This is a personal, single-user local inference setup — shell scripts and configuration, not a hosted service. Security issues here are almost always about the documented configuration choices (network exposure, container privileges, API keys), not exploitable code.

Vulnerabilities in llama.cpp, Open WebUI, Open Terminal, SearXNG, or MeTube themselves belong upstream, not here — this repo only owns the launch scripts and flags wrapping them.

Email onkarbadve@gmail.com with what you found, which script/doc it's in, and why it's a security issue rather than a documented trade-off (see below). This is maintained solo in spare time, so response times vary.

• Open Terminal, SearXNG, and MeTube bind 127.0.0.1 only, even over Tailscale — Open Terminal is shell execution, SearXNG has no auth of its own, and MeTube can write arbitrary files; none are worth exposing beyond localhost.
• Open WebUI binds 0.0.0.0:3000 as a side effect of --network host — mitigated by keeping WEBUI_AUTH on and relying on Tailscale as the only realistic path to that port from outside the LAN.
• API keys stored as plaintext files (e.g. ~/.config/open-terminal/api-key, chmod 600) — fine for a single-user local machine, not a shared-hosting posture.

If one of these is actually unsafe even under that threat model (a single trusted user, Tailscale-gated), that's still a valid report — explain the scenario where it fails.