people were part of still make sense.
• Access tokens for services you connect, encrypted with AES-256-GCM before they reach the database.
• Token counts from an app you built, if it has an assistant. Four numbers and a model name per answer — how many tokens in, how many out — so your Usage page can show what your app costs you. Nothing about who asked or what they asked.
• Model provider API keys, if you ask us to install one into an app you build. They live under API keys, separately from the Connections Crited itself uses. Same encryption, same storage, and decrypted only to write them into your own Vercel project at deploy time. You can skip this entirely and add the variables in Vercel yourself.
• Whether you have dismissed the introduction. One timestamp against your account, so it is not shown again on another device.
• Repository file trees, and Notion or Slack content, but only what you explicitly import and only within the scope you granted.
• A record of each AI model call — model, tokens, cost — which is how workspace budgets are enforced.
• Standard server logs from our host, including IP address, kept for security and debugging.

• No advertising and no cross-site tracking.
• No third-party analytics following you off this site.
• No content from the end users of applications you build. Their questions and the answers they get go from your deployment straight to the model provider; we never see either.
• Registrant details for a domain you buy. Name, address, email and phone go from the form to Vercel and its registrar in a single request. We do not write them to our database, our logs or an agent transcript, and they are gone from memory the moment the purchase returns.

A key you add under API keys is never used by Crited. We write it into your own Vercel project as an encrypted environment variable, and your deployment reads it from there. It is kept apart from Connections for that reason: those are tokens Crited uses on your behalf to build and deploy, and these are not.

Removing a key here stops it going out on future deploys. It does not reach into a deployment that already has it — Vercel keeps the copy it was given until you redeploy, and revoking the key at the provider is the only thing that takes effect immediately. We say so on the screen rather than letting “removed” imply more than it does.

We cannot see what your app says on that key. Its requests go from your deployment straight to the provider, so the conversation and the bill are between you and them.

This page used to say the app reported nothing back, and promised that if that changed we would state exactly what is sent. It has changed, so: each answer sends the model name and two token counts, and nothing else. The next section is that promise being kept — read it rather than taking this sentence for it.

If an app you build has an assistant, each answer it gives sends us three things: the model name, how many tokens went in, and how many came out. That is the whole message. There is no field in it for the question, the answer, who asked, or where they asked from — and the endpoint that receives it rejects anything else.

It exists so your Usage page can show what your app is costing you. The assistant runs on your key in your Vercel project and bills you directly, so without this we would have no way to show you that number and you would have to read it off your provider's dashboard instead.

The report is sent and forgotten. If we are slow or unavailable your app does not wait and does not fail — the person asking still gets their answer, and the count is simply lost. You can also remove the two variables from your Vercel project, and the app stops reporting with nothing else changing.

A domain is bought with your Vercel token into your Vercel account. Crited never handles the money: the charge is between you and Vercel, the registration is in your name, and we are not a registrar or a reseller.

ICANN requires the registrar to hold registrant contact details, so a purchase cannot happen without them. We pass them through and keep none of them — see the previous section. What we do record is the domain name itself, against the project it was bought for, so the deploy screen can show it.

Domain registrations are not refundable. Nothing here buys anything without you pressing a button that shows the exact amount first, and no agent or background job can trigger a purchase.

Your specifications and agent conversations are sent to Anthropic to produce responses. Your content is not used to train models, by us or by our providers.

Applications you build can carry their own assistant. When they do, it runs on your API key in your deployment, and its conversations go straight from your infrastructure to the model provider without passing through Crited.

The key itself, however, does pass through us if you ask us to install it. When you paste an API key into Crited we encrypt it (AES-256-GCM) and store only the ciphertext; we decrypt it at deploy time to write it into your Vercel project, where Vercel holds it as an encrypted environment variable. So the key exists in plaintext inside our servers for the moment it is submitted and for the moment it is deployed. We never log it, never return it to the browser, and never show it again — not even truncated.

If you would rather we never held it at all, you can add the variable in Vercel yourself. Crited will tell you its exact name and link you to the right settings page, and will not ask you for the key. The assistant works identically either way; the only difference is whether the key ever touches our infrastructure.

All are established in the United States and maintain their own subprocessor lists, which change without us being told.

• Account data: for as long as you have an account. Delete it yourself in Settings, and it goes immediately rather than on a queue.
• Projects and specification versions: while the project exists. Deleting a project deletes its versions, conversations and artifacts.
• Integration tokens: until you disconnect. Disconnecting destroys the stored secret immediately; a record that the connection once existed remains.
• Usage and billing records: kept while the workspace exists, because budgets and invoices are computed from them. Deleting a project does not remove them — the spend already happened.
• Server logs: for as long as our hosting provider retains them, which is their setting rather than ours.

We share your data with the processors above and nowhere else. We do not sell personal information, and we do not share it for cross-context behavioural advertising — neither in the ordinary sense of those words nor as California defines them. We would disclose data if legally compelled, and we will tell you when we are permitted to. If Crited is acquired your data moves with the business, and you would be told before that happens.

Deletion you can do yourself, in Settings, and it happens on the spot. For access, correction, a machine-readable export, or restriction of processing, and to object to processing based on legitimate interest, write to info@crited.com and we respond within 30 days. We will not treat you differently for asking. If you are in the EU or UK you may also complain to your national data protection authority.

• Tenant isolation is enforced in PostgreSQL row-level security. Every table carries policies, and an automated check on each build fails if a table is left without them, if two policies overlap, or if one is reachable by an unauthenticated caller. A separate proof runs against real fixtures to confirm one workspace cannot read another's rows.
• Integration tokens are encrypted with AES-256-GCM before storage.
• All traffic is over TLS, and production access is limited to people who need it.

No system is perfectly secure. If we discover a breach affecting your personal data we will notify you and any required regulator without undue delay. Report a vulnerability to the same address.

Only the ones the service needs: keeping you signed in, and protecting against cross-site request forgery. No advertising or tracking cookies.

Crited is not for anyone under 16 and we do not knowingly collect their data. If you believe a child has given us personal data, write to us and we will delete it.

The rights above are the ones the CCPA gives you: to know what we collect, to have it deleted, to correct it, and to receive a copy. We do not sell or share personal information, so there is nothing to opt out of. Asking will never get you worse service or a different price.

Any change is posted here with a new date at the top. For changes that materially affect your rights we will email you at least 30 days beforehand.