Skip to content
Documentation menu

Share your model · about 5 minutes

Connect your computer.

Share the model running in LM Studio, Ollama or another local server. The connector page in Chrome finds it, checks it and makes it available to your account. No ports to open.

Open the connector

Needs the Chrome extension. Install it first if you haven't.

  1. Start your model server

    • LM Studio — load a model, then Developer → Start server (port 1234).
    • Ollama — runs after installation; download a model, for example ollama pull qwen3 (port 11434).
    • llama.cpp — llama-server -m model.gguf --port 8080.
    • Jan, vLLM, KoboldCpp, text-generation-webui, GPT4All — turn on the OpenAI-compatible API server.

    Keep the server on localhost. You don't need “Serve on local network”, router changes or a public IP.

  2. Choose the server and model

    Open the connector page. It checks the usual ports and lists what it finds. Pick your server, then a model — models that are already loaded come first.

    A list of local models with the loaded model marked

    Nothing found? Start the server and click Scan again, or enter its address under Advanced settings.

  3. Connect

    Click Connect (the first time, also confirm the privacy note). Ompera sends a few short test prompts to check chat, streaming, tools and JSON, then makes the model available to your account.

    The page says “Your model is live”.

  4. Test it and use it

    Click Send a test message, then Create an API key. Use the base URL, the key and the address my/local in your app, as in Use AI in your apps.

Keep it running

  • After Connect the tab shares your model until you click Stop sharing or close the tab. A sleeping computer, a network drop or an Ompera update interrupts it only for a moment: the tab says “Restoring the connection…” and comes back by itself.
  • Don't reload the tab while it shares. A reload takes your model offline for a few seconds and cuts off a request that is being answered; both lower your rating in Your sharing record on the Devices page. The tab asks before it reloads, and sharing comes back by itself afterwards. To pause, click Stop sharing: it is not counted as a break.
  • While the computer sleeps or the tab is closed, the model is offline; your apps then get a clear “model is offline” error.
  • Chrome can unload tabs that stay in the background to save memory. From version 0.4.3 the extension keeps the connector tab loaded while it shares.
  • If your model server asks for an API token, a reload asks you to enter it again: the token is kept only in the tab's memory.
  • One Chrome profile shares one model at a time. Opening the connector page again shows the live tab with Switch to that tab and Stop sharing; the extension's toolbar icon can also Stop all connections.
Advanced settings
  • Server address — any OpenAI-compatible server on this computer, ending in /v1 (for example http://127.0.0.1:1234/v1).
  • Server API token — only if your server requires one (LM Studio “Require authentication”). It stays in the tab's memory and is not your Ompera API key.
  • Model address — what your apps call, my/local by default. Keep it stable: when you switch models behind the same address, your apps keep working. Pool members use an address built from the model's name (HomeLab/qwen3-4b), so theirs changes with the model. Use a different address for a second model.
  • Context limit — set automatically from the model loaded in LM Studio. For other servers, use the context your model was loaded with.
My model shows “Not loaded”

LM Studio lists every downloaded model. Load the one you want (or turn on Just-in-Time loading in LM Studio), then connect. Selecting a model name alone does not load it.

Requests to this model pass through Ompera to your model server; both can read them. What this means.