CLI

Install the base command with ParqDB:

python -m pip install parqdb
parqdb --help

parqdb serve

Run the experimental HTTP server:

parqdb serve
parqdb serve --config ./parqdb-server.toml

The server reads ./parqdb-server.toml when it exists. See Run a server for authorization and deployment guidance.

parqdb config init

Write a default server configuration:

parqdb config init
parqdb config init --path ./parqdb-server.toml

parqdb publish

Build or reuse an immutable browser-queryable IVF-LVQ index and publish it to a local directory or S3-compatible object storage:

python -m pip install "parqdb[publish]"

parqdb publish \
  --source documents.parquet \
  --key chunk_id \
  --vector-column embedding \
  --nlist 4096 \
  --destination s3://my-bucket/kb/v1 \
  --public-url https://data.example.com/kb/v1

Required options

Option	Meaning
--source PARQUET	Source Parquet file
--key COLUMN	Dense, ordered, non-null int64 key starting at zero
--destination PATH_OR_S3_URI	New publication prefix

When building a new index, provide --nlist and exactly one embedding input:

• --vector-column COLUMN for existing vectors; or
• one or more --text-column COLUMN options for pinned MiniLM embeddings.

Use --index-manifest PATH to reuse an index that has already been built.

Index options

Option	Default	Meaning
--encoding	lvq8	Posting encoding: lvq4 or lvq8
--metric	cosine	cosine or l2_squared
--threads	8	Build threads, from 1 through 16
--embedding-batch-size	128	MiniLM embedding batch size
--work	.parqdb-publish	Resumable local build directory

Publication options

Option	Meaning
--include-source	Publish the source table for browser payload lookup
--include-model	Publish the pinned model used by --text-column
--public-url	Public HTTPS base URL used in the manifest
--s3-endpoint	S3-compatible endpoint
--s3-region	S3 region; use auto for Cloudflare R2
--cors-origin	Origin used for CORS verification
--no-verify-http	Skip public HTTP Range and CORS verification

The command never overwrites an existing destination prefix. See Publish for the browser for the full workflow.