Skip to content

Releases: 4thel00z/konan

v0.2.4

07 Jun 17:32
6e031cb

Choose a tag to compare

0.2.4 (2026-06-07)

Bug Fixes

  • drop dead code left by the hot-path change (7e0e426)

Performance Improvements

  • core: ICU4X sentence segmentation + lazy memchr unit scanners (276678d)
  • core: stream fixed-size units lazily (49ffe7c)

Documentation

  • refresh benchmark figures after the hot-path fixes (eadb77e)

v0.2.3

07 Jun 08:15
14ea1d4

Choose a tag to compare

0.2.3 (2026-06-07)

Documentation

v0.2.2

07 Jun 03:28
d2801c7

Choose a tag to compare

0.2.2 (2026-06-07)

Bug Fixes

  • include workspace-root LICENSE in sdist, skip-existing on publish (84a090a)

v0.2.1

07 Jun 03:19
539ebed

Choose a tag to compare

0.2.1 (2026-06-07)

Performance Improvements

  • direct char counting for non-ASCII text + wider benchmark field with plots (1c355a5)

v0.2.0

07 Jun 02:54
d06148c

Choose a tag to compare

0.2.0 (2026-06-07)

Features

  • async-openai embedder backend, richer configs, config-revealing reprs (6caf2c2)
  • benchmarks — bench.py, criterion benches, README results (5ed3bfa)
  • core: Chunk, KonanError, span/offset utilities (3f40368)
  • core: Chunker port and rayon chunk_many (f6a62d1)
  • core: Embedder port and OpenAI-compatible adapter (7a7d7c2)
  • core: FixedSizeChunker with overlap and sentence awareness (fe36188)
  • core: MarkdownChunker with breadcrumbs and atomic code fences (c4bd62b)
  • core: NaiveChunker (word-based) (99f2738)
  • core: RecursiveChunker with separator hierarchy (3a1cbb1)
  • core: SemanticChunker over the Embedder port (22621f8)
  • core: SentenceChunker (unicode segmentation) (bf57195)
  • core: TokenChunker (tiktoken cl100k/o200k) (d24ef27)
  • py: bind six chunkers with sync, parallel and async methods (f115929)
  • py: package surface with typed stubs and py.typed (b5fb787)
  • py: SemanticChunker, OpenAIEmbedder, Python-callable embedder port (a8324ba)
  • scaffold konan workspace (konan-core + konan-py, maturin) (8e2e74f)

Bug Fixes

  • absolute logo URL so it renders on PyPI (618e185)
  • clippy is_multiple_of lint in benches (CI stable toolchain) (1e7bb44)
  • core: MarkdownChunker setext heading breadcrumbs (a242ce4)
  • harden semantic chunking edge cases from review backlog (9e8e673)

Performance Improvements

  • bpe-openai tokenizer + ASCII OffsetMap fast path (e877ea2)

Documentation

  • add async-openai adapter + repr polish design spec (f396cbe)
  • add konan chunkers design spec (dc9b24a)
  • add konan chunkers implementation plan (0384847)
  • modern README with logo, strategy table and examples (d3106e3)