What

Reframes the README hero and docs (homepage, overview, FAQ) around LocalAI's composable architecture instead of language that reads as a monolith.

The README led with "36+ backends" and the docs described LocalAI as "Your All-in-One Complete AI Stack" / "a single binary that gives you everything you need" - phrasing that fuels a "heavy / does-too-much" impression. In reality LocalAI is a small core, and each backend is a dedicated gRPC service built around a best-in-class engine (llama.cpp, vLLM, whisper.cpp, stable-diffusion, MLX...), shipped as a separate OCI image and pulled only when a model needs it.

Changes

• README hero: leads with "a small core, not a bundle"; adds Composable by design and Open and extensible (bring your own model / backend) bullets; drops the kitchen-sink "36+ backends" lead.
• _index.md: tagline → "A small, composable AI stack: run any model locally and install only what you use."
• overview.md: distinguishes the small core from on-demand backends; explains the gRPC/OCI design as benefits (isolation, independent upgrade, run on another machine); adds bring-your-own model/backend + flexibility.
• faq.md: adds "Do I need to install all the backends?" and "Can I bring my own model or backend?"

Messaging principles applied

• Lead from strength - state the architecture as a positive; never name the accusation (no "bloated" anywhere, which would only plant it).
• Altitude - gRPC/OCI mechanics live on evaluator surfaces (overview/FAQ), framed as benefits; the hero stays jargon-free.
• Consistency - README and docs share one vocabulary and the same backend example set.

A composable-architecture diagram will be added in a follow-up commit once hosted.