Security and your data

Kichma holds the record of how your product came to be — requirements, specifications, decisions, and who made them. This page says what happens to that record: where it is kept, who else sees it, how it is protected, and how to take it out or delete it.

It also says what we have not done yet. A security page that rounds "planned" up to "done" is worse than none, and you would find out at the worst moment.

Where your data lives

Everything you store in Kichma is in one Postgres database, run by Neon. It is hosted in AWS us-east-1 (Northern Virginia, United States); we do not offer an EU region today. The application runs on Vercel. Its server code runs in AWS us-east-1 (Northern Virginia, United States), the same region as the database.

All customers share that database. Your organization's data is kept apart from everyone else's by the application: every request checks that you are a member of the organization that owns what you asked for, and a request for something you are not a member of is answered as if it does not exist. There is no separate database per customer.

Who else handles it

These kinds of service receive customer data so that Kichma can work. Which company provides each one, where it processes your data, and how we know it is in use are on the subprocessor list.

AI model provider

The text of the documents you are working on, and your prompts. No account credentials and no payment data. The interview and drafting agent. Generates and critiques document content at your request.

Hosting

Everything in a request: IP address, user agent, URLs, and request bodies in transit. Application hosting and compute. Serves every request.

Database

Account details, organisation and project data, every graph node and its history, and the audit trail. The Postgres database. Stores the product record itself.

Payments

Billing name, email and address, and payment details — card numbers are entered on Stripe’s own checkout and never reach Kichma. Payments and subscription billing.

Email

Recipient email address and the message content. Transactional email: sign-in links, invitations, billing notices.

Error monitoring

Error messages, stack traces, the URL in use, and the identifiers of the organisation and project involved. No user identity and no document content. Error reporting, so faults are found before they are reported.

Sign-in and repository integration

Your GitHub account identity, and metadata about the repositories you connect. Kichma reads repository content only to check claims about it. Sign-in, and reading the repository activity that advances your plan — commits, branches and pull requests.

The AI model provider sees your content. When an agent drafts a document, interviews you or checks coverage, the text it works on is sent to our model provider. That is the product working as designed, not an edge case, so it is the first thing on this list worth weighing. It handles that text under its commercial terms: it does not use it to train models, and it deletes it within 30 days, except content flagged under its usage policy, which it may keep longer, or where the law requires it. We do not have a zero-data-retention agreement with it today.

Health data. Kichma is not set up to hold protected health information. Please do not enter patient data unless we have first signed a business associate agreement with you — write to hello@kichma.ai before you do.

What we deliberately do not send

Error monitoring is configured to see as little of your content as it can still be useful with:

  • No session recordings. Session replay is off — on Kichma, the screen is your specifications.
  • No request bodies. Collection of HTTP bodies, on by default in our error monitor, is turned off: our request bodies are your documents.
  • Identifiers, not content. Error reports are tagged with the organization and project identifiers involved and the page path — not document text, people's names or email addresses. Two caveats, stated plainly: a project's identifier is derived from its name, so a report can reveal a project's name; and an error message can occasionally quote the value that caused it.

These are commitments in our configuration, and an automated test fails if a dependency update ever turns one of them back on.

How it is protected

In transit. Everything between your browser or the CLI and Kichma travels over HTTPS. The connection from the application to the database is encrypted and verifies the database's certificate.

At rest. Neon encrypts the stored database with AES-256, with keys held in AWS Key Management Service. That protects against stolen disks and compromised infrastructure. It does not stop Neon itself from reading the data, because Neon holds the keys — that protection is contractual, not cryptographic.

No passwords to steal. You sign in with GitHub or a single-use link sent to your email; Kichma stores no passwords. Sessions live in the database, so signing out ends yours immediately rather than when a token expires, and an admin removing you from an organization takes effect on your very next request.

Long-lived credentials are stored only as hashes. CLI credentials, CI scan tokens, invitation links, document-share links and device sign-in codes are kept as SHA-256 hashes, never in plain text, so a copy of the database would not contain a working one of them. Two exceptions, as in most web applications: the browser session identifier, which is kept so a session can be looked up and ended; and, if you sign in with GitHub, the access token GitHub issues for that sign-in.

The audit trail

Every change to the record — who made it, when, and what it was before and after — is written to an audit log the database will not let anyone edit or delete, including us through the application.

The log is sealed: entries are chained together in batches, and each batch is signed with an Ed25519 key whose public half is published at /.well-known/kichma-audit-keys.json. An exported log carries its own proof, which you can check on your own machine — without trusting us — with kichma audit verify. The full export includes every audit log on every plan (see export).

The log records the titles of things that changed and the facts about the change — not the full text of your documents.

How long we keep it

Your content is kept for as long as your organization uses Kichma. When a subscription ends, the organization becomes read-only: everything in it stays readable and can be exported, and subscribing again restores it exactly as it was. We keep it that way for at least 30 days. We do not currently delete content after that period, and we will update this page before that changes.

Kichma does not keep the files you upload for the agent to read. The text extracted from one is held only while it is processed — deleted when processing finishes, and cleared within 24 hours if a job is interrupted. What you accept from it becomes part of your record like anything else you write.

Getting it out, and deleting it

Exporting the audit log. An auditor-ready export (CSV or JSON) of the audit log, stating its own range and caveats, is part of the Regulated plan, which is not available to buy yet.

Exporting everything. An organization admin can download all of the organization's data from Settings → Export, on every plan and after cancelling: one zip with every document as Markdown, the complete record as JSON — every version of everything — conversations, comments, members, and the audit logs with their seals, which verify with kichma audit verify. Sign-in credentials and payment details are not in it; the export page lists exactly what is left out and why.

Deleting a project. A project admin can delete a project. It is immediate and cannot be undone: every document, requirement, task, conversation and share link in it is removed. The organization's audit log keeps a record that the project existed and was deleted — its name and the titles of things that changed in it — because an audit trail that forgets its own deletions is not one.

Deleting your account. If you have ever authored anything in Kichma, your account cannot currently be deleted outright: the record keeps who wrote each version, and that attribution is the audit trail the product promises. We are working out how to honour a deletion request without breaking that record. If you need your account removed, contact us and we will tell you exactly what can and cannot be done for it.

Certifications and compliance

Kichma has no SOC 2 report. No audit has been started and no auditor engaged. We are not ISO 27001 certified either. If your procurement process requires either, we cannot meet it today, and we would rather you knew that from this page than from a questionnaire.

Our database provider, Neon, holds SOC 2 Type II and ISO 27001 certifications. Those cover Neon's infrastructure, not Kichma: running on certified infrastructure does not make us certified, and we do not present it as if it did.

What we do have is listed above and can be checked: named subprocessors, long-lived credentials stored only as hashes, an append-only audit trail you can verify yourself, and error monitoring configured not to collect your content.

For teams working to ISO 9001 or 21 CFR Part 11, Kichma is designed to support those practices — an attributed, append-only change history, and sign-offs that record who approved what, and when. That is not the same as being certified or validated for them, and we do not claim it is.

When something goes wrong

We aim to tell you about a problem before you notice it. Our incident process sets how quickly we contact affected customers and how often we update them, and every serious incident gets a written account of what happened and what we changed.

Reporting a vulnerability. Email hello@kichma.ai. Please give us a chance to fix an issue before disclosing it; we will acknowledge your report and keep you updated.