vexor 0.9.1

A vector-powered CLI for semantic search over files.

Vexor

---

Vexor is a vector-powered CLI for semantic file search. It supports configurable remote embedding models and ranks results by cosine similarity.

Usage

Vexor is designed to work seamlessly with both humans and AI coding assistants through the terminal, enabling semantic file search in autonomous agent workflows.

When you remember what a file does but forget its name or location, Vexor's semantic search finds it instantly—no grep patterns or directory traversal needed.

Skill

This repo includes a skill that guides an AI agent to use Vexor effectively: plugins/vexor/skills/vexor-cli.

Install it into Claude Code / Codex skill folders:

vexor install --skills claude
vexor install --skills codex

Install

Download from releases without python, or with:

pip install vexor # or use pipx, uv

The CLI entry point is vexor.

Configure

Set the API key once and reuse it everywhere:

vexor config --set-api-key "YOUR_KEY"

Optional defaults:

vexor config --set-model gemini-embedding-001
vexor config --set-batch-size 0   # 0 = single request
vexor config --set-provider gemini
vexor config --set-base-url https://proxy.example.com  # optional proxy support for local LM Studio and similar tools; use --clear-base-url to reset

Provider defaults to gemini, so you only need to override it when switching to other backends (e.g., openai). Base URLs are optional and let you route requests through a custom proxy; run vexor config --clear-base-url to return to the official endpoint.

Environment/API keys can be supplied via vexor config --set-api-key, VEXOR_API_KEY, or provider-specific variables (GOOGLE_GENAI_API_KEY, OPENAI_API_KEY). Example OpenAI setup:

vexor config --set-provider openai
vexor config --set-model text-embedding-3-small
export OPENAI_API_KEY="sk-..."   # or use vexor config --set-api-key

Configuration is stored in ~/.vexor/config.json.

Inspect or reset every cached index:

vexor config --show-index-all
vexor config --clear-index-all

Workflow

1. Index the project root (includes every subdirectory):

   vexor index --path ~/projects/demo --mode name --include-hidden
   

2. Search from anywhere, pointing to the same path:

   vexor search "api client config" --path ~/projects/demo --mode name --top 5
   

   For script/agent-friendly output, add --format porcelain (TSV) or --format porcelain-z (NUL-delimited), default format rich (table). Porcelain formats emit: rank, similarity, path, chunk_index, start_line, end_line, preview (line fields are - when unavailable). Output example:

   Vexor semantic file search results
   ──────────────────────────────────
   #   Similarity   File path                      Lines   Preview
   1   0.923        ./src/config_loader.py        -       config loader entrypoint
   2   0.871        ./src/utils/config_parse.py   -       parse config helpers
   3   0.809        ./tests/test_config_loader.py -       tests for config loader
   

Tips:

• Keep one index per project root; subdirectories need separate indexes only if you explicitly run vexor index on them.
• Toggle --no-recursive (or -n) on both index and search when you only care about the current directory; recursive and non-recursive caches are stored separately.
• Hidden files are included only if both index and search use --include-hidden (or -i).
• By default, Vexor respects .gitignore (including nested .gitignore files and .git/info/exclude) while scanning. Use --no-respect-gitignore on both index and search to include ignored files.
• Use --ext/-e (repeatable) on both index and search to limit indexing and search results to specific extensions, e.g. --ext .py --ext .md.
• Re-running vexor index only re-embeds files whose names/contents changed (or were added/removed); if more than half the files differ, it automatically falls back to a full rebuild for consistency.
• Specify the indexing mode with --mode:
  • auto: smart default routing (Python → code, Markdown → outline, small files → full, large files → head/name).
  • name: embed only the file name (fastest, zero content reads).
  • head: grab the first snippet of supported text/code/PDF/DOCX/PPTX files for lightweight semantic context.
  • brief: summarize PRDs/high-frequency keywords (English/Chinese) in requirements documents enable quick location of key requirements.
  • full: chunk the entire contents of supported text/code/PDF/DOCX/PPTX files into windows so long documents stay searchable end-to-end.
  • code: chunk Python .py files by module globals + class/function/method boundaries (AST-aware); other files fall back to full.
  • outline: chunk Markdown files by heading outline and embed heading breadcrumbs + a snippet from each section; other files fall back to full.
• Switch embedding providers (Gemini by default, OpenAI format supported) via vexor config --set-provider PROVIDER and pick a matching embedding model.

Commands

Command	Description
vexor index --path PATH [--mode MODE] [--include-hidden] [--no-recursive] [--no-respect-gitignore] [--ext EXT ...] [--clear/--show]	Scans PATH (recursively by default), respects .gitignore by default, embeds content according to MODE (defaults to auto), and writes a cache under ~/.vexor.
vexor search QUERY --path PATH [--mode MODE] [--top K] [--include-hidden] [--no-recursive] [--no-respect-gitignore] [--ext EXT ...] [--format rich/porcelain/porcelain-z]	Loads the cached embeddings for PATH (matching the chosen mode/recursion/hidden/gitignore/ext settings), shows matches for QUERY.
vexor doctor	Checks whether the vexor command is available on the current PATH.
vexor update	Fetches the latest release version and shows links to update via GitHub or PyPI.
vexor config --set-api-key/--clear-api-key	Manage the stored API key (Gemini by default).
vexor config --set-model/--set-batch-size/--show	Manage default model, batch size, and inspect current settings.
vexor config --set-provider/--set-base-url/--clear-base-url	Switch embedding providers and optionally override the remote base URL.
vexor config --show-index-all/--clear-index-all	Inspect or delete every cached index regardless of path/mode.

Documentation

See the docs for more details.

Contributions, issues, and PRs are all welcome!

Star this repo if you find it helpful!

License

This project is licensed under the MIT License.