Privacy

Privacy Policy

Draft · last updated 27 August 2026

Draft — pending legal review

Draft

This document describes how TAZM actually works today and is written to be accurate rather than generic. It has not yet been reviewed by a lawyer. It will be finalised before TAZM is generally available, and the date below will change when it is.

TAZM coordinates AI workers to build software for you. That means it touches three things most products don’t: the credentials you use with AI providers, the code and documents you are building from, and machines you designate for work to run on. This policy is specific about all three.

What we hold

Your account: email address, account status, and the dates the account was created and verified. Your password is stored only as a salted hash — it cannot be read back, by us or by anyone else.

Your work: the projects, documents and files you upload, the missions you run, the requirements derived from your instructions, the evidence collected, the decisions you make when a mission escalates, and the artifacts produced.

Your operations record: mission events, authorization records and security events. These are append-only by database constraint, which means they cannot be silently altered or removed — including by us. That is deliberate: an audit trail that can be edited is not an audit trail.

Your usage: turn counts and provider costs per mission, so you can see what a mission cost while it runs.

Provider credentials

When you connect an AI provider, the credential is written to disk under your account’s own directory with file permissions set so only the service can read it. It is not stored in the database, and the API never returns it — requests for your connection get a masked value such as sk-…4f2a, never the key. Our records of it hold a storage path and a fingerprint, not the secret.

You can disconnect a provider at any time, which removes the stored credential. We do not use your provider credentials for anything other than running your missions.

What leaves TAZM

Running a mission means sending instructions and relevant project content to the AI providers you have connected. Their handling of that content is governed by your agreement with them, not by this policy. TAZM does not choose your providers for you and does not resell their capacity.

Missions execute on build servers you designate. Your code is worked on where you told TAZM to work on it. We do not copy your repositories to infrastructure of our own for execution.

What we do not do

×We do not sell your data
×We do not train models on your code or documents
×We do not read your projects except where needed to operate the service or where you ask us for support
×We do not put credentials or secrets in URLs, logs or analytics

Deletion, and its limits

You can delete projects and files, and disconnect providers, at any time. Be aware of one deliberate constraint: mission history is append-only. Deleting a mission tombstones it — it stops appearing in your lists, and its events, requirements, decisions and evidence remain in the record rather than being erased. This is what makes the audit trail trustworthy, and it means “delete” here does not mean “never happened”.

If you close your account, tell us and we will explain exactly what is removed and what is retained, rather than making a promise this architecture cannot keep.

Security

Workers run in isolated environments with restricted privileges, and tenant separation is tested adversarially rather than assumed. The current state of each control is published on our Security page, including the controls that are not finished. We would rather tell you what is proven than imply everything is.

Changes and contact

If this policy changes materially we will say so rather than quietly updating the date. A published contact route and security disclosure process arrive with general availability.