Legal
Privacy Policy
Effective October 10, 2026
See also the Terms of Service.
1. About Covin and this policy
Covin (“Covin”, “we”, “us”) keeps AI agents inside the job they were given, on a developer’s laptop or across a company’s cloud. Covin is built by a team of engineers from Meta, Palo Alto Networks and Microsoft, and is based in San Francisco, California, USA. (Covin is not affiliated with or endorsed by those companies.)
This policy covers:
- the website, covin.dev, including the meeting booking form;
- Covin Local, the app that watches coding agents on a developer’s computer;
- Covin accounts, the dashboard at app.covin.dev that Covin Local computers sign in to; and
- Covin Enterprise, which monitors AI agents in an organisation’s AWS environment through a collector that runs in the customer’s own account.
It explains what we collect, why, who processes it for us, how long we keep it, and the choices you have. The Terms of Service govern use of the service.
2. Our role: controller and processor
For organisations (Covin Enterprise and company accounts), the customer decides what is monitored and Covin processes agent activity on the customer’s behalf and under its instructions. The customer is the controller and Covin is the processor. The customer’s configuration (which buckets and networks are monitored, the setup level, the retention period, and whether AI features are on) forms part of those instructions. A Data Processing Addendum (DPA) is available for enterprise agreements.
For individuals who sign up for a Covin account, book a meeting, or visit the website, Covin decides how that information is used and is the controller.
3. The website and meeting bookings
If you book a meeting, we collect your name, work email, optional message, and the time and time zone you pick. We use them only to schedule, hold and follow up on that meeting.
Our infrastructure keeps standard request logs (such as IP address and browser) for security and operations.
We do not use advertising trackers, third-party analytics or session recording on this site. Fonts are served from covin.dev itself, so your browser does not contact a font provider.
4. Covin Local
Without an account, nothing leaves the computer. Until you run covin login, Covin Local works entirely on your computer and sends nothing to us. What it records there (including prompts and judged actions in ~/.covin/local.db) stays on the computer, under your control.
Activity metadata. After you sign in, Covin Local uploads event metadata about what your AI agents did: the agent session, the action, the resource address (for example a file path, command name or host, which can be sensitive in itself), the result, and a tool-call id. Anything that looks like a secret in an address is removed before upload.
Heartbeats and inventory. Every few minutes a signed-in computer also sends a heartbeat: the Covin Local version, its upload queue, and an inventory of the coding agents and MCP servers installed (agent names and versions, MCP server names, the package a known launcher runs, and the scheme and host of a server URL, which can include internal host names). The inventory never includes environment or header values, credentials, argument values or URL paths; some identifiers in it are sent only as keyed hashes. The inventory is on by default; inventory: false in ~/.covin/local.yaml turns it off and deletes the stored list and the alerts built from it.
Content stays on the computer by default. Covin Local does not upload your prompts, file contents, command text or agent output to Covin, with two exceptions you choose: rule text you type to have a rule drafted, and short task titles if you turn on title sharing (both below). Where Covin needs to match content (for example, the same task across two sessions), it receives keyed hashes (HMAC fingerprints under your account’s key) instead of the text. A hash is not the metadata above: the metadata is sent as is, the hashes stand in for the content.
Task titles, opt-in only. Title sharing is off by default. If you run covin share-titles on (it takes effect once you have run covin login), Covin Local sends Covin a short title for each task, so your dashboard can say what a session was for. The title is the first non-empty line of the prompt that states the task (your latest prompt that is not just “continue” or similar), with spacing collapsed, anything that looks like a secret replaced by a placeholder, and cut to about 80 characters. Secret masking matches patterns, so a title can still contain names or other details you typed in that line. The rest of the prompt is not sent. covin share-titles shows the setting and covin share-titles off stops new titles from being sent; it does not remove titles already sent. Covin Local refuses the command when an agent runs it.
Optional AI review with your own key. If you add your own Anthropic API key to Covin Local’s settings, an action Covin Local cannot settle is sent from your computer directly to Anthropic (not to Covin), under your key and your agreement with Anthropic. That request contains the session’s task (its prompts, trimmed to 4,000 characters) and the action (tool, resource and command, trimmed to 2,000 characters), with obvious secrets masked. Without a key, nothing is sent.
