patchvec 0.5.6

Patchvec — A lightweight, pluggable vector search microservice.

PatchVec — A lightweight, pluggable vector search microservice.

Upload → chunk → index (with metadata) → search via REST and CLI.

Highlights

• Multi-tenant collections: /collections/{tenant}/{name}
• Upload and search TXT, CSV, and PDF
• Chunking per format:
  • PDF → 1 chunk/page
  • TXT → configurable chunk size + overlap
  • CSV → 1 chunk/row
• Metadata filters on POST search ({"filters": {"docid": "DOC-1"}})
• Health, metrics, and Prometheus endpoints
• Configurable auth modes: none or static (Bearer)
• Default backends are local (vendor-neutral "default"); embedders & stores are pluggable

Requirements

• Python 3.10+

Install

python -m venv .venv
source .venv/bin/activate
pip install patchvec

CPU-only by default. If you have a CUDA setup and want GPU-accelerated deps, install the GPU-enabled packages in your environment before running PatchVec.

Quickstart

# Start the server (installed entry point)
pavesrv

# Or run with uvicorn manually if you prefer:
uvicorn pave.main:app --host 0.0.0.0 --port 8080

Minimal config (optional)

By default PatchVec runs with sensible local defaults. To customize, create config.yml and set:

vector_store:
  type: default
embedder:
  type: default
auth:
  mode: none    # or 'static' with per-tenant Bearer keys

Then export:

export PATCHVEC_CONFIG=./config.yml

REST example

# Create a collection
curl -X POST http://localhost:8080/collections/acme/docs

# Upload a TXT document
curl -X POST http://localhost:8080/collections/acme/docs/documents   -F "file=@sample.txt" -F "docid=DOC1"

# Search (GET, no filters)
curl -G --data-urlencode "q=hello" http://localhost:8080/collections/acme/docs/search

# Search (POST, with filters)
curl -X POST http://localhost:8080/collections/acme/docs/search   -H "Content-Type: application/json"   -d '{"q":"hello","k":5,"filters":{"docid":"DOC1"}}'

License

GPL-3.0-or-later — (C) 2025 Rodrigo Rodrigues da Silva