Expose a Local LLM

Q: How do I share Ollama with my team?

Start Ollama with OLLAMA_HOST=0.0.0.0 , run localport http 11434 --token tok_YOUR_TOKEN --region eu , and give teammates the tunnel's address. Add header authentication first so only they can call it.

Summary

An HTTP tunnel gives the API of a model running on your machine a public HTTPS address. Teammates, coding tools and your other devices call it from anywhere while the model runs on your own hardware. A model API has no login of its own, so put header authentication or an IP allowlist in front of it.

How it works#

Ollama and LM Studio serve an HTTP API on a local port. The agent carries requests from the tunnel's address to that port and returns the model's response, streamed as it is produced.

You need a Localport account, the Localport agent, and Ollama or LM Studio running on your machine.

Ollama#

Ollama listens on port 11434. Create an HTTP tunnel in the dashboard, then run:

localport http 11434 --token tok_YOUR_TOKEN --region eu

Call the API on the tunnel's address:

curl https://my-application.eu.localport.dev/api/generate -d '{
  "model": "llama3.2",
  "prompt": "Explain TCP tunneling in one paragraph",
  "stream": false
}'

Ollama answers 403 until it listens on all interfaces

Ollama started with its default settings refuses a request addressed to a host name other than its own. Start it with OLLAMA_HOST=0.0.0.0 and it accepts requests sent to the tunnel's address. That setting also makes Ollama reachable from your local network.

LM Studio#

LM Studio's local server listens on port 1234 and speaks the OpenAI API format:

localport http 1234 --token tok_YOUR_TOKEN --region eu

curl https://my-application.eu.localport.dev/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "local-model",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Any tool that takes an OpenAI base URL can point at https://my-application.eu.localport.dev/v1.

Keep the endpoint private#

A tunnel is public by default, and a model API answers anyone who calls it. Add one of these before you share the address:

Header Authentication admits only requests that carry a header and value you set. Most API clients can send a custom header, so this suits a model endpoint.
IP Allowlists accept connections only from the addresses you name, such as your office.
Basic Auth asks for a username and password.

Point your tools at it#

Tool	Setting to change
Continue	The model's `apiBase`
Open WebUI	The Ollama API URL
Your own scripts	The base URL, in place of `http://localhost:11434`

Set a subdomain in the tunnel's settings so the address in your team's configuration never changes.

Keep the model loaded#

Ollama unloads a model after a period with no requests, and the next caller waits while it loads again. Start Ollama with OLLAMA_KEEP_ALIVE=-1 to keep the model in memory.

Frequently asked questions#

How do I share Ollama with my team?

Start Ollama with OLLAMA_HOST=0.0.0.0, run localport http 11434 --token tok_YOUR_TOKEN --region eu, and give teammates the tunnel's address. Add header authentication first so only they can call it.

Why does Ollama return 403 through the tunnel?

With its default settings Ollama refuses requests addressed to a host name other than its own. Start it with OLLAMA_HOST=0.0.0.0.

Do streamed responses work?

Yes. A response is passed on as the model produces it.

Can Localport read my prompts?

Localport does not log or store what you send. On an HTTP tunnel a request is decrypted in memory to route it and is never written to disk.

Should I use an HTTP tunnel or a TCP tunnel?

An HTTP tunnel. Ollama and LM Studio serve HTTP, and an HTTP tunnel gives you HTTPS with a trusted certificate and the per-request controls. A TCP tunnel carries the same traffic without them.

Does the address stay the same?

Yes. The address belongs to the tunnel and is the same after every restart of the agent.