Run a private AI server directly on your iPhone or iPad

Running AI locally on your phone or tablet is now practical for many workflows.
With ai.local, you can run models on-device and expose local endpoints compatible with both OpenAI-style (/v1) and Ollama-style (/api) clients.

This guide shows the fastest setup.

1. Check device requirements

For a good experience, use:

• iOS 18.2+ or iPadOS 18.2+
• Newer high-performance devices (for example iPhone 15 Pro, iPhone 16 series, iPad Pro M1/M2)

2. Install and prepare ai.local

1. Install ai.local on the App Store.
2. Open the app and choose a model that matches your device performance target.
3. Start the server in the app.

The server is exposed on your local network with:

• Default port: 11434
• Base URL format: http://<iphone-ip-address>:11434
• LAN binding: 0.0.0.0

You can use either the device IP address or Bonjour host shown by the app.

3. Pick from famous LLM families

Popular model families to look for in ai.local:

• Llama 3.2 (for example 1B/3B Instruct)
• Qwen 2.5 (for example 1.5B/3B Instruct)
• Mistral 7B Instruct
• Gemma 2 (2B/9B)
• Phi-3.5 Mini
• DeepSeek distilled models (small variants)

On iPhone/iPad, smaller quantized models usually feel best for latency and memory.
If you are unsure what is installed, query the model list first:

curl http://<iphone-ip-address>:11434/v1/models
curl http://<iphone-ip-address>:11434/api/tags

4. Verify the server is up

From another device on the same network:

curl -I http://<iphone-ip-address>:11434/
curl http://<iphone-ip-address>:11434/status

Expected status response:

{
  "status": "Running",
  "message": "Server is currently running."
}

5. Use OpenAI-style endpoints (/v1)

ai.local supports OpenAI-compatible routes like:

• GET /v1/models
• POST /v1/chat/completions
• POST /v1/completions
• POST /v1/audio/transcriptions
• POST /v1/audio/speech

Example chat request:

curl http://<iphone-ip-address>:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer local-ai" \
  -d '{
    "model": "Llama-3.2-1B-Instruct-4bit",
    "messages": [{"role":"user","content":"Give me 3 tips to write cleaner Swift."}]
  }'

Notes:

• Authentication is not enforced by the server.
• If your client requires an API key, any non-empty value works.
• For /v1 model names, do not include the mlx-community/ prefix.

6. Use Ollama-style endpoints (/api)

ai.local also supports Ollama-compatible routes like:

• GET /api/tags
• POST /api/chat
• POST /api/generate
• POST /api/show

Example Ollama-style chat request:

curl http://<iphone-ip-address>:11434/api/chat \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Llama-3.2-1B-Instruct-4bit",
    "messages": [{"role":"user","content":"Write a short release note."}],
    "stream": false
  }'

Resources

• ai.local website
• ai.local API docs
• ai.local on the App Store