Use a model on your own machine

Connect Ollama, vLLM, LM Studio or any OpenAI-compatible server that is not on the internet.

Reason calls your model from the cloud, so a Base URL like http://127.0.0.1:11434 points at Reason’s servers, not yours. Give the server a public https address, for example with a Cloudflare Tunnel:

OLLAMA_CONTEXT_LENGTH=32768 ollama serve
cloudflared tunnel --url http://localhost:11434 --http-host-header localhost:11434

In Settings → Models, choose Custom. Enter any API key and set the Base URL to the tunnel address plus /v1. Leave the other fields blank. Reason lists the server’s models when you save.

• For vLLM, LM Studio or a server behind an SSH tunnel, point cloudflared at its local port.
• Agent prompts run past 20,000 tokens. Raise the server’s context window, because Ollama’s default silently cuts off the rest.
• Ollama only answers requests addressed to localhost, which is why the tunnel needs --http-host-header.
• Anyone with the address can use the server. If that matters, put a proxy that checks a key in front of it and enter that key in Reason.
• A trycloudflare.com address changes whenever cloudflared restarts. To enter the new address, use Replace key on the Custom connection. A named tunnel keeps the same address.
• Small models struggle with tool calls. If one makes up a tool result, start a new session instead of following up.

FAQ

Can Reason reach a server on my private network or VPN?

No. Reason calls the server from the cloud, so it only reaches a public https URL. localhost, private addresses such as 10.x or 192.168.x, VPN and Tailscale addresses all fail. Put the server behind a tunnel that gives it a public address.

How do I connect a server I reach over Tailscale?

A 100.x or *.ts.net address only works inside your tailnet. On the server’s machine, run tailscale funnel 8080 with the server’s port, then enter the printed https://….ts.net address plus /v1 as the Base URL.