Connect a Local Model: Ollama and LM Studio

If you already run models on your computer with Ollama or LM Studio, you can connect them to Owlfy and let the AI Agent think with a local model.

Owlfy doesn’t include a local model itself. Ollama or LM Studio downloads and runs the model, and Owlfy connects to it through their API.

Before you start

1. Install Ollama or LM Studio.
2. Download a model that fits your computer’s memory.
3. Start the local server:
   • Ollama usually runs in the background once installed;
   • in LM Studio, start the local server from the Developer tab.

If you want the AI Agent to work with files and use tools, pick a model that supports tool calling.

Add it to Owlfy

1. Open Settings → AI and turn on Third-party models.

2. Click Add third-party model and choose Ollama Local or LM Studio Local.

3. Check the API address; usually you don’t need to change it:

   Provider	Default address
   Ollama Local	http://127.0.0.1:11434/v1
   LM Studio Local	http://127.0.0.1:1234/v1

4. Click Refresh models and pick a model you’ve downloaded (or type its name).

5. Click Test and save and wait for the success message.

6. In the AI Agent, switch to your new local model in the model picker.

Start with a simple question, then try tasks that need tools.

Other model providers

Besides local models, Third-party models also supports:

• OpenAI, Anthropic, Google Gemini, Grok / xAI
• DeepSeek, Zhipu GLM, Alibaba Cloud Bailian, Volcengine Ark, Kimi / Moonshot
• OpenRouter, CC Switch Gateway
• Any OpenAI-compatible API (choose Third-party API (OpenAI-compatible))

Enter your API key and model, and you’re set. Owlfy doesn’t charge Credits for these models; usage is billed by the provider.

Results depend on the model

How well a local model works comes down to the model and your hardware:

• A model that chats well may still struggle with tools. For complex tasks, use a larger model that supports tool calling.
• To work with images, choose a model that accepts image input.
• Long documents need a model with a long enough context window.
• Bigger models are slower and need more memory.

Keep Ollama or LM Studio running while you use it.

Local model doesn’t mean fully offline

A local model handles the “thinking” on your computer. If a task searches the web, opens pages, uses online skills, or replies through Mobile Control, those steps still go online.

A local model also only affects the AI Agent. Speech-to-text is a separate setting: to keep voice recognition on your computer too, turn on Local Speech Recognition in Settings → Privacy.

Can’t connect?

• Is Ollama or LM Studio running?
• Is the address and port right? If you changed the port, use the address the app shows.
• Is the model downloaded, and does the name match?
• Chats fine but won’t take actions? Try a model that supports tool calling.

For server setup, see the Ollama docs and LM Studio docs.

AI Agent · Privacy