harbi256/galaxdb repository overview

⁠GalaxDB

An AI-native database that unifies SQL, vector search, and local embeddings in a single process. Speaks the PostgreSQL wire protocol, so any Postgres client or driver connects with no changes. No external embedding API, no separate vector database, no data pipeline — one connection string.

Apache-2.0. Source: https://github.com/zentrix-innovative-labs/galaxdb⁠

⁠What's in this image

• galaxdb-server — the database (PostgreSQL wire protocol on 5433, HTTP /health + /metrics on 9090)
• galaxdb-sidecar — the local embedding model runner (opt-in; see below)

Ubuntu 24.04 base. On Linux the engine uses io_uring and transparently falls back to tokio where it isn't available (e.g. inside Docker Desktop).

⁠Quick start

docker run -p 5433:5433 -p 9090:9090 -v /data:/data \
  harbi256/galaxdb:latest --data-dir /data

Then connect with any Postgres client:

psql "host=localhost port=5433 dbname=galaxdb user=galaxdb sslmode=disable"

Check health:

curl http://localhost:9090/health
# {"status":"ok","version":"0.4.0","subsystems":{"disk_full":false,"sidecar_healthy":false,"connections_active":0}}

⁠Semantic search with local embeddings

Embedding columns and SEMANTIC_MATCH need the sidecar attached. Pass --sidecar and a HuggingFace model id; the model is downloaded on first run, so mount the HF cache to persist it across restarts:

docker run -p 5433:5433 -p 9090:9090 \
  -v /data:/data \
  -v ~/.cache/huggingface:/root/.cache/huggingface \
  harbi256/galaxdb:latest \
  --data-dir /data \
  --sidecar /usr/local/bin/galaxdb-sidecar \
  --model sentence-transformers/all-MiniLM-L6-v2

CREATE TABLE docs (
  id   INT PRIMARY KEY,
  body TEXT EMBEDDING MODEL 'sentence-transformers/all-MiniLM-L6-v2' DIM 384
);
INSERT INTO docs (id, body) VALUES (1, 'machine learning and neural networks');

-- Top-k semantic search; add LIMIT to control how many matches come back
-- (without LIMIT, the default is the 10 nearest)
SELECT id, body FROM docs
WHERE SEMANTIC_MATCH(body, 'artificial intelligence', 0.4)
LIMIT 20;

⁠Features

• Full SQL — joins, aggregates, GROUP BY, transactions (snapshot isolation)
• Local embeddings + SEMANTIC_MATCH semantic search (HNSW vector index)
• Time-travel queries (AT VERSION), version tags, FOR TRAINING Lance export
• Near-duplicate detection (WHERE NOT DUPLICATE)
• SCRAM-SHA-256 auth, TLS, AES-256-GCM encryption at rest
• WAL crash recovery, PostgreSQL wire protocol, /health + /metrics

⁠Security — read before exposing this on a network

By default the server runs in trusted-local mode: any connection is accepted as the superuser with no password. This is fine for local development but must never be exposed to an untrusted network as-is.

For anything networked, enable authentication and set a real password:

docker run -p 5433:5433 -p 9090:9090 -v /data:/data \
  -e GALAXDB_AUTH=1 \
  -e GALAXDB_INITIAL_SUPERUSER=admin \
  -e GALAXDB_INITIAL_SUPERUSER_PASSWORD=<a-strong-password> \
  harbi256/galaxdb:latest --data-dir /data

Also consider --tls-mode require (with --tls-cert / --tls-key) to encrypt the wire protocol in transit.

⁠Tags

• latest — most recent stable release
• 0.4.0 — pinned version

⁠Other install options

• Python client: pip install galaxdb-client
• Homebrew: brew tap zentrix-innovative-labs/galaxdb https://github.com/zentrix-innovative-labs/galaxdb && brew install galaxdb
• Binaries + checksums: GitHub Releases⁠
• One-liner installer: curl -fsSL https://raw.githubusercontent.com/zentrix-innovative-labs/galaxdb/main/install.sh | bash

⁠Ports & volumes

Mount /data for persistent storage. If using the sidecar, also mount ~/.cache/huggingface to avoid re-downloading the embedding model on restart.

⁠Documentation

• Getting Started⁠
• Benchmarks⁠
• Roadmap⁠

License: Apache-2.0