Developer Guide

How mistral.rs works inside:

• Architecture - how the server, engine, and pipeline layers fit together.
• MoE expert backends - how Mixture-of-Experts (MoE) experts are dispatched and which kernel runs.
• Multimodal pipeline - how image, video, and audio inputs reach the model.
• Session memory - how agentic conversation state is stored and matched.
• Build from source - local build setup and feature flags.