Skip to content

v0.1.0

Choose a tag to compare

View all tags
@usize usize released this 28 Jun 06:36
· 3 commits to main since this release
v0.1.0
990881a
This commit was signed with the committer’s verified signature.
usize Morgan Foster
SSH Key Fingerprint: 8VPWkk6TUubYORoHUcFqwHu1OaLXomgAx0Zi9nHQyt4
Verified
Learn about vigilant mode.

One 45 MB binary. No dependencies. No runtime. No network. No GPU.

vecfile is a self-contained hybrid search engine: SQLite + FTS5 + sqlite-vec + llama.cpp with bge-small-en-v1.5 baked in. Copy the binary anywhere and it works.

What's in it:

  • Namespaced corpus with frozen config per namespace
  • Inline chunking + embedding at ingest (no separate index step)
  • Hybrid RRF query: FTS5 BM25 + vector KNN fused with Reciprocal Rank Fusion
  • Semantic-only and lexical-only query modes
  • SHA256 content dedup (re-running add on unchanged files is instant)
  • Wildcard/multi-file add: vecfile add --db x --ns y *.txt
  • Chunk-level retrieval with -C N context expansion (designed for LLM consumption)
  • Full CRUD: add, delete, get by id/tag/chunk
  • Claude Code memory skill + one-line installer

Install as Claude Code memory:

curl -sL https://raw.githubusercontent.com/usize/vecfile/main/install.sh | sh

Verify:

shasum -a 256 vecfile
e3705a59db21679f6d75a4da6ada01297e2307988fce3c7a452a07e8327a98dd  vecfile
Assets 4
Loading