Legal
Graphide Privacy Policy
- Version
- Version 1.2
- Effective date
- 2026-11-02
- Last updated
- 2026-11-02
1. The short version
- We keep your data only to provide the service to you. Your code, your builds, your test results and your reviews are used to do the work you asked for, and for nothing else.
- We never train on your data. Not our models, not anyone else's. We do not let the model providers we use train on it either.
- We collect no telemetry, with one exception you can turn off. Graphide's app, command line tool and daemon send us no usage analytics, no session content, no screen recordings and no crash reports. Our website runs no analytics and no session replay. The exception is agent signals (4.13): your coding agents tell Graphide CI when they start work or are about to push, so machines are ready. They send event names with the repository, branch and commit, never your code, your prompts or your agents' conversations. If your agent asks Graphide to test a commit, that commit is sent for the test.
- We never sell your data, and we do not use it for advertising.
- When you delete something, or leave, we delete it. Section 9 says how, and section 10 says how long each thing is kept.
2. Who we are
Graphide, Inc. is a Delaware corporation with a business address at 2215 Channing Way, Berkeley, CA 94704, United States. You can reach us athello@graphide.com, and about your data or this policy at privacy@graphide.com. For your account data we decide what is kept and why. For the code and data in the repositories you connect, we act on your instructions: we process it to run your builds, tests and reviews, and for no purpose of our own.
3. What this policy covers
- The service. Your Graphide account; the hosted API; the Graphide GitHub App; the CI/CD service, which builds and tests your code on machines we run; pull request reviews; and the graph of your code that all of these use.
- The software you run. The Graphide desktop app, the
grcommand line tool, thegruglocal daemon, and the earlier Graphide Editor where it is still installed. These talk to the service to do what you ask, and section 5 says what they no longer send. - The website. graphide.com, including its sign-in, signup and contact forms.
It does not cover a third party's own product that you sign in to through Graphide. If you use your own subscription to another provider inside Graphide, that relationship is between you and that provider.
4. What we keep
Everything below is kept to provide the service, and is used only for the purposes in section 6.
4.1 Your account
Your user id and email address, from the sign-in run by Supabase. We never see your password. If you sign in with GitHub, we also keep your GitHub user name and id.
4.2 Billing
If you buy a plan, Stripe takes the payment and is the seller of record. We never see or store your card details. We keep your plan, its status, the Stripe customer and subscription ids, and the usage counts your plan is billed or limited on (build minutes, reviews, model tokens and cost). These counts hold none of your code.
4.3 Your repositories, builds and tests
When you install the Graphide GitHub App on a repository, or connect it to CI/CD, we fetch the repository from GitHub and keep a copy of it, and of its git history, on our servers, so we can build it, test it, review its pull requests and keep its graph current. We run your builds and tests in isolated sandboxes on machines we operate, and we keep what they produce: logs, test results, timings, and the artifacts and dependency caches that make the next run faster.
When our CI/CD runs your jobs, GitHub hands each job the secrets it needs, such as tokens for a package registry. We use a secret only to run that job, on a machine that is deleted when the job ends. To start your jobs faster we keep a prepared machine image per repository, holding a checkout of its default branch and its dependency caches. As part of the service, we may work on private copies of your repository and its CI runs, on our machines, to make your CI faster. We do this work with the coding agents and AI models named in section 11. Treat build logs as you would on any CI service: if your build prints a secret, the log holds it.
4.4 Pull request reviews
To review a pull request, we read its changes and the code around them, send what a model needs to an AI model provider (section 11), and post the review on GitHub. We keep the review and its findings so you can see them and so the next review of the same pull request knows what was already said.
Before a new version of the reviewer reaches you, we may try it on your recent pull requests on our own machines, to check that it finds real problems and does not add noise. A trial can only read from GitHub and posts nothing; its results are kept like your reviews and seen only by our engineers. We may also have a model grade a review, to check its quality, and when our engineers correct a review before it is posted we keep the original beside it. No model is trained on any of this.
4.5 The cloud copy of your code and its graph
When you open a project in the Graphide app while signed in, the app uploads the files git tracks at your current commit, plus changed and new files git does not ignore, so a file such as .env that is not in your.gitignore is uploaded too. It also uploads the graph built from them: every symbol, signature, file path and model-written description. This is how the cloud half of the product works, and you can see what is stored withgr cloud code. Signing out stops the upload. Members of a project can read its stored files, and each read is logged.
4.6 What you look at in your own graph
To rank what matters in your graph, the app records which parts of it you open, edit or ask about, with a time and a weight. We keep these signals with your project, use them only to rank that project's graph for you and the people you share it with, and you can delete them at any time (section 9).
