Skip to content

Security: AlkeraAI/Databench

SECURITY.md

Security policy

Reporting a vulnerability

Report privately, never in a public issue or pull request. Either:

  • use GitHub's Report a vulnerability button on this repository's Security tab, or
  • email security@alkera.ai with the details and, if you can, a proof of concept.

We acknowledge a report within 2 business days and send a first assessment within 5 business days. We keep you updated until it is fixed, and credit you in the advisory if you want.

Give us a reasonable window to fix the problem before you disclose it. We publish a GitHub security advisory, with a CVE where one applies, when the fix is released.

Scope

In scope is the code in this repository: the backend, worker, model gateway, web app, CLI and daemon, box supervisor, notebook packages, and the Docker images and Compose files built from it.

Out of scope:

  • vulnerabilities in a dependency or vendored project that this repository does not make worse. Report those upstream. Tell us too if this project is exposed.
  • denial of service by volume alone.
  • findings that need an already compromised host or physical access.
  • a deployment's own configuration, such as a missing TLS proxy or a weak secret chosen by the operator.

Supported versions

Fixes land on main and in the latest release. Self-hosted installs should track the latest release. GET /health/info reports the running version and build.

Hardening a deployment

  • Set every secret in the reference Compose file to a fresh random value, including SECRET_BOX_KEY, which encrypts stored credentials.
  • Terminate TLS in front of the web container. Keep the Temporal UI bound to localhost; it has no authentication.
  • APP_ENV=production refuses to start with insecure defaults. Do not work around that check.

There aren't any published security advisories