Skip to content

semtree v0.4.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 22 Jul 07:06
· 10 commits to main since this release
v0.4.0
c9fe70b

Breaking changes

  • Embedder now requires dimension() and model_id(), and gains max_batch_size() and a derived fingerprint(). Custom embedders must add the two required methods.
  • VectorStore now requires metric(); the crate adds a Metric enum.
  • FileManifest::new takes an embedder and a store fingerprint, and the manifest is versioned. Indexes built with 0.3 are treated as incompatible and rebuilt on the next run.
  • Chunk IDs are now blake3(path + span) instead of a content hash, so existing indexes rebuild.
  • semtree-parse drops the public Extractor trait; extraction is query-driven.

New features

Languages (semtree-parse)

  • 20 languages, up from 5: adds Java, C, C++, C#, Ruby, PHP, TSX, Kotlin, Scala, Swift, OCaml, Solidity, Lua, Zig and Emacs Lisp.
  • Extraction is driven by tree-sitter query files (one .scm per language), so adding a language needs no Rust. Queries are compiled once per language and validated by a build-time test.

Umbrella crate (semtree)

  • New crate re-exporting the default stack. semtree::default_backends() builds a fastembed embedder and a usearch store sized to it, and semtree::prelude::* brings in the indexing and search types.

Embedder (semtree-embed)

  • dimension(), model_id() and a derived fingerprint() let an index reject vectors from a different model.
  • max_batch_size() bounds embed() batch sizes so a large file no longer risks memory blow-up or an oversized remote request.

Store (semtree-store)

  • Metric enum and VectorStore::metric(); usearch and qdrant are metric-configurable and refuse to reopen an index under a different metric.

Incremental indexing (semtree-rag)

  • The manifest is versioned (schema and chunker version, embedder and store fingerprints) and rebuilds instead of mixing incompatible vectors.

Fixes

  • Stable, location-scoped chunk IDs fix cross-file collisions and clean deletion of a renamed file's chunks.
  • The CLI sizes the vector store to the embedder's dimension (OpenAI/Ollama indexes were mis-sized at 384).
  • index_dir defaults to .semtree when there is no config file (it was writing into the current directory).
  • No phantom "incompatible index" warning on a first index.
Assets 10
Loading