Skip to content

Security: KelianS/vyzio

SECURITY.md

Security policy

Vyzio watches a home and keeps the footage there. A weakness in it is not an inconvenience, it is the failure of the one promise the product makes. This document says what is supported, how to report a flaw, and what Vyzio does and does not defend against today.

Supported versions

Vyzio is pre-1.0 and moves fast. Only the latest release receives fixes, and there is no backport to an earlier tag.

Version Supported
Latest release Yes
Anything older No, update first

Updating is docker compose pull && docker compose up -d, after setting the new number when .env pins one in VYZIO_VERSION; the data survives it.

Reporting a vulnerability

Report privately, through GitHub Security Advisories. That channel is private between you and the maintainer until a fix ships. Please do not open a public issue for a vulnerability, and do not disclose it elsewhere before a fix is available.

Useful in a report: the version, how Vyzio is deployed, the steps that reproduce the problem, and what an attacker gains from it. A proof of concept helps more than a description.

This is a single-maintainer project with no support commitment, so expect an acknowledgement within a week rather than within a day. You will be credited in the advisory unless you ask otherwise.

What Vyzio assumes

Vyzio is built for a home network you trust, on a machine you own. The whole system sits behind one entry point and one authentication boundary: the dashboard container is the only service published, the API is reachable only from the Docker network, and Frigate is bound to 127.0.0.1 and never routable from outside. Nothing is reachable before an owner password has been set.

Vyzio also assumes the machine is yours to give. Its API container carries the host's Docker socket, because writing a configuration is only half of applying it: something has to restart Frigate. A container holding that socket can do anything on the host, and mounting it read-only changes nothing there, since the restriction applies to the file and not to the commands sent through it. Whoever executes code inside vyzio-api therefore holds the machine.

That is a deliberate trade for a product meant to install itself without asking anyone to write YAML, and it is bounded on the side that matters: the container is not published, and the command it runs is read from the environment when the process starts, never from a request. No route can choose it.

On the home network, the interface is served over plain HTTP, and that is deliberate. There is no TLS, no certificate and no redirect, so the session cookie, the password and every preview image travel unencrypted, and anyone able to observe that network can read them. An installation is exactly as private as the network it sits on: its Wi-Fi encryption, and who is on it. Remote access is not delivered yet (issue #62): when it is, the interface is reached from outside only through the overlay network below, whose tunnel encrypts end to end. Why, and the certificate options set aside: ADR-16.

The threat model, surface by surface, lives in docs/SAD.md § Threat model and is not repeated here.

Known gaps, stated plainly

Camera accounts and messaging bot tokens are stored unencrypted in the database file, and the camera accounts again in the Frigate configuration Vyzio generates. Whoever gets a copy of either file, a backup for instance, can read them: protect both as you would the passwords themselves. Tracked in issue #247.

Out of scope

  • Exposing the interface directly to the internet. Port-forwarding the dashboard is not a supported deployment, and reports based on it will be closed. Remote access is not delivered yet (issue #62): when it is, it goes through a user-operated overlay network, where the hub is a peer and not a gateway (ADR-16).
  • An attacker who already has the machine. Vyzio does not defend against someone with a shell on, or physical access to, the host: they hold the database and the disk.
  • The reset window, while it is open. Clearing a forgotten password deliberately reopens the installation for 30 minutes to anyone on the local network. That is a bounded, documented trade-off, not a flaw. The README says so where the command is given.
  • Vulnerabilities in Frigate itself. Report those to the Frigate project. Report to us how Vyzio exposes one.

What runs on every change

The CI workflow runs on every pull request, on every push to main and daily. It builds both images through the Images workflow, which scans them with Trivy from high severity up, and which the release runs too, so the release pushes no image it has not scanned. CI also calls the Security workflow: CodeQL on C# and TypeScript, Semgrep static analysis, and a dependency audit on both the .NET and the frontend package trees. What depends on the change blocks a pull request: Semgrep fails on a finding its base did not have, and CodeQL on a new alert. What depends on advisories published since, the dependency audit and the image scan, only warns on a pull request and fails main and its daily run; the image scan also fails a release. Semgrep, CodeQL and Trivy report to code scanning. Dependabot proposes updates for every dependency ecosystem, base images and workflow actions included, as pull requests, except a runtime major or a Frigate minor or major, which an issue decides.

There aren't any published security advisories