v0.4.3

What's Changed

• feat(cli): match evidence, relative scores, and context tips in human search output by @miguelarios in #16

Full Changelog: v0.4.2...v0.4.3

v0.4.2

What's Changed

• fix: gpuLayers "auto" — LlamaProvider was running 100% on CPU by @miguelarios in #13

Full Changelog: v0.4.1...v0.4.2

v0.4.1

What's Changed

• chore: Homebrew formula → published 0.4.0 tarball + real sha256 by @miguelarios in #10
• feat: live download progress bar + fix llama context-size crash on long chunks by @miguelarios in #12

Full Changelog: v0.4.0...v0.4.1

v0.4.0

What's Changed

• docs: TypeScript rewrite plan by @miguelarios in #8
• feat: TypeScript rewrite (node-llama-cpp + better-sqlite3 + sqlite-vec) by @miguelarios in #9
• fix: package.json repository field for npm provenance; version → 0.4.0 by @miguelarios in #11

Full Changelog: v0.3.0...v0.4.0

v0.3.0

feat: split ingestion into `qkb ingest` (structural) + `qkb embed` (v…

v0.2.3

feat: ingest shows current file in progress + graceful Ctrl-C (#6)

- on_progress gains the current file (relative posix path); the progress
  bar description updates in place to "Indexing <…/file.md>" (truncated to
  the useful tail) so you can see where it is without a scrolling list.
- Ctrl-C during ingest now prints a clean "Aborted. The index is safe —
  re-run `qkb ingest --full` to resume." and exits 130, instead of Click's
  bare "Aborted!". (An interrupted --full is already recoverable via the
  in-progress sentinel; nothing is corrupted.)

210 unit tests pass. Version 0.2.3.


Claude-Session: https://claude.ai/code/session_01TnBf6NQWckc7q1sUXnstG7

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

v0.2.2

feat: progress bar + clean skip reporting for ingest (no tracebacks) …

v0.2.1

feat: qkb status shows what model built the index + mismatch warning …

v0.2.0

docs: full first-run steps in Quickstart (config.toml, frontmatter op…

v0.1.0

feat: local in-process embedding provider (llama.cpp / GGUF) (#1)

Adds a `provider = "local"` embedding option running in-process via
llama.cpp on an auto-downloaded GGUF model, alongside the existing Ollama
provider. Ingest/search and the EmbeddingProvider protocol are unchanged.
Implemented via subagent-driven development (per-task review + whole-branch
review); README/CI/gitleaks setup included.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TnBf6NQWckc7q1sUXnstG7