Security
Your code stays yours
Hivekeeper reads your repositories to find what will break. Here is exactly what happens to them, and to everything else you trust us with.
In short
- We don't keep your code. Every scan and fix works in a temporary checkout that is deleted as soon as it finishes. Nothing is stored on local disks between runs.
- Everything is encrypted: TLS in transit, and the database, its backups and every stored credential at rest.
- The AI engine is sandboxed to that temporary checkout: no shell, no network, and it treats your files as data, never as instructions.
- Your code isn't used to train AI models, by us or by our AI provider.
- SOC 2 is planned. We're building toward a SOC 2 report; until then we're happy to answer your security questionnaire.
Your code
| What we read | Only the repositories you choose (GitHub App: only selected repositories), one branch at a time. |
| How | A shallow clone (the latest commit only) into a fresh temporary directory, deleted when the scan or fix finishes, whether it succeeded or not. |
| What we keep | Results only: the inventory of APIs, packages and resources, versions, file paths and line numbers, the short line of evidence for each finding, and fix PR details. Not your source code. |
| Fix PRs | Pushed to a new branch in your repository. We never push to your existing branches or merge anything; your review and checks decide. |
| Git credentials | Sent only to your git host, as an HTTP header: never in URLs, command lines or the checkout's git config. Git hooks and similar settings in a repository are never run. |
The AI engine
AI deep scans and fix PRs use Claude through Amazon Bedrock, in AWS. Amazon Bedrock doesn't store your prompts or outputs to train models, and doesn't share them with model providers. We don't train on your code either.
A repository is untrusted input: anyone who can commit to it can write text aimed at an AI. So every AI session:
- can only read files inside the temporary checkout (fixes can also edit files there). Anything outside it, including symbolic links pointing out of it, is denied.
- can't run commands or reach the network, and can't write into the repository's git configuration or hooks.
- ignores settings, hooks, agent instructions (such as CLAUDE.md) and tool servers shipped in the repository.
- is told that file contents are data, never instructions, and reports instructions aimed at it as a possible prompt injection in the scan summary.
- runs with none of our application's secrets in its environment.
Encryption
| Data | Protection |
|---|---|
| In transit | TLS 1.2+ for every connection to Hivekeeper (HTTP redirects to HTTPS), and TLS to your git host, AI provider and integrations. |
| Database and backups | Encrypted at rest with AWS KMS; automated backups are encrypted too. |
| Credentials and secrets | Git tokens, integration and registry credentials, SSO client secrets, two-factor secrets and billing contacts are additionally encrypted field by field with AES-256-GCM before they reach the database. |
| Passwords | Hashed with Argon2id. We never store or see your password. |
| API keys | Only a hash is stored; the key is shown once when it's created. |
Infrastructure
- Hosted on AWS in the United States (us-east-1).
- The application runs in private subnets; the database is in isolated subnets with no internet route, reachable only from the application.
- Only the load balancer is public, and only on HTTPS (HTTP redirects).
- Encrypted automated database backups, with deletion protection.
Access control
- Roles (owner, admin, manager, developer, viewer) and teams that limit which repositories people see.
- Two-factor authentication for every account; single sign-on with Microsoft Entra ID, Okta, Google Workspace or any OpenID Connect provider on Enterprise, which can be required for everyone.
- A tamper-evident audit log of every action: each entry is chained to the previous one by hash, so an edited or deleted entry shows. Enterprise can stream it to a SIEM.
Payments
Payments are handled by Stripe. Card numbers go straight to Stripe and never touch our servers. Pay-as-you-go charges for scheduled scans happen only after an owner or admin explicitly turns on automatic charges, which is recorded and confirmed by email.
Compliance and SOC 2
SOC 2: planned. We're putting the controls and evidence in place for a SOC 2 audit, and will publish the report here when it's available. Until then we'll complete your security questionnaire and walk your team through our architecture. Ask at support@hivekeepr.com.
How we handle personal data is described in our Privacy Policy.
Bring your own Bedrock Coming soon
Soon you'll be able to connect Hivekeeper to Amazon Bedrock in your own AWS account, so AI scans and fixes run on your Bedrock quota, under your account's policies, logging and region choices, and are billed by AWS to you. Want it early? Email support@hivekeepr.com.
Reporting a vulnerability
If you think you've found a security issue, email support@hivekeepr.com with “Security report” in the subject. We'll acknowledge it promptly and keep you updated until it's fixed. Please don't access other customers' data or degrade the service while testing.