AI that sees your screen, listens to your conversations and tells you what to do

Moshi is a speech-text foundation model and full-duplex spoken dialogue framework. It uses Mimi, a state-of-the-art streaming neural audio codec.

Model-agnostic code memory MCP server. Budget-aware graph retrieval for AI agents. Sub-millisecond queries, token budgeting, deterministic results. Built on CodeGraphContext.

AI coding agent optimized for small LLMs. 87% benchmark with 4B-active model.

Clive — a friendly, local-first CLI for open-source LLMs via Ollama. Streaming chat, interactive coding sessions, safe file edits, and autonomous multi-file agent workflows, all on your own machine.

Local AI model manager

The Unified Model Registry for all your local AI apps.

LLM inference gateway with model management — search, download, and serve GGUF models via llama.cpp.

Next Generation Intelligent Browsing System

PyTorch building blocks for the OLMo ecosystem

Sonorita — installeurs téléchargeables (alpha). Radio musicale IA 100% locale. Code source : privé pour le moment.

Manipulate DNS records on various DNS providers in a standardized way.

Manipulate DNS records on various DNS providers in a standardized way.

Building the world's first true agent together with the community, using federated local reinforcement learning!

MCP server to provide Figma layout information to AI coding agents like Cursor

Open-source Figma MCP server with REST reads, a companion canvas-write plugin, design tokens, multi-framework codegen, a desktop app, and no Dev Mode seat or telemetry.

Run Qwen 3.8 MAX (2.4T MoE) on a 50GB-RAM consumer machine in pure Rust, experts streamed from disk. Tiny engine, immense model.

A full GUI experience on top of llama.cpp: all-knobs model tuning, one-click build/update from upstream, HuggingFace discovery with VRAM-fit ratings

Zero-config application builder that automatically analyzes and turns your code into an image

Distributed AI/LLM for the people. Share compute privately or publicly to power your agents and chat.

Run MoE models bigger than your RAM. A 284B on a 12 GB phone, CPU only, lossless, on stock llama.cpp

🔦 Le SLM souverain français Luciole-1B tourne 100 % sur le téléphone (hors-ligne) et pilote de vraies actions Android — application native de démonstration.

Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦

Universal LLM Deployment Engine with ML Compilation

Winamp 2 reimplemented for the browser