4.7 Embeddings of your code
To find similar code, our server sends each function of three lines or more, with its own name masked, to a model that turns it into 1,024 numbers (an embedding), and keeps the numbers per project. The request goes through OpenRouter to DeepInfra, pinned to that provider, and asks for routing only to endpoints that keep no copy and do not collect the request. You can turn this off withenabled = false under [embeddings] in Graphide'sconfig.toml.
4.8 The phone app
If you use Graphide on a phone, we keep each phone's name, platform, push address and when it was last seen, and an account-wide list of your agents (each agent's title, the folder and branch it works in, its state and times, and your computer's name) so the phone can show them while your computer is away. Conversations, code and terminal output travel end to end encrypted between your phone and your computer through our relay, which sees only that a connection happened, when, and how much data it carried. Notifications carry no content.
4.9 Clean-up and swarm runs
When you run a swarm or a clean-up tool, we keep only what each run cost, as a usage count for billing (4.2). We do not keep the prompts, the answers or the changes.
4.10 When you write to us
If you email us, use the website's signup or contact form, or report a problem withgr report, we keep what you send so we can answer it. A report bundle holds only what the command shows you before it is sent. A report also alerts our team, in our own chat and in Sentry, with its title, a short excerpt of what you wrote and your email address.
4.11 Our server logs
Our servers log each request they receive, including the IP address it came from. We use these logs only to keep the service running and secure: rate limiting, abuse prevention, and finding the cause of an incident. When our own servers hit an error, the error report (which can include a repository name or a file path) goes to Sentry so we can fix it. These are reports from our servers, not from your machine.
4.12 On our website
graphide.com runs no analytics. What it does keep is what its pages need to work: the details you type into the welcome form after signing up (your name, how you heard of us, and an access code if you have one), stored on your account; and, in your browser only, your sign-in session and a few preferences such as the colour theme. To stop automated sign-ups, the sign-in page asks Cloudflare Turnstile whether you are a person, which means Cloudflare sees your IP address and browser details for that check. Pages that show pull requests load avatars from GitHub.
4.13 Agent signals
Agent signals are on by default, and you can turn them off. They come from a small script and hook settings in a repository (.graphide/agent-signal). With them, your coding agents tell Graphide CI when they start work, change a dependency file, or are about to push or merge, so machines are ready when the push arrives.
- What they send to api.graphide.com: the repository's name, the branch, the head commit, which coding agent it is, a random session id and secret, and the kind of event. For a changed dependency or lock file, its path and a SHA-256 hash of its contents. For a merge, the pull request's number.
- What they never send: your code, your prompts or your agents' conversations. The one exception: if your agent asks Graphide to test a commit, the script pushes that commit to a hidden ref so we can test it.
- Proof of access. Once per session the script pushes an empty commit to a hidden ref outside your branches and tags. We read it to check that the session can push to the repository, then delete it.
- How long: 30 days.
- How to turn them off: for an organization, the switch in its Settings on your dashboard; on one machine,
GRAPHIDE_SIGNALS=offin the environment; in one repository, a.graphide/signals-offfile.
CI security has the detail.
5. What we do not collect
- No telemetry from your machine, apart from agent signals. No usage events, no usage pings, no feature analytics, no crash reports, no recordings of your screen or of the app's window. The one exception is agent signals (4.13), which say when your coding agents start work or are about to push, and which you can turn off.
- No session content. We do not collect your prompts, your agents' conversations, their tool calls or your edits. When you ask the service to do something with a model, what that request needs is sent and the answer comes back; we do not keep a copy to study.
- No website analytics. graphide.com sets no analytics cookies, runs no session replay and loads no tracking scripts.
- No passwords and no card details. Supabase runs sign-in and Stripe takes payment.
- No model provider credentials. If you sign in to your own model provider account inside Graphide, that sign-in goes through the provider's own flow and never reaches us.
- Nothing about your other tools. To suggest projects to open, the app reads on your own machine which folders other coding agents were used in. That stays on your machine.
6. What we use it for, and what we never do
We use the data in section 4 only to:
- provide the service to you: run your builds and tests, review your pull requests, keep your graph, show you results, and bill your plan;
- support you: look into a problem you report, or a failure of the service on your account, and fix it;
- keep the service secure: prevent abuse and enforce our Terms; and
- comply with the law, where a law or a binding order requires it.
We never use your code, your builds, your reviews or anything else you give us to train or fine-tune an AI model, ours or anyone else's. We never sell it, never use it for advertising, and never use one customer's data to serve another.
Our people look at your data only when one of the four purposes needs it, for example when you ask us why a build failed, and only as far as that purpose needs.
