Crate tt_retrieval 

tt-retrieval — RAG / context-compression engine.

See docs/superpowers/specs/2026-05-28-trackE-rag-context-compression-design.md.

Re-exports

pub use error::RetrievalError;

pub use store::RetrievalStore;

pub use substitute::substitute_in_messages;

pub use substitute::SubstitutionReport;

pub use substitute::DEFAULT_MIN_SIMILARITY;

pub use types::Chunk;

pub use types::Corpus;

pub use types::Document;

pub use types::RetrievableTag;

pub use types::RetrievalResult;

Modules

chunking

512-token chunks with 64-token overlap. Tokenizer: tiktoken cl100k_base.

embed

OpenAI text-embedding-3-small client. 1536-dim output.

error

search

Top-k cosine over a RetrievalStore.

store

substitute

Orchestrator: take a message body, parse retrievable tags, embed the rest, retrieve top-k above a similarity floor, splice the retrieved chunks into the tag spans.

tags

Parse <retrievable corpus="X" k="N">...</retrievable> tags from message text. Returns each tag’s corpus, k, and span in the text.

types

Data shapes shared across the crate.