Repository navigation
[Feature]: Add configurable bind host (--host) for local network access #63
Description
Area
Local server (tuff serve or the Server screen)
Problem or use case
Summary
Currently, the server binds exclusively to 127.0.0.1. When running TUFF / the inference server on a dedicated host (e.g., an Apple Silicon Mac Mini) and accessing it from client frontends hosted on another machine within the local network (such as OpenWebUI running on a Docker host), the server is unreachable without setting up workarounds like SSH port forwarding or reverse proxies.
Please consider adding a --host <address> CLI flag (defaulting to 127.0.0.1) so users can explicitly bind to all network interfaces (0.0.0.0) or a specific LAN IP address.
- Hosting the server on a headless or desktop Mac on the local network.
- Querying the OpenAI-compatible API from external clients/frontends (e.g., OpenWebUI, LibreChat, or automated scripts) on other local devices.
- Avoiding the need for persistent SSH tunnels (
ssh -L) or third-party reverse proxies when operating inside a trusted private LAN.
Suggested change
Upstream turbo-fieldfare already has an open Pull Request implementing this exact functionality cleanly across the argument parser, the HTTP server bootstrap, and tests:
Model details (for new model support)
No response
Memory and hardware
No response
Alternatives or evidence
No response
Would you like to work on this?
None
Activity
rexmhall09 commentedon Nov 8, 2026
Great Idea! I'll look into it and try to get an update out ASAP :D
TUFF 10.2.1: serve other devices on the network with an API key
rexmhall09 commentedon Nov 8, 2026
This is in 10.2.1. Turn on Allow other devices on your network on the Server screen, or run
tuff serve --host 0.0.0.0. Other devices need the API key the Server screen shows (Authorization: Bearer <key>), so Open WebUI just needs the network URL and that key. Setup details: https://github.com/tuff-app/TUFF/blob/main/docs/LOCAL_SERVER.md#other-devices-on-your-network