Rule drafting, through Covin. When you describe a rule in your own words (the dashboard’s Rules box or covin rules add "..."), that text is sent to Covin’s server and from there to Anthropic’s Claude API, which drafts the rule. Both places say so before you send. Only what you type (and your answers to any question Covin asks to clear up what you meant) is sent, with Covin’s fixed drafting instructions; no activity, file paths from your uploads or secrets are added. Covin does not log the text. If you save a drafted rule, your words are kept with it (shown on the rule and synced to your computers) until you delete the rule. Drafting has an hourly limit per account and, once enabled on the server, is available to every account without a separate opt-in. If you would rather not send any text, add rules by hand or from the ready-made ones.
5. Covin accounts
To sign you in we receive your email and identity from the sign-in method you choose: GitHub, or a sign-in link sent to your email. We record the computers you enrol, so you can see and revoke them.
We send sign-in links, the alert emails you opt into, and a weekly summary. The weekly summary is on by default for Covin Local accounts, is sent at most once a week, and only in a week when your agents ran. It says how many agent sessions ran and which agents (for example Claude Code or Codex), the names of the computers they ran on, how many files they edited and pushes they made, how many actions Covin stopped, the severity and title of open alerts, credentials an agent read and the sites it then reached, rules Covin suggests, and how many rules you have. You can turn it off at any time in Settings or with the unsubscribe link in the email.
6. Covin Enterprise
How data reaches Covin. A read-only Covin collector runs in the customer’s AWS account. It reads CloudTrail log files (and, at the recommended setup level, the VPC Flow Logs and Route 53 Resolver query logs the customer names), keeps only AI agents’ activity on the resources the customer chooses, and sends a short record of each action to Covin over HTTPS. Covin has no role in the customer’s account and nothing can connect to the collector.
What each record contains: the agent (as named in the customer’s agent map, or the IAM role name), the action, the resource (bucket and object name, including account id and region, or the public address or name an agent reached), the result, timestamps, and, for uploads and downloads, the number of bytes moved. Object names can be meaningful; customers can choose to keep them in their own account, in which case they leave it only as keyed hashes under a key held in the customer’s own AWS Secrets Manager.
What stays in the customer’s account: raw CloudTrail records, file contents (the collector reads log files, not the customer’s data), credentials, keys and tokens, request parameters, source IP addresses and user agents, activity by people (IAM users and root), traffic inside the private network, and activity on resources the customer did not choose. Covin Enterprise does not collect agent prompts or messages. Covin’s receiving endpoint refuses any record that carries fields beyond the documented set.
Sign-in. The customer’s team signs in through Google, or the customer’s own identity provider (such as Okta or Microsoft Entra ID), with roles the customer assigns.
Optional features, off unless the customer turns them on: the AI investigator, which may send event metadata to Anthropic and reads object contents only if the customer explicitly allows content reading; and bait files, which the collector places only where the customer approves, using a separate scoped role. The fast-lane containment role, which can only add or remove a deny-all policy on the agent roles the customer lists, is on by default and can be set to alert-only or off.
A customer-hosted option keeps detection and storage inside the customer’s own account.
7. How we use data
We use the data described above to provide, secure and support the service: to show you your agents’ activity, detect and stop risky actions, send the alerts and summaries you asked for, respond to you, and fix problems.
- We do not sell personal information.
- We do not share personal information for cross-context behavioural advertising.
- We do not use customer data to train AI models.
- We do not use customer data for any purpose other than providing the service to that customer.
8. Service providers (subprocessors)
We use these providers to run Covin. Each receives only what it needs for the service listed.
- Cloud infrastructure providers: hosting, storage, backups and logs for Covin accounts and Enterprise.
- Content delivery providers: the website and Covin Local downloads.
- Email delivery providers: sign-in links, alerts, the weekly summary and meeting booking notifications.
- Identity providers: sign-in for Enterprise dashboards, where the customer uses Google sign-in.
- GitHub: sign-in for Covin accounts, when you choose it.
- Anthropic: drafting rules from your own words (through Covin, as described above), and the AI review and investigation features, when enabled. Covin Local’s own-key review goes from your computer to Anthropic directly, not through Covin.
Enterprise customers can request the full list of named subprocessors, and are told before we add or replace one that processes their data, as set out in their agreement.
