Skip to content

v0.20.0 — Context-optimizer compression upgrades

Choose a tag to compare

View all tags
@doramirdor doramirdor released this 12 Jun 20:29
· 18 commits to main since this release
v0.20.0
fadd907
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Context-optimizer compression upgrades (#65), now documented and released.

Added

  • Pluggable optimizer backend — NADIRCLAW_OPTIMIZE_BACKEND=native (default) or headroom (opt-in via pip install nadirclaw[headroom], Apache-2.0). Headroom is fail-open: absent/error → transparent fallback to native, requests never fail.
  • Progressive (staged) compression — --optimize progressive escalates native_safe → native_aggressive → headroom_structural → headroom_ml and stops as soon as NADIRCLAW_OPTIMIZE_TARGET_TOKENS is met. Dependency-free and lossless by default. New compress_progressive() library entrypoint.
  • Columnar JSON-array packing (json_array_pack, aggressive mode) — lossless, reversible packing of homogeneous object arrays (~68% vs pretty JSON).
  • Native CCR (nadirclaw/ccr.py) — deterministic offload + nadir_retrieve fetch-back loop (library-only for now).

Fixed

  • Whitespace normalization corrupted unfenced source code — it flattened leading indentation of file-read tool outputs into invalid syntax; now preserves indentation in both safe and aggressive modes.

See the CHANGELOG and savings analysis.

Assets 2
Loading