Skip to content

The app: what stays, what leaves

The app starts a database (the memory graph: plans, tasks, notes, decisions) and a search engine for you in Docker Desktop, on your computer. They only answer connections from your own machine. Your project files stay where they are; configuration and documents live in a folder of your user account; the secrets you put in the vault are encrypted with your passphrase and the vault locks at every restart: the server cannot read them on its own.

What the assistant produces with you (notes, decisions, plans, steps) is written into that graph, at home. That is what we mean by staying sovereign over your knowledge: it is structured on your machine, not in an online service, and you can export it or delete it by removing the Docker containers.

The only outgoing flow is the chat: what you write, and the context the app attaches so the assistant does not start from scratch (notes and decisions found by search, up to three cited files, the current plan), goes to Claude Code, installed on your computer and signed in to your own account, then to Anthropic under its terms. The app does not modify Claude Code and holds no account on your behalf.

  • Model list: the app queries api.anthropic.com with your API key or, failing that, your Claude Code token read on your machine; the answer is kept for 12 hours.
  • New version: at start and then every 6 hours, the app asks api.github.com whether a newer version exists (PO_UPDATE_CHECK=0 turns it off); the desktop app can update itself from github.com, setting auto_update_app, on by default.
  • Semantic search: its model is downloaded once from huggingface.co (several hundred MB) and kept in ~/.fastembed_cache; the search itself runs on your computer.
  • If Claude Code is not installed, the app can download it from claude.ai or the npm registry; it does not update it on its own (auto_update_cli, off by default).

Only if you turn it on: another AI provider (refused until password login is enabled, then allowed project by project; the app keeps a journal of what went where, without the content); automatic pull requests with gh; a remote skill registry; Google or OpenID sign-in; MCP servers you declare yourself.

No telemetry, analytics or crash-reporting code was found in the app’s source. Logs stay on your machine.

An honest reserve: the app’s server listens on all network interfaces of your computer and, by default, without a password. On a shared network (office, café), anyone who reaches that port can read and change your data: the setup wizard says so and offers to add a password. Do it as soon as your machine is not alone on its network.

Beyond one computer

Everything above describes the desktop app on your own machine. These setups change where the data goes; each has its real status, as on the Vision page.

  • Your own server

    Available

    The server runs on a machine of yours, with password, Google or OpenID sign-in. The data is on that server, not on your desktop.

    src/auth, docs/guides/authentication.md
  • Claude Code on another machine, over SSH

    Available

    The chat runs through a Claude Code installed elsewhere, reached over SSH: the key stays in the vault, the host fingerprint is confirmed by hand. The project context therefore passes through that machine. Very recent.

    src/chat/provider/native_factory.rs
  • Another AI provider

    In progress

    OpenAI-compatible endpoint, Codex or ACP agent, with your key. The project content then goes to that provider. Only tried against test servers.

    src/chat/provider
  • MCP servers you declare

    Available

    The app can call tools of other MCP servers you configure; on connect it only probes read-only tools.

    src/mcp_federation
  • Sharing between two instances

    In progress

    Off by default. You decide project by project what is shared; copies can expire and a share can be withdrawn. The transport between instances is not wired yet.

    src/sharing
  • Remote MCP access to your instance

    Available

    An MCP client on another machine can query your instance, with OAuth 2.1. Off by default.

    src/lib.rs (remote MCP)

In every case the rule stays the same: what leaves your machine is what you chose to send, to a service you chose, with your account or key. The app does not route your data through a Freedom From Scratch server: there is none.

This site: who is responsible

The controller is Freedom From Scratch (FFS), 7 chemin de la Peyregoue, 06600 Antibes, France (see the legal notice). For any question or request: Théotime Rivière on LinkedIn.

The rest of this page covers the website only. The app runs on your computer: what you write in it does not go through this site.

This site: what is processed

ProcessingDataPurposeLegal basisRetention
Serving the pages (web server)No personal data: the server (Caddy) writes no access log; its technical logs (start-up, certificates) contain neither IP address nor visited pageSecurity and proper operation of the siteLegitimate interest (GDPR article 6(1)(f))None: the server keeps no access log (no IP address retained); only technical logs without personal data exist. The ISP box and any upstream network equipment may keep their own logs, outside our control.
Remembering your language choiceThe language code you chose (key po-locale), in your browserNot redirecting you again after an explicit choiceYour own choice; storage strictly necessary for a feature you asked forUntil you clear it; never sent to this site
Avoiding repeat requests to GitHubPublic repository figures and latest version (keys po:gh-stats, po:gh-stats-rl, po-latest-release), and the closing of the download bar (po:download-bar-closed), in your browserLimiting requests and respecting your closing of the barLegitimate interest; technical storage6 h, 1 h, 10 min (session), lifetime of the tab (session)
Showing the latest version and repository figuresYour browser queries api.github.com: GitHub receives your IP address and the usual information of a web requestShowing up-to-date download links and figuresLegitimate interestNothing kept by this site; see GitHub’s policy
Activity pageA public file of this site (/activity.json): titles of public commits and pull requests, dates, status of checks; no author name or e-mail addressShowing the real progress of the projectLegitimate interest; information already publicReplaced every 5 minutes

The page language can also be guessed from your browser’s languages (read on your device by a small script, never sent) so the page is offered in your language.

What the site does not use

  • No cookie, set or read.
  • No analytics tool, pixel, advertising or embedded social network.
  • No font, image or script loaded from a third-party service: the fonts are served by this site.
  • No form, no account, no newsletter.

The only information written in your browser is the one in the table above: it only serves to make a feature you use work or to avoid an unnecessary request, and is not used to follow or profile you. For this reason the site does not show a cookie consent banner.

Recipients and transfers

The site is hosted by the publisher itself: no hosting provider is involved, and the server keeps no access log. The ISP box and any upstream network equipment may keep their own logs, outside our control. GitHub (GitHub, Inc., a company established in the United States) receives your IP address when your browser asks it for the latest version or the repository figures: this is a transfer outside the European Union, governed by GitHub’s privacy policy. A request blocker or blocking api.github.com does not break the site: it then shows the values frozen at its last release.

Links to GitHub, Discord, LinkedIn or other sites take you off this site: their own rules apply from the click.

Your rights

You can ask for access to your data, its rectification, its erasure, restriction of processing, and object to processing based on legitimate interest. Write to the contact above; we answer within one month. To find your data in logs, give your IP address and the period.

You can also lodge a complaint with the CNIL (https://www.cnil.fr/fr/plaintes). We have not appointed a data protection officer: the site does no large-scale tracking and processes no sensitive data.

Updates

Any change to the site or the app that adds an outgoing flow or a processing (analytics, third-party service, form) first changes this page; the last-verified date is shown at the top.

Spotted an error?

Report it through the repository issues: it gets corrected or removed.

Open an issue