Privacy Policy
Last updated: November 12, 2026
Protecting your personal data is a priority.
1. Data Controller
Jean-Baptiste CHAUVIN Contact: [email protected]
2. Data Collected
When you create an account (email sign-up, Google Sign-In, Sign in with Apple, Sign in with Microsoft or Sign in with GitHub), we collect your name, email address, and either a password (stored encrypted) or a Google, Apple, Microsoft or GitHub account identifier (the identifier provided by that service). By default, your source code and projects stay on your machine and are not uploaded to our servers. There are two exceptions. Commit Context, on by default and confirmed the first time: when you attach the agent conversation to a commit, your prompts, the agent's replies and the tools it ran, which can quote your code, are stored on our servers behind an unlisted link that anyone with the URL can open, until you delete it. Uncheck the option and nothing is uploaded. Cloud Agents, an optional feature you trigger yourself, task by task: for that run only, a disposable machine we rent clones the repository from your own git host, carries out the task, pushes a branch, and is destroyed together with the clone on it. We never read that code, and no copy of it is kept once the machine is destroyed. Outside these two features, your source code never leaves your machine. Newsletter: if you opt in to our newsletter, we additionally record whether each email is delivered, opened (via an invisible tracking pixel) and whether its links are clicked (via link redirects), only to measure engagement and improve our communications. You can unsubscribe at any time from the link in every email or from your account settings. Desktop and mobile apps: to improve the product and detect regressions, the apps send anonymous usage events and, when you close the desktop app, a downsized screenshot of the application window (which may show whatever was on screen at that moment), together with basic technical and diagnostic information (operating system, app version, device characteristics). All of this is governed by the "Share usage data" switch in Settings, under "Privacy & data", and is never sent when that switch is turned off. Your code, file contents, prompts, terminal output and project names are never transmitted in clear through this usage measurement, whatever the switch is set to; the only exceptions are the AI assistance features described below, which you trigger yourself or can disable. AI assistance features: some features send a bounded, relevant excerpt of your data to an AI provider in order to produce a result for you. The AI commit message generator sends the git diff of the files you are committing (used for the single generation and never stored). The agent advisor sends the task description you typed. The onboarding assistant, on the welcome screen of the desktop app, sends the sentence you type there when you ask it what to do, and nothing else from your machine; clicking one of the suggested answers instead sends nothing at all. That sentence, an English rendering of it, the answer you were given and an English rendering of that answer are kept with the call so we can see what new users are looking for and improve the feature. The welcome chat has since taken the place of that assistant on the first screen of the desktop app and works the same way over a short conversation: each message you type or tap there and each answer you receive, each with an English rendering, are kept together as one conversation, along with the action you then run from it, so we can see how new users get started and improve the feature. When the chat looks for one of your folders, the search runs on your own machine: the only thing sent is the folder name you gave in the conversation, and the only thing reported back is how many folders were found, never a path. The project planner, also on the welcome screen, sends the sentence in which you describe the project you want to start, together with your interface language and the name of your operating system, and nothing else from your machine: there is no project yet. It returns a plan (a project name, a technology choice, the reasons for that choice, security rules and a first brief) which is written into the project folder the app then creates on your own machine; unlike the onboarding assistant and the AI error diagnosis, neither your description nor the answer you were given is stored, only technical metadata about the call (model, number of tokens, duration, cost) is recorded. AI Suggestion, the screen shown before an agent session opens, sends the task you describe there, the folders you ticked and the optional framing you filled in, so it can choose the model, name the session and, when no agent is set, pick one; nothing leaves your machine until you click the analyze button, and setting the complexity level yourself sends nothing at all. That description, an English rendering of it and what was proposed are kept with the call so we can improve the suggestions. You can turn the screen off globally, per project or per agent. The AI error diagnosis ("CLI doctor") sends the tail of the failing terminal output and the launch command of the agent, with detected secrets and system prompts redacted before anything leaves your machine; it can run automatically when an agent fails to start, and you can turn that off at any time in Settings, under "Privacy & data". For the AI error diagnosis, the redacted excerpt and the produced answer are stored with your account so we can improve the feature. AgentsRoom Help, the assistant built into the desktop app, normally runs on your own agent CLI, on your machine, so what you ask it does not leave your computer in order to be answered. When "Share usage data" is on in Settings, under "Privacy & data", the app additionally reports each exchange to us (the question you asked, the answer you were shown, which engine wrote it and how long it took) so we can measure how the feature is used and check that its answers are actually helpful. Nothing else from your machine is sent with it, and turning that switch off stops those reports entirely. In non-dev mode, the first conversation in your room of assistants works the same way as the welcome chat: each message and each answer, each with an English rendering, are kept together as one conversation, along with the assistants you then choose and whether you start a mission, so we can understand who new users are and improve the feature. If you give a company name there, you are told it will be searched: that name is sent to our AI provider, which searches the web for the company's website, and our server then reads that public website to find its logo. The company you confirm is kept with the conversation. The country