finds.dev← search

// the find

nobodywho-ooo/nobodywho

★ 991 · Rust · EUPL-1.2 · updated Jun 2026

NobodyWho is an inference engine that lets you run LLMs locally and efficiently on any device.

NobodyWho is a Rust-core inference engine that wraps llama.cpp and exposes it through idiomatic bindings for Kotlin, Swift, Python, Flutter, React Native, and Godot. The primary audience is mobile and game developers who want to ship on-device LLM features without writing C++ or managing llama.cpp directly. It handles the FFI plumbing so you can call `chat.ask()` from Dart or GDScript.

The Godot plugin is genuinely useful — game devs have had almost no good options for local LLM integration, and shipping this through the AssetLib with tool calling and vision support is real work. The grammar-based tool calling that auto-generates structured output from function signatures means you're not hand-rolling JSON schemas. Vulkan and Metal GPU acceleration out of the box covers the full desktop matrix without per-platform configuration. The `hf://` model URL shorthand that downloads directly from Hugging Face is a small thing that saves a lot of friction for new users.

Six language targets at once is a credibility problem — maintaining parity across Kotlin, Swift, Python, Flutter, React Native, and Godot means every feature ships six times or ships inconsistently, and you can already see this in the versioned docs being out of sync. Web/WASM support is tracked in issue #111 with no clear timeline, which is a hard blocker for React Native Web and Flutter Web use cases that people will assume work. The underlying llama.cpp dependency means you're inheriting its memory management and threading model through an FFI layer, and the evals directory full of CSV files suggests quality of output varies a lot by model and quantization — this is documented but not surfaced clearly enough at the decision point where someone picks a model. No mention of how context shifting behaves at the boundary: the README claims 'full conversation memory without any message length limits' but the actual tradeoff (what gets evicted, and when) isn't documented.

View on GitHub → Homepage ↗

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →