Remote_Access_Tunneling

Remote Access & Tunneling

How to watch (and optionally control) a running Orrin from another device. Treat this as an advanced configuration: the backend is not security-hardened, so never expose control endpoints publicly without the tokens below.

The one-tunnel path

./expose_orrin.command

This opens a single public tunnel to the Vite dev server, which proxies both /ws (telemetry WebSocket) and /api (REST) through to the backend on :8800. The frontend derives both URLs from the page origin, so one tunnel carries everything. The resulting URL is written to tunnel_url.txt; open <url>/brain on the remote device. tailscale_funnel.command does the same over Tailscale Funnel for a stable, authenticated alternative.

For plain LAN/Tailscale viewing without a tunnel, bind the backend beyond loopback (ORRIN_BACKEND_HOST=0.0.0.0, pin ORRIN_BACKEND_PORT) and point the UI at it with VITE_TELEMETRY_HOST=<ip>:8800.

Security model

Three independent tokens (all zero-config on localhost, enforced by backend/server/auth.py):

• ORRIN_READ_TOKEN / VITE_READ_TOKEN — require a token to read telemetry (REST and the WebSocket handshake) from non-loopback clients.
• ORRIN_CONTROL_TOKEN / VITE_CONTROL_TOKEN — require a token on /api/control/* (the Stop button and any steering). Set this before opening a tunnel, so a viewer can't stop or steer the agent.
• ORRIN_INGEST_TOKEN — require a token on POST /ingest, so a remote-exposed backend only accepts telemetry from the real cognitive loop.

If the UI is served from a tunnel with its own public origin, allowlist it with ORRIN_EXTRA_ORIGINS=https://... (comma-separated); the local Vite origin and the backend host are trusted automatically.

Remember: the tunnel URL itself is effectively the read secret — anyone who has it can watch. Stop the tunnel when done.

Code pointers

• expose_orrin.command, tailscale_funnel.command, tunnel_url.txt
• backend/server/auth.py — token enforcement
• docs/CONFIGURATION.md — the full remote-access reference