7. Data collected before 29 October 2026
Until 29 October 2026, earlier versions of Graphide's app, editor and website collected telemetry: usage events and a daily usage ping, crash and error reports from your machine, by default the content of agent sessions (prompts, replies, tool calls and edits, each item cut to 4 KB), screen recordings of the editor window, and analytics and session replay on our website. Earlier terms allowed us to use session content to train models.
- We have never trained a model on any of it, and we never will. That permission is withdrawn.
- We no longer accept it. Our servers stop taking telemetry from every version of the software, including versions that still try to send it.
- We are deleting it. All of it, including the copies held by PostHog and Sentry, will be deleted by 30 November 2026. Until then it is not used for anything.
- To have yours deleted sooner, email privacy@graphide.com.
8. Legal bases, for users in the EU, the EEA and the UK
| What | Basis |
|---|---|
| Your account, your repositories, builds, tests, reviews, agent signals, the cloud copy and graph, embeddings, ranking signals, the phone app, run counts | Performing our contract with you: we cannot provide the service without it |
| Billing records | The contract, and our legal obligation to keep financial records |
| Server logs and our servers' error reports | Our legitimate interest in keeping the service running and secure |
| What you send when you write to us | Our legitimate interest in answering you, or the contract where it concerns your account |
9. Deleting your data
| What | How |
|---|---|
| A repository's code, history, graph, builds, test records and reviews | Uninstall the GitHub App, or remove the repository from it. We delete them by hand within 30 days of the uninstall, or at once if you ask: email privacy@graphide.com |
| A project's cloud copy, its graph and its ranking signals | "Remove Graphide data" in the project's menu, or "Delete cloud copy…" in the app's sidebar, while signed in. Deleted at once |
| Only the ranking signals | DELETE /rank/behaviour, for one node, a subtree, or all of them |
| Embeddings | Turn them off (4.7); the stored numbers go with the project's cloud copy, or after 90 days unused |
| Agent signals | Turn them off (4.13); what was sent is deleted after 30 days, or at once if you ask |
| Your account and everything attached to it | Email privacy@graphide.com from the address on the account |
| Anything collected before 29 October 2026 | Section 7 |
10. How long we keep things
| What | How long |
|---|---|
| A repository's copy, git history, graph, build and test records and reviews | While the GitHub App is installed on it, and no longer than any period below. After you uninstall the App or remove the repository, we delete them by hand within 30 days of the uninstall, or at once if you ask |
Test results a gr ci run step sends us: test names, pass or fail, timings and, for a failing test, its error message of at most 2 KB (we keep no full CI log text) | 90 days |
| CI run and job records shown on your CI board | Runs 90 days, jobs 30 days |
| Records of the jobs Graphide CI ran: which job, on which class of machine, when and for how long | 90 days |
| Each test's verdict in your own CI runs, read from their logs, which we do not keep | 30 days, and the list of the runs and jobs they came from 90 days |
| Graphide CI's events about your jobs: when each job was queued, started and finished, and the test results it gave (each test's name, file, outcome and duration, and a fingerprint of its failure) | 400 days |
| The work of Graphide CI's agents: what they proposed, the proof behind it, a transcript of each run, and the other evidence they work from | 180 days. Decision records, and the certificates behind an adopted change, which hold no code, 1 year. Code an agent rewrote, while the change is in use and 30 days after |
| Agent signals (4.13) | 30 days |
| Records of each review's progress on a pull request: queued, posted, skipped or failed | 90 days |
| GitHub's webhook deliveries | 14 days after they arrive, once handled |
| Backups of Graphide CI's database | 35 days |
| Build machines and sandboxes | Deleted when the job ends |
| Prepared machine images of a repository (4.3) | While you use our CI/CD. When you stop, we delete them by hand within 30 days |
| Private copies we made to speed up your CI (4.3) | Deleted within 30 days of that work ending |
| Dependency caches and checkouts used to speed up builds | Deleted after 14 days unused |
| The cloud copy of a project's code, its graph and ranking signals | Until you delete them. Deleted files leave our server backups within 7 days |
| Embeddings and model-written descriptions of your code | While the project exists, and at most 90 days after last use |
| Your account | Until you ask us to delete it |
| Billing records and usage counts | As long as the law requires us to keep financial records |
| Server logs | 30 days |
| Our servers' error reports at Sentry | 90 days |
| Email, form messages and report bundles | As long as we need to answer you, and report bundles at most 90 days |
| Data collected before 29 October 2026 | Deleted by 30 November 2026 (section 7) |
11. Who else receives it
We use these companies to run the service. Each receives only what it needs for its part. The AI model providers receive your code only for the work you asked for, and only on terms, or with request settings, under which they may not train on it.
