PistaDBThe embedded vector database for LLM-native apps

Pure C99 · Zero dependencies · Single .pst file · Sub-millisecond search on a laptop

Get Started

Why PistaDB?

View on GitHub

🥜

Small, dense, full of value

One .dll / .so / .dylib. One .pst file. No Docker, no API keys, no data leaving the machine.

⚡

8 production index algorithms

HNSW, IVF, IVF-PQ, DiskANN, LSH, ScaNN, SQ, and exact linear scan — pick the right one per workload.

🧠

Built for RAG and agents

Cosine, L2, IP, L1, Hamming. Covers OpenAI, Cohere, BGE, CLIP, and binary embeddings out of the box.

🚀

SIMD-accelerated

AVX2+FMA on x86-64, NEON on ARM, runtime-dispatched. 4–8× faster than scalar.

🔁

ACID transactions

Atomic multi-op groups with full undo-on-failure rollback. Crash-safe WAL.

🎭

Multi-modal & hybrid search

Multiple named vector fields per record + RRF fusion. Text + image + audio in one bundle, atomic writes.

🌐

10 language bindings

C, C++, Python, Go, Java, Kotlin, Swift, Objective-C, C#, Rust, Julia, WASM. Same API everywhere.

Quick taste

python

import numpy as np
from pistadb import PistaDB, Metric, Index

with PistaDB("docs.pst", dim=1536, metric=Metric.COSINE, index=Index.HNSW) as db:
    db.insert(1, np.random.rand(1536).astype("float32"), label="chunk_0001")
    results = db.search(np.random.rand(1536).astype("float32"), k=10)
    for r in results:
        print(r.id, r.distance, r.label