

Unknown · 5× · vu le 03 sept. 2026
3
Momentum
Exo (also "EXO" or formerly "exo labs") is an Open Source software for distributed LLM Inference that automatically connects multiple Apple Silicon devices (and increasingly Linux computers) into a shared Inference cluster. Models are distributed layer by layer across available devices in the network, enabling even very large models to run that wouldn't fit in memory on a single device. The project is developed by Exo Technologies Ltd (exo labs) and is available on GitHub as Apache-2.0-licensed code. It offers compatible APIs to OpenAI, Claude, and Ollama as well as an integrated dashboard for cluster management.
Historique du momentum
18.08.16.11.
Fonctionnalités
| Throughput/Latency | Example: DeepSeek R1 (8-bit) at ~20 tok/sec on 2x M2 Ultra 512GB via Thunderbolt 5 |
| License | Apache License 2.0 (Copyright 2025 Exo Technologies Ltd) |
| Platform | macOS (Apple Silicon, primary platform via MLX), Linux (currently CPU-only, GPU support in development) |
| Price | Free, open source (no commercial pricing listed on official site) |
| Protocol Compatibility | OpenAI Chat Completions API, Claude Messages API, OpenAI Responses API, Ollama API |
| Release Date | Initial release June 2024; EXO 1.0 (Apache-2.0) released December 19, 2025 |
| Supported Models/Providers | LLaMA, Mistral, LlaVA, Qwen, DeepSeek, etc.; custom models loadable from HuggingFace Hub |