| Who | What they receive | Where |
|---|---|---|
| GitHub | The GitHub App reads your repositories from GitHub and posts checks and reviews back to it | United States |
| Microsoft Azure | Hosts our servers and the machines that build and test your code, and therefore everything on them | United States |
| Supabase | Sign-in, your account identity, and the service's database | United States |
| Anthropic | The changes under review and the code around them, to write pull request reviews; and the code we work on to speed up your CI (4.3). Anthropic does not train on API data | United States |
| OpenAI | The code we work on to speed up your CI (4.3), under settings that do not allow training | United States |
| OpenRouter and Straitly, and the model hosts they route to | Short excerpts of your code (at most a few kilobytes per function), to write descriptions and pick the code relevant to a question; prompts and context for features you use in the app. Requests through OpenRouter ask it to use only hosts that do not keep the data for their own use or train on it. Straitly does not store prompts or answers and does not train on them, and bars its hosts from doing so | United States, and the hosts' own regions |
| DeepInfra, through OpenRouter | The code of each function, with its name masked, to compute embeddings (4.7), with zero data retention routing | United States |
| Depot | For repositories whose tests also run on macOS: the repository at the commits being tested, on a machine started for that job and discarded when it ends | United States |
| Cloudflare (Turnstile) | On the sign-in page only: your IP address and browser details, to check you are a person | Global |
| Stripe | Your purchase and payment, as the seller of record | United States |
| Sentry | Error reports from our own servers (4.11) | United States |
| Google (Firebase Cloud Messaging), or the UnifiedPush service you choose | For the phone app: the push address and, with each notification, which agent and computer changed (identifiers only) | United States, or varies |
We will update this table when it changes, and note the change under section 16.
12. Where your data is
The service runs in the United States. If you are outside the United States, using Graphide means your data is transferred to the United States and processed there.
13. Children
Graphide is not directed at anyone under 18, and we do not knowingly collect personal information from anyone under 18. If you believe a child has given us personal information, write to privacy@graphide.comand we will delete it.
14. Security
Security lead: Benjamin McIntyre, CTO, is responsible for Graphide's security, including infrastructure security and incident response. Contact him at ben@graphide.net.
Data in transit is protected with TLS. Builds and tests run in isolated sandboxes, one job at a time per sandbox, and are discarded after the job. Access to the service's database is limited to the service itself: row level security blocks the public database roles from every table. Your sign-in token is kept in your operating system's keychain where one is available. Deleting a project removes its data and the files that hold it, so a later open cannot bring it back.
We are a small company and say plainly what we do not yet have: a formal security certification, or a tested restore of the production database. No method of transmission or storage is completely secure. If you find a security problem, write to security@graphide.com (how to report one).
15. Your rights
If you are in the EU, the EEA or the UK, you have the right to ask for a copy of your personal data, to have it corrected, to have it deleted, to restrict or object to how we use it, and to have it ported. You can also complain to your data protection authority.
If you are in California, you have the right to know what we collect and why, to get a copy, to have it deleted, to have it corrected, and not to be treated differently for exercising any of those rights.We do not sell your personal information, and we do not share it for cross context behavioural advertising.
To exercise any of them, write to privacy@graphide.comfrom the address on your account. We answer within 30 days.
16. Changes to this policy
We can change this policy. When we do we will post the new version atgraphide.com/privacy with a new effective date, and for a material change we will tell you at least 30 days before it takes effect, by email or in the product.
A change that lengthens how long we keep your data, or widens what we use it for, applies only to data collected after the change takes effect. A change that only narrows what we do takes effect at once.
Versions
- Version 1.2, effective 2 November 2026. Says that working on private copies of your code to speed up your CI is part of the service, rather than something you turn on (4.3). Says agent signals are on by default, and how to turn them off, and that a commit your agent asks Graphide to test is sent for the test (4.13).
- Version 1.1, effective 30 October 2026. Names agent signals as the one exception to "no telemetry" and says what they send and keep (4.13). Says that after you uninstall the GitHub App we delete a repository's data by hand within 30 days, rather than by an automatic sweep. Sets the retention table (section 10) to the periods the service keeps each kind of data. Names Anthropic as receiving the code Graphide CI's agents work on. Requests about your data now go to privacy@graphide.com, and security reports to security@graphide.com.
- Version 1.0, effective 29 October 2026. The first version of this policy.
17. Contact
Graphide, Inc.
2215 Channing Way, Berkeley, CA 94704, United States
About your data or this policy: privacy@graphide.com
Security problems: security@graphide.com
Anything else: hello@graphide.com
See also the Terms of Serviceand the software licence.