rust-llm-stack

Five small, focused Rust crates for the LLM / RAG / agent niche. Each one solves a single problem well, ships independently on crates.io, and has zero weird transitive deps.

Crate	Purpose	Lines
embedrank	Cosine / dot / L2 distance for f32 embeddings + heap top-k	~200
promptbudget	Token-budget-aware text truncation, BYO tokenizer	~210
stopstream	Streaming-safe stop-sequence detector with partial-match handling	~210
citecite	Citation-marker [1] [2] injector + parser for RAG outputs	~230
ragmetric	IR metrics: recall@k, hit@k, MRR, NDCG@k	~250

Each crate is its own publishable unit. Pick the one you need:

cargo add embedrank promptbudget stopstream citecite ragmetric

Why a single workspace?

These crates share a problem space (one piece of an LLM/RAG pipeline each), an author, and dual MIT / Apache-2.0 licensing. Putting them in one repo gives you shared CI, one README to find them all, and the option to depend on multiple without juggling git remotes.

Status

v0.1.0 alpha. APIs may change in 0.x. Each crate has its own tests and docs.

License

MIT OR Apache-2.0.