# ToshLLM

> ToshLLM is an open-source native macOS app that runs local AI models on Intel Macs with AMD GPUs. The website documents the app and publishes signed community benchmarks. English is the primary language.

ToshLLM is not a cloud model provider. Chat, model execution and media generation run on the user's Mac. Benchmark submissions are optional and must be initiated in the app. A signed community submission is not the same as a verified ToshLLM Lab result. Performance depends on model, quantization, GPU, memory, macOS version and runtime settings; quote measurements with their configuration and source.

The complete Spanish site starts at [ToshLLM en español](https://toshllm.com/es/). Spanish-language documentation and benchmark indexes have their own URLs and `llms.txt` guides.

## Product and documentation

- [Product overview](https://toshllm.com/): Supported workflows, platform requirements and real application screenshots.
- [Documentation index](https://toshllm.com/docs): Starting point for the current app guides.
- [Getting started](https://toshllm.com/docs/getting-started): Installation, architecture compatibility and first launch.
- [Models and performance](https://toshllm.com/docs/models-and-performance): Model formats, quantization, memory and GPU settings.
- [Chat and projects](https://toshllm.com/docs/chat-and-projects): Conversations, attachments and local workflows.
- [Image and video](https://toshllm.com/docs/image-and-video): Local generation workflows and their separate runtimes.
- [Local API](https://toshllm.com/docs/local-api): App-hosted OpenAI-compatible and Anthropic-compatible endpoints.
- [Integrations](https://toshllm.com/docs/integrations): Editors, coding assistants and SDK connections.
- [Agents and MCP](https://toshllm.com/docs/agents-and-mcp): Tools, permissions and MCP support.
- [Privacy and security](https://toshllm.com/docs/privacy-and-security): Local data boundaries and optional sharing.
- [Troubleshooting](https://toshllm.com/docs/troubleshooting): Common compatibility and runtime issues.

## Benchmarks

- [Benchmark index](https://toshllm.com/benchmarks): Filterable public results and grouped comparison pages.
- [Benchmark guide](https://toshllm.com/docs/benchmarks): Measurement procedure, interpretation and submission provenance.
- [Benchmark submission](https://toshllm.com/contribute): How the app reviews, signs and shares a run.
- [Public benchmark JSON API](https://toshllm.com/api/v1/benchmarks?limit=100): Machine-readable public results. Check pagination and response fields instead of assuming a fixed dataset.

Individual benchmark runs contain evidence and are linked from the public index and grouped pages. Treat their measurements as observations, not universal model speed claims. Admin and private API routes are not public documentation.

## Community and contact

- [Source and license](https://github.com/engeldlgado/toshllm): ToshLLM's public repository.
- [Latest macOS release](https://github.com/engeldlgado/toshllm/releases/latest): Official download location.
- [Contact](https://toshllm.com/contact): Personalized help and company inquiries.
- [Contacto en español](https://toshllm.com/es/contacto): Formulario de contacto en español.
- [Support development](https://toshllm.com/support): Sponsorship and donation options.

## Optional

- [Sitemap](https://toshllm.com/sitemap.xml): Indexable page inventory.
- [Robots policy](https://toshllm.com/robots.txt): Public crawl access and restricted routes.