Help→Local Model

Connect your own AI model (Ollama / LM Studio)

Last updated: 17 July 2026 · Applies to Lampira for iPhone, iPad and Mac · Читать по-русски

Before you start

Run a free AI model on your own computer and let Lampira use it over your Wi-Fi. It's private — your notes travel from your device straight to your computer and nowhere else — with no subscription and no API keys. You'll need:

💡

Have an iPhone 15 Pro or newer, or an M1+ iPad or Mac? Apple Intelligence needs no computer at all — this guide is for everyone else, and for anyone who prefers to pick their own model.

A.The Ollama way

  1. Install Ollama

    Download it from ollama.com and install it on your computer.

  2. Download a model

    Open Terminal and pull a model — llama3.2 is a good small starting point, but any model from the Ollama library works:

    ollama pull llama3.2
  3. Start the server so your iPhone can reach it 
    OLLAMA_HOST=0.0.0.0 ollama serve
    ⚠️

    The OLLAMA_HOST=0.0.0.0 part is what makes Ollama reachable from your iPhone or iPad — without it, Ollama listens only to the computer itself and Lampira's connection test will find nothing.

  4. Find your computer's name

    Open System Settings → General → Sharing — the local hostname is shown at the bottom, something like my-mac.local.

  5. Connect Lampira

    In Lampira, open Settings → Intelligence → Local Model, choose the Ollama chip, enter my-mac.local (your computer's name) and tap Test Connection.

  6. Pick a model — done

    After a successful test, Lampira lists the models your server offers live. Pick one, keep or edit the suggested name (for example “Ollama on my-mac”), and you're set. From then on Lampira shows only that friendly name — never the address.

    💡

    Nothing is saved until a connection test succeeds — you can't end up with a half-working setup.

B.The LM Studio way

  1. Install LM Studio and get a model

    Download it from lmstudio.ai, then use its built-in browser to download a model.

  2. Start the server

    Open the Developer tab and click Start Server.

    ⚠️

    Turn Serve on Local Network ON — it's off by default, and while it's off your iPhone can't see the server at all.

  3. Connect Lampira

    In Lampira, open Settings → Intelligence → Local Model, choose the LM Studio chip, enter your computer's name (from System Settings → General → Sharing), tap Test Connection, pick a model — done.

C.Something else entirely

Any OpenAI-compatible server works. Choose the Custom chip and enter the full base URL — for example http://localhost:1234/v1 — plus an API Key if your server requires one. The key is only needed for protected servers and is stored in your device Keychain, nowhere else.

Troubleshooting

The messages below are exactly what Lampira shows under Test Connection.

Lampira saysWhat to try
“Nothing answered at this address. Check that the server is running and both devices are on the same Wi-Fi.”
  • The server isn't running — start Ollama with OLLAMA_HOST=0.0.0.0 ollama serve, or press Start Server in LM Studio.
  • Ollama is running without OLLAMA_HOST=0.0.0.0, or LM Studio's Serve on Local Network is off — the server exists but only answers the computer itself.
  • Phone and computer are on different Wi-Fi networks (a guest network counts as different).
  • The computer's firewall is blocking incoming connections — on macOS check System Settings → Network → Firewall and allow the app.
“The server asked for an API key. Add it under Custom.” Your server is protected. Switch to the Custom chip, enter the full URL and the key — it's stored in your device Keychain.
“That doesn't look like a valid address.” Check for typos. Pasting localhost:11434-style addresses works — Lampira honors an explicit port. Just remember that on an iPhone or iPad, localhost means the phone itself: use your computer's name (my-mac.local) instead.
“The server answered, but not like an AI server (error …).” Something is listening at that address, but it isn't an OpenAI-compatible endpoint. Check the port (Ollama 11434, LM Studio 1234) and, for Custom, that the URL ends in /v1.
Test succeeds, but the model list is empty The server has no models yet — run ollama pull llama3.2, or download a model inside LM Studio, then test again.