suggested to you there comes from your connection or your computer's language, and only the one you confirm is kept with the conversation. It also helps find the right company. When you then open an assistant in this mode, its job and your room's description are sent to our AI provider to suggest first requests, which are kept with the assistant in your account. You can turn all of these features off, together with the voice features and the SQL assistant described below, in the desktop app, in Settings > Privacy & data > AI helpers. That computer then sends nothing to these AI services and these features stop working on it; dictation and read-aloud only keep working with offline dictation and an on-device voice. Your coding agents are not affected: they keep talking directly to the provider you chose. AgentsRoom assistant and memories: three kinds of content written by you or by your agents are kept on our servers, each with its own protection. Conversations with the assistant and its specialists are saved to your account so you can pick them up on your other computers. They are end-to-end encrypted: encrypted on your computer with a key that only your computers hold, so we store data we cannot read. Their content is deleted after 30 days without activity; we only keep technical information (dates, size, number of times a conversation was resumed). This sync is on by default and can be turned off in Settings, under Assistant, or from the mobile app: nothing is then sent or retrieved. The assistant's memory (the notes it keeps about you and your topics) is only created if you allow it, when you are asked the first time. It is encrypted at rest on our servers with a key specific to your account, but not end-to-end: our servers decrypt it to deliver it to your devices. It rejects passwords, tokens and card numbers. When the assistant looks up a note, that note is sent to the AI provider that runs the assistant, like the rest of the conversation. You can view it, export it or erase all of it at any time. Project memory (the notes your agents write about a project: architecture, decisions, conventions) belongs to the project, not to your account: it is shared with the members of that project: the accounts that already work on it, the accounts invited to it and the team it is published to. A copy of its repository is not enough to read it. It is encrypted at rest on our servers with a key specific to the project. It is not end-to-end encrypted and is kept until it is deleted from the app. It is on by default and can be turned off project by project; do not store secrets in it. Voice dictation: if you use the voice features, the audio you record is sent to a speech-to-text provider to be transcribed; the resulting text is returned to you and the audio is not retained for any other purpose. Database and server connections: by default, the password of a connection never leaves the machine where you typed it: it is encrypted in that machine's own keychain and is never sent to us. If you turn on synced passwords (off by default, in Settings, under Account), passwords are also saved to your account end-to-end encrypted: they are encrypted on your computer with a key that only your computers hold, so we store data we cannot read and could not hand over in readable form. You can keep any password on one computer only, and turning the option off erases every synced password from your account. The rest of the connection (its name, host, port, user, default database, scope and options) is saved to your account, encrypted at rest, so your other machines list the same connections instead of having them typed again. The way you arranged a table is saved to your account, so your other machines open the same layout: hidden columns, pinned columns and column widths, identified by connection, schema, table and column names. No row and no value from your databases is ever transmitted. Kubernetes clusters: the name you give a saved cluster, its kubeconfig context, its default namespace and the path of its kubeconfig file are saved to your account, encrypted at rest, so your other machines list the same clusters. No token, certificate or cloud login is ever sent to us: they stay in your kubeconfig, on your machine. What the cluster view shows (workloads, pods, events, logs) is read by your desktop and, when you open it on your phone, relayed to it end-to-end encrypted; it is never stored on our servers. Saved queries: a statement you explicitly save in the console is stored with your account, with the name you gave it and the scope you chose (that connection, that project, or your whole account), so you find it again on your other machines; nothing is stored unless you save it. SQL assistant: when you ask it to write a query, the request you typed and the shape of the focused database (table and column names, types and keys, never a row or a value) are sent to an AI provider for that single generation. The shape is not stored; the request and the statement produced are kept with your account so we can improve the feature. Live data for HTML views: if you turn it on for a project, that choice, the connections you switched on, the URL prefixes you allowed and the request headers you typed are saved to your account, so your other machines follow (a secret appears there only as a reference to its name, unless you typed a value in clear). So are the paths of the views that used it, inside the project, with dates and counters. The statements a view runs, the rows and the responses it reads never reach our servers: your computer queries your databases and your endpoints directly. Data saved to your account: so that your computers and your phone show the same work, we keep on our servers the name of each project, its folder path and the name of each computer that holds it; your tickets and their attachments; your saved agents, teams, prompts and skills; and per-project statistics (time, prompts, tokens, cost). Unless this policy says otherwise, this data is encrypted in transit but not at rest. Deleting your account does not remove project memory, since it belongs to the project: delete its notes first, or ask us to erase it. Usage per machine: each desktop signed in to your account also sends it, per session, per day and per model, the token counts, the estimated cost, and the project, agent and computer names, including sessions started outside AgentsRoom on that computer. It never sends a prompt, a file, a path or the session identifier of the CLI. This is not the usage measurement of section 9, so the "Share usage data" switch does not govern it. You can turn it off in Settings, under "Privacy & data": nothing more is sent and what was already sent is erased from your account. Otherwise these figures are kept 90 days.
