ios-local-llm

# OnDevice LLM

> OnDevice LLM is a Swift/SwiftUI local-first AI workbench for
> iPhone and Apple silicon Macs. Run language, vision, speech, and
> image-generation models on-device, with optional authenticated local APIs for
> AI clients and agents.

Product name: OnDevice LLM
Product line: OnDevice
Repository: https://github.com/Mesutcydev/ios-local-llm
Website: https://mesutcydev.github.io/ios-local-llm/
License: MIT for original project code and documentation
Swift module and Xcode scheme: IOSLocalLLM

## OnDevice product line

- OnDevice LLM — main open-source workbench (this repository)
- OnDevice Local API Server — coming soon
- OnDevice CoreAI Local API Server — beta

## Primary documentation

- [Project overview](https://github.com/Mesutcydev/ios-local-llm/blob/main/README.md)
- [Coding-agent instructions](https://github.com/Mesutcydev/ios-local-llm/blob/main/AGENTS.md)
- [Architecture](https://github.com/Mesutcydev/ios-local-llm/blob/main/ARCHITECTURE.md)
- [Reusable component catalog](https://github.com/Mesutcydev/ios-local-llm/blob/main/Docs/REUSABLE_COMPONENTS.md)
- [Coding-agent implementation handoff](https://github.com/Mesutcydev/ios-local-llm/blob/main/Docs/CODING_AGENT_IMPLEMENTATION.md)
- [Local runtime foundation notes](https://github.com/Mesutcydev/ios-local-llm/blob/main/IOSLocalLLM/LocalAIRuntimeFoundation/README.md)
- [Development setup](https://github.com/Mesutcydev/ios-local-llm/blob/main/SETUP_INSTRUCTIONS.md)
- [Agent and local API integration](https://github.com/Mesutcydev/ios-local-llm/blob/main/Docs/AGENT_INTEGRATION.md)
- [Contributing](https://github.com/Mesutcydev/ios-local-llm/blob/main/CONTRIBUTING.md)
- [Security policy](https://github.com/Mesutcydev/ios-local-llm/blob/main/SECURITY.md)
- [Third-party notices](https://github.com/Mesutcydev/ios-local-llm/blob/main/THIRD_PARTY_NOTICES.md)
- [Structured metadata](https://github.com/Mesutcydev/ios-local-llm/blob/main/codemeta.json)

## Capabilities

- On-device LLM chat and code assistance with MLX and llama.cpp
- Camera, screenshot, OCR, and vision-language-model analysis
- Local transcription, voice activity detection, and text-to-speech
- On-device image generation
- Hugging Face model discovery, admission checks, downloads, and lifecycle
- Thermal-aware generation and coordinated model residency
- OpenAI-, Anthropic-, and Ollama-compatible local HTTP routes
- Structured tool calling and a pairing-authenticated iPhone-to-Mac agent link

## Important boundaries

- Model weights and generated native frameworks are not included in Git.
- Downloaded models have independent licenses.
- Local API access is opt-in, bearer-authenticated HTTP for trusted networks.
- The project does not silently send local prompts or captures to a cloud
  inference service.
- Simulator results do not prove physical-device thermal or Jetsam behavior.