Skip to content

Privacy

What we store, who else sees it, and how to remove it. Written from the code rather than from a template, so everything below is checkable against what the product actually does.

Last updated 16 September 2026.

What we store

Your work, and what it cost.

Your account. Email address, and the name and team details you enter during onboarding. Sign-in is passwordless, so we never hold a password.

Your work. Projects, the briefs you write, the agents built from them, the files those agents produce, the messages in each thread, and a record of every run — including which model each agent used and what it cost.

Your connections. When you connect a third-party system, we store the access tokens needed to act on your behalf. They are encrypted at rest and are never included in anything sent to a model.

Billing. Seat counts and usage totals. Card details are held by Stripe and never reach our servers.

Who else sees it

The processors, named.

We do not sell data, run advertising, or share anything with data brokers. These are the services the product depends on to work, and what each one receives:

Model providers, via OpenRouter and CheaperInference
Your briefs, the contents of files agents are working on, and results returned by connected tools are sent to the model chosen for each agent. This is the core of how the product works, and it is the most important line in this policy. Requests reach a model through one of two gateways depending on which model it is. Neither is permitted to train on what we send, and OpenRouter additionally receives an instruction, on every request, to route only to providers that do not retain or train on prompt content.
Supabase
Hosts the database and handles sign-in. Everything in the section above lives there.
E2B
Provides the disposable Linux container an agent runs code in. It receives whatever the agent puts in that container, and the container is destroyed when the run ends.
Resend
Sends sign-in links. Receives the recipient address and the message.
Stripe
Handles payment. Receives billing details directly from you; we receive only the resulting subscription status and totals.
Render and Cloudflare
Run and front the application. They see ordinary request data — addresses, timestamps, paths.
Systems you connect
Anything you connect a credential to receives whatever the agent sends it. You choose which of those exist and can revoke any of them at any time.

Google data

Gmail, Drive and Calendar, specifically.

Google asks that apps touching Workspace data describe it separately rather than folding it into the general case, which is fair. Each of the three connections below is optional, connected on its own, and can be removed on its own.

What we access. Only what the tools below call for. Every scope is listed with the tools that use it, so this is checkable rather than reassuring.

Google Calendar
calendar.readonly and calendar.events.readonly. Two tools, both read-only. One lists the calendars you have. The other reads events in a date range and returns the start and end, the title, the location, and how many people are attending — not who they are. No email address of any attendee is read or stored. Nothing can create, move or cancel an event.
Google Drive
drive.readonly. Two tools. One searches your Drive by name, type or contents. The other reads a file the search found; Docs, Sheets and Slides are exported to markdown, CSV and plain text in order to read them. Nothing creates, edits, moves or deletes anything in your Drive.
Gmail
gmail.modify. Four tools. Two read: searching your mail, and reading a message including its body. Two write: saving a draft in Gmail, and sending a message you asked to be sent. Nothing changes labels, archives, trashes or deletes mail — the scope permits it, we never do it, and no code in the product calls those endpoints.

What it is used for. Performing the task you asked for, in the run you asked for it in. Nothing is read speculatively, on a schedule you did not set, or to build a profile of you. It is not used for advertising and there is no advertising here to use it for.

Where it goes.To the model running your agent, because reading a thread and writing a reply is a thing a model does. That means the sub-processors named above — the model gateways, and E2B if the agent writes it into its container. It goes nowhere else. We do not sell it, and it never reaches a data broker or an advertiser.

How it is protected. The tokens that let us reach your Google account are encrypted with AES-256-GCM before they are stored; the key lives in the application environment and never in the database, so a copy of the database is inert without it. The table holding them has row-level security on with no policies at all, which means no browser session can read it under any circumstance. Tokens are never included in anything sent to a model.

How long it is kept, and how to remove it. Disconnecting Gmail, Drive or Calendar deletes the stored tokens for it immediately, and the agents lose access at once. It does not revoke the grant on Google’s side, so do that too, at your Google account permissions. Content an agent already read lives on in the run record and in any file it produced, for as long as you keep them — deleting the project removes both. To delete everything, see Removing things below.

Limited Use.traven’s use and transfer of information received from Google APIs to any other app will adhere to the Google API Services User Data Policy, including the Limited Use requirements. Specifically: Google user data is not used to develop, improve or train generalised AI or ML models, and it is not transferred to anyone for that purpose.

Your control

Removing things.

Deleting a project removes it and everything under it — agents, runs, files and messages.

Disconnecting an integration deletes the stored tokens for it. Revoking access from the other system’s side also works, and we would rather you did both.

To delete your account and everything in it, email admin@traven.ai. We remove it within 30 days. Backups roll off on their own schedule and are not searchable in the meantime.

We keep what you have not deleted for as long as the account exists, because it is the work itself — a project you cannot open is not privacy, it is data loss.

Plainly

What this document is not.

This is an accurate description of how the system behaves, written by reading it. It is not legal advice and it has not been reviewed by a lawyer. If you need contractual guarantees — a DPA, specific retention terms, a named sub-processor list under contract — write to admin@traven.ai and we will deal with it properly rather than pointing at this page.

We hold no security certification and say so here rather than leaving you to notice. See the security page for what is actually in place.