3. Purpose of Processing
Your data is used exclusively for: managing your user account, providing access to the application features, and sending service-related emails (password reset, important notifications).
4. Legal Basis
Data processing is based on contract performance (use of the service), on your consent (newsletter, usage data sharing in the desktop and mobile apps, optional features) and on our legitimate interest in measuring the audience of the website and of the apps (anonymous counts) and keeping the service secure (section 9).
5. Data Retention
Your data is retained for the duration of your registration. When you delete your account, you are signed out at once and the account is erased after a 30-day grace period; signing back in during that period cancels the deletion. Project memory is the exception: it belongs to the project, not to your account, so delete its notes first or ask us to erase it. Usage per machine figures are kept 90 days. The database is backed up every night, and deleted data leaves those backups within 7 days. Screenshots captured by the desktop app (when you close it, or when you attach one to a bug report) are automatically deleted within 30 days; only an anonymous, non-visual summary may be kept for product statistics.
6. Your Rights
Under the GDPR, you have the right to access, rectify, delete, restrict, port, and object to the processing of your data. To exercise these rights, contact us at: [email protected] If you believe your rights are not being respected, you can lodge a complaint with the CNIL, the French data protection authority (www.cnil.fr), or with the authority of your own country.
7. Hosting
The website, your account data and their backups are hosted by OVH SAS, 2 rue Kellermann, 59100 Roubaix, France. The other service providers involved are listed in section 10.
8. Data Transfers
Your account data is stored in the European Union (see Hosting). Some of the service providers listed in section 10 are based outside the European Union, mainly in the United States, and therefore receive data for the part of the service they provide: website traffic (Cloudflare), payments (Stripe, Apple), notifications (Expo, Apple, Google), signing in with Google, Apple, Microsoft or GitHub if you choose that button, and the AI assistance and voice features described in section 2 (OpenAI, TypeSafe), solely to produce the requested result. These transfers are covered by the safeguards provided for by the GDPR: the EU-U.S. Data Privacy Framework when the provider is certified under it, otherwise the European Commission's standard contractual clauses. The end-to-end encrypted traffic exchanged between your own devices (desktop and phone) may pass through a relay located outside the European Union, chosen for latency; the relay only forwards encrypted messages and can never read their content.
