Your own models, on your iPhone.
Run a model on your Mac or PC with Ollama, LM Studio, llama.cpp, or vLLM, and chat with it from your iPhone with Pelican.
A model that supports tool calling can do what any other does in Pelican: check your calendar, add reminders, search the web, and find places in Maps. Your messages go only to your own computer.
Set up in one field.
In Settings, add a custom provider and type your computer’s name or address. Pelican finds the server and fills in the rest. It tries the ports common servers use, works out which server answered, and lists its models:
- Ollama (11434)
- LM Studio (1234)
- llama.cpp (8080)
- vLLM (8000) and SGLang (30000)
- KoboldCpp (5001), text-generation-webui (5000), Jan (1337), and Exo (52415)
- LiteLLM (4000)
With Tailscale, your computer’s name alone is enough: Pelican tries its MagicDNS name over https, for a model you share with tailscale serve, so you can use it away from home too.
Any compatible server.
Pelican speaks the OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages APIs, so a proxy or a gateway works too. Choose models from the server’s list or add them by name, and set each one’s context window, output limit, reasoning, and image input. Plain http works for addresses on your local network. Anywhere else, use https.
Free.
Connecting your own models is free. Mix them with providers you have keys for and Apple’s model on Private Cloud Compute, and pick a different one for any chat.