Privacy

What Axiom stores

Your notes are stored in Axiom's database as readable Markdown, protected by your account and by encryption in transit and at rest on the hosting provider's disks. That is not end-to-end encryption: the search index reads your note text on the server, and Axiom holds the keys, so staff with database access could in principle read it. Axiom does not sell your notes, train models on them, or disclose them to anyone else.

Your account is the note. The browser keeps no database, no vault, and no offline copy: signing in from any browser shows the same notes, and signing out leaves nothing behind on that machine.

• Notes — Markdown and frontmatter, kept until you delete them.
• Account metadata — ownership, note identifiers, versions, and timestamps.
• Template extractions — what a template recognised in a capture, until you file or dismiss it.
• Aggregate operations data — service, status, latency, and counts. Never content, search text, tokens, or identifiers.

What Axiom does not keep

Raw capture media is never written to disk. A photograph or recording lives in bounded request memory for the length of one compilation and is dropped on success, failure, timeout, and disconnect alike. A voice transcript is treated the same way: produced, used to build the compilation request, and discarded.

Prompts, model responses, transcripts, credentials, and note content are never logged, traced, or returned in an error body.

Who else sees a capture

Compilation happens once, when you capture, and it sends the capture to GitHub Copilot using Axiom's own credential. There is nothing for you to connect and no key for you to supply.

A recording reaches two companies rather than one, because it is transcribed into words by a speech-to-text service before it is compiled. A photograph reaches only the compiler. Axiom cannot control what either company does after the data crosses that boundary, and production use is blocked until their terms have been reviewed.

A capture template does not add a second look at your capture. What a template recognises comes back in the same single response, and nothing is filed anywhere until you say so.

AI runs once

AI runs exactly once, during capture compilation. Reading, searching, filtering, the calendar, workspaces, templates, filing, and export never call a model.

Forwarding is separate and off

Upload forwarding sends a chosen note to a service outside Axiom. It is off by default and controlled per note. Storing a note in Axiom does not forward it anywhere.

Disconnecting deletes Axiom's encrypted copy of your token and stops new forwarding. It does not retract notes already sent — revoke the token and request deletion at that provider separately.

Leaving

Export downloads every note as Markdown files in a ZIP, on the same format Axiom stores. Deleting your account removes the account and everything it holds, and there is no other copy — export first.

Deleting your Axiom data does not delete your separate sign-in identity or revoke every session held by the authentication provider. That is a distinct procedure and is not part of the current release.