9. Audience measurement
To improve the product and detect regressions, we collect usage events across the website, the desktop app and the mobile app (pages viewed, screens opened, features used). On the website, this measurement falls under the French CNIL "audience measurement" exemption. In the desktop and mobile apps, it relies on your consent: it is off by default, the app asks for your choice on first launch, and nothing from it is sent until you accept, apart from the anonymous counts described below: • Your IP address is never stored. It is used on the fly to derive an approximate country and city and is then immediately discarded. • No project name, filename, prompt content or command ever leaves your machine through this usage measurement. Anything potentially identifying is hashed locally before sending. The AI assistance features described in section 2 are separate, feature-triggered processing and are not part of this measurement. • When you close the desktop app, it also sends a downsized screenshot of the window so we can understand where users stop. It is accessible only to our administrators, follows the same switch as the rest of this measurement, and is automatically deleted within 30 days. • When the desktop or mobile app hits a fatal error and shows its recovery screen, it sends the error message, its technical stack trace and the app version so we can fix the bug. Your username is removed from any path they contain, and this send follows the same switch as the rest of this measurement. • If the desktop app stops without having been quit (a crash, or closed from outside), its next launch sends how it ended, the version concerned and a few technical counters (time open, memory used, number of open terminals, internal crashes) and, for a crash on Windows, the name of the software component at fault. No path, no process name and nothing you typed is included, and this send follows the same switch as the rest of this measurement. • Independently of this switch, and on the basis of our legitimate interest, our servers record the days on which your signed-in account used the desktop or mobile app, from the requests the app already makes to provide the service, and count each day the installed desktop apps that check for updates, without storing their IP address. These figures are only read as totals (active users per day), are never used to profile you, and the per-account record is erased with your account. • Anonymous counts: unless you have declined, the desktop and mobile apps also send plain daily totals (app opens, first installs, fatal errors, features used, agent launches with their provider and model) under the same CNIL "audience measurement" exemption. They carry no account, no identifier, no content and no time finer than the day; the country is derived from the IP on the fly. They are sent from the first launch, before you answer, and stop as soon as you decline (on the first-launch card or in "Privacy & data"). • No cross-referencing with other services, no sharing with third parties, no advertising profiling. • Raw events are kept for 13 months maximum; derived data (sessions, aggregates) for 25 months maximum. • An anonymous identifier is generated locally on first launch to distinguish sessions from the same device. It cannot be traced back to your identity. • The feedback you give yourself (thumbs, "Did this answer help?") is sent when you click, even with this sharing off: only your verdict, attached to the suggestion it rates. For AgentsRoom Help, your question and the rated answer go with it, and the app says so under the buttons. • In the desktop and mobile apps, you accept or decline on first launch, then you can change your mind at any time under Settings in "Privacy & data" (the "Share usage data" switch); on the website, please email us to request opt-out.
10. Service providers
We do not sell your data. We use the following service providers, who act as processors under GDPR Art. 28 and only receive what they need for their task: • OVH SAS (France): hosting of the website, the database, its backups, the European relay and our email servers (emails and newsletter). • Cloudflare, Inc. (United States): protection and delivery of the website and the relays. Traffic goes through its network, which sees your IP address and tells us your country. • OpenAI (United States): AI assistance features, voice dictation and read-aloud, as described in section 2. • TypeSafe: helps with some decisions of the AI assistance features (model choice, agent suggestion, task title). It receives the text of the task concerned. • Stripe: payment of subscriptions and gift cards. We never see your card number. • Apple: in-app purchases on iOS, Sign in with Apple and iPhone notifications. • Google: Google Sign-In and Android notifications. • Expo (650 Industries, Inc., United States): delivery of notifications to the mobile app. • Fly.io, Inc.: temporary Cloud Agents machines, located in Paris, only if you start a run. • DigitalOcean, LLC: regional relays in Singapore and New York, which only forward end-to-end encrypted traffic. Installing an AI tool: when you ask AgentsRoom to install one, it is downloaded directly from its publisher's website (for example Anthropic, OpenAI or Google). That publisher receives your IP address, as with any download, and you use the tool under its own terms with your own account. AgentsRoom collects nothing more for this; the optional usage statistics described in section 9 only record that an installation happened.
11. Public feedback boards
AgentsRoom customers can publish a feedback board for their own clients and users. A board can carry its owner's name, logo and domain and hide any mention of AgentsRoom, including in the emails it sends. When you write to a board, its owner decides what happens to your message: they are the data controller for it, and you can contact them directly. AgentsRoom hosts the board, stores your message, your email address and the technical context sent with it, and sends the related emails on the owner's behalf, as their processor. To exercise your rights, write to the board's owner, or to [email protected] if you cannot reach them. Signing in on a board uses an AgentsRoom account, for which AgentsRoom is the data controller.
Chrome Extension: AgentsRoom.dev : AI Dev Remote Control (Chrome companion)
This section applies specifically to the "AgentsRoom.dev : AI Dev Remote Control (Chrome companion)" Chrome extension (item ID molooephodmdppiakonadbhihdfbajhg). It details the collection, processing, storage and sharing of data by the extension. Every submission is triggered explicitly by the user: the extension never captures data in the background, never logs browsing history, and only runs after a user action.