9. Where data is stored
Covin stores and processes data in the United States. Each customer’s data is kept in its own storage and processed by its own processes. Enterprise customers who need data to stay in their own environment can use the customer-hosted option.
10. How long we keep data
Event data is kept for the account’s retention period (90 days by default; Enterprise customers can agree a different period) and then deleted. Some records derived from events, such as analysis results and the dashboard audit trail, can currently be kept longer; Enterprise customers can ask us for the details for their deployment.
Rules, alert settings and weekly summary totals are kept while the account exists and are deleted with it. You can delete one computer’s logs, or your whole account, sooner (section 11).
At the end of an Enterprise agreement, we delete the customer’s data from our live systems within 30 days and confirm in writing. If asked first, we export the customer’s event data and findings to them.
Backups are daily disk snapshots, and deleted data stays in the snapshots taken before it was deleted. We keep a set number of the newest snapshots, not a set number of days, and older ones are removed only when new ones are taken. So deleted data usually leaves the backups after about two weeks, but we do not guarantee a time limit: if snapshots stop being taken (for example the disk is removed, or the backup policy is turned off or deleted), the remaining snapshots, and the data in them, can be kept indefinitely until we delete them.
Operational logs are kept for 30 days. Booking details are kept while needed to follow up, and deleted on request.
11. Deleting your logs and your account
Removing Covin from one computer. When you run covin uninstall, it asks where that computer’s Covin logs should be deleted: from the computer and Covin’s server, from the computer only, from the server only, or nowhere. Nothing is deleted without your answer (--logs answers it in a script). You can do the same from the dashboard under Laptops > Remove from this laptop.
- On Covin’s server, “this computer’s logs” means everything that computer uploaded and everything Covin worked out from it: its recorded events, sessions and heartbeats, the alerts and your answers that name those sessions, and the analysis results that mention them. Your account, your rules and your other computers’ logs stay.
- On the computer, it means Covin’s activity log and the background agent’s log. Your settings, rules and the backups of your agents’ config files stay, because they are not logs.
Deleting your account in Settings deletes your sign-in details, enrolled computers and sessions at once, and then removes your account’s data from Covin’s server, including every computer’s logs. Each signed-in computer is told, the next time it is online, to remove Covin and delete its own logs. So that a computer that is switched off still gets that message, we keep a record of the instruction and the revoked device key it proves itself with for 30 days, and nothing else about the account. Computers running an older version of Covin Local only stop uploading; their local logs stay until you delete them.
Deleted data stays in backups until they rotate out (see section 10), and emails we already sent stay in your mailbox.
12. Security
We protect data with measures including:
- encryption over the internet (TLS 1.2 or 1.3) and encryption of stored data and backups;
- separate storage and processes for each customer;
- signed uploads: every request from a collector or computer is signed with its own key, must be recent, and is refused if replayed;
- no standing SSH access to production; operator access goes through managed cloud identity;
- only signed, versioned software images run in production; and
- a read-only collector with no inbound access in the customer’s account.
No system is perfectly secure. If we learn of a breach affecting your data, we will notify you as the law and your agreement require. Enterprise customers can request our security documentation.
13. Your choices and rights
- You can use Covin Local without an account, in which case nothing is sent to us.
- Uploads from a computer stop when you run
covin logouton it or revoke that device in Settings. - You can uninstall with
covin uninstalland choose where that computer’s logs are deleted, remove a computer from the dashboard, or delete your account at any time in Settings (section 11). - Optional data flows (task titles, inventory, AI review) can be turned off as described above.
You can ask us to access, correct, export or delete your personal information, including under the California Consumer Privacy Act, by booking a meeting with us (see below). We will not discriminate against you for making a request. If your data reaches Covin through an organisation that uses Covin Enterprise, please direct your request to that organisation; we will help them respond.
14. Children
Covin is a business product. It is not directed to children under 16, and we do not knowingly collect their data.
15. Changes to this policy
We will post changes here and update the date above. For material changes affecting account holders or Enterprise customers, we will also notify you by email or as your agreement provides.
16. Talk to us
Questions about this policy, a privacy request, our DPA or security documentation? Book a meeting with us and we will walk you through it.
Book a meeting