guides
Practical write-ups on reaching your own models from anywhere, and honest comparisons against the tunnelling tools you'd otherwise reach for.
A complete walkthrough for pointing Cursor at your own Ollama install instead of a hosted model — Ollama for the models, Amallo for the endpoint, Cursor for the editor.
Aug 24, 2026
OLLAMA_HOST=0.0.0.0 answers requests from anyone who can reach the port — with no login, no rate limit, no log. Here's the actual threat model, and what a safe setup looks like.
Aug 24, 2026
These tools already support a custom OpenAI-compatible base URL for exactly this. The missing piece is getting your own Ollama reachable from wherever you're coding.
Aug 24, 2026
No public IP means port forwarding isn't just annoying — it's not on the table. Here's why CGNAT breaks the usual advice, and what actually works instead.
Aug 24, 2026
Off your home WiFi, your phone can't see your Ollama box at all. Here's how to reach it from cellular data without running a VPN client or paying for hosted inference.
Aug 24, 2026
Ollama answers on your LAN the moment you set OLLAMA_HOST — reaching it from outside is a different problem. Here's the actual fix, not another port-forwarding guide.
Aug 23, 2026
Cloudflare Tunnel is most people's first answer for reaching a machine from outside the LAN. Here's an honest comparison against 41tunnels for the specific case of a local LLM — Cloudflare wins some of this.
Aug 23, 2026