Skip to content

Security: senssei/prism-local

SECURITY.md

Security Policy

Reporting a vulnerability

Please report security issues privately through GitHub's private vulnerability reporting rather than a public issue. Include the version, how you run prism serve (host, flags), and steps to reproduce. This is a small alpha project, so expect a best-effort response, but reports are taken seriously.

Threat model

prism serve is meant to be a local, single-user inference server.

  • It binds to 127.0.0.1 by default, sends no CORS headers, rejects non-loopback Host headers, and limits request bodies to 10 MB.
  • Setting --host to a non-loopback address exposes the model, and your GPU time, to that network. Use --api-key and put a TLS-terminating reverse proxy in front of it; Prism does not speak TLS itself.
  • /health is unauthenticated by design and returns GPU model, memory and the active model name.
  • --cors-origin '*' lets any web page call the server from a browser; only use it together with --api-key.
  • The MCP server exposes read-only tools (generation, listing, status, benchmark). It does not read or write files.

Supported versions: the latest release only.

There aren't any published security advisories