Data the extension collects
The extension processes only what is strictly needed to file your feedback: • Account credentials: the email and password typed on the sign-in form, sent over HTTPS to agentsroom.dev to obtain a session token. • Session token: a short-lived bearer token returned by the server, stored locally so you do not sign in on every submission. • Optional remembered password: stored locally only if you tick "Remember password on this device". Off by default. Cleared when you untick the box or sign out. • Page context, captured only at submission time: current tab URL, page title, any text you have selected, and (in "Point on page" mode) the outer HTML of the clicked element (truncated to 1.5 KB) plus a CSS selector. • Content you type: ticket title, description, type, and any screenshot file you deliberately attach. • Bug report to your agent, only when you click "Send to my agent": the text you typed, the tab's URL and, unless you untick the box, a screenshot of the visible tab. They go over the local loopback to the AgentsRoom desktop app on your computer, never through our servers, and reach your agent like a message you typed, so the AI provider of that agent receives them. • Local UI state: last used public backlog slug, list of slugs you explicitly joined, UI preferences.
Data the extension does NOT collect
No browsing history, no keystrokes, no clipboard content, no form inputs on other sites, no analytics, no advertising identifiers, no telemetry, no fingerprinting, no device IDs. The extension does not run on pages you do not interact with and does not read content from pages you have not opened the popup or picker on.
How the data is processed
• Authentication: credentials are verified against agentsroom.dev and exchanged for a session token. Plaintext passwords are never stored server-side beyond the standard salted hash in the user account table. • Ticket submission: the payload (title, description, type, URL, selection, HTML snippet, selector, optional image) is sent to the agentsroom.dev API and written to your backlog database. • AI-assisted title generation ("Point on page" mode only): the description is forwarded from the agentsroom.dev backend to the OpenAI and TypeSafe APIs to generate a short title. Only your description is sent. You can opt out by writing the title yourself. • No automated decision-making, no profiling, no behavioral analytics.
How and where the data is stored
Local storage (on your device): the extension uses chrome.storage.local, sandboxed by Chrome to this extension ID and not readable by other extensions or websites. It holds the session token, your email, the optional remembered password, your last used public slug, your joined slugs, and UI preferences. chrome.storage.local is not encrypted at rest: do not tick "Remember password on this device" on shared computers. Server-side storage: tickets, user accounts and optional screenshots are stored in a MariaDB database hosted by OVH SAS, 2 rue Kellermann, 59100 Roubaix, France (European Union). Screenshots are saved as files on the same infrastructure. No data is transferred outside the European Union.
How the data is shared
We do not sell your data, we do not share it for advertising, and we do not use third-party analytics or tracking. Recipients are strictly limited to: • OVH SAS: infrastructure provider acting as processor under GDPR Art. 28. No access to ticket content. • OpenAI and TypeSafe: invoked only for the optional title-generation call. Only the description text you wrote is sent, for the sole purpose of generating a ticket title. Avoidable by writing the title manually. • Public-backlog recipients: when you submit to a public backlog, the ticket is visible to the owners and collaborators of that backlog, which is the intended behavior. Your email is attached so the backlog owner can reply. • Legal authorities: only if compelled by a valid legal order under French or EU law, limited to what the law strictly requires.
Retention, user controls and rights
Locally stored data persists until you sign out, clear the extension's storage, or uninstall the extension. Server-side tickets are retained for the lifetime of the associated backlog. Account data is retained during your registration and erased after the 30-day grace period that follows an account deletion request. You can sign out at any time (deletes the token and remembered password) or uninstall the extension (removes all local storage). To exercise your GDPR rights (access, rectification, deletion, portability, restriction, objection) or to lodge a complaint, contact [email protected] or the CNIL (www.cnil.fr).
YouTube API Services
AgentsRoom uses YouTube API Services through an internal tool, "AgentsRoom Captions Publisher", only to add translated titles, descriptions and subtitles to the videos of our own YouTube channel, AgentsRoom Dev. Our apps and our website never call YouTube API Services with your account. When you watch our videos on YouTube, the YouTube Terms of Service (https://www.youtube.com/t/terms) and the Google Privacy Policy (https://policies.google.com/privacy) apply.
Data accessed
The tool only acts on our own channel, with the authorization of its owner: it reads the list of our subtitle tracks and updates the titles, descriptions and subtitles of our videos. It does not access, collect or store any personal data about you, about other YouTube users or about other channels.
Storage and sharing
The authorization token is kept only on our own computer. No data obtained through YouTube API Services is stored on our servers, sold or shared with third parties, and the tool sets no cookie on your device.
Revoking access
The owner of the channel can revoke this access at any time from the Google security settings page (https://myaccount.google.com/permissions). For any question about this use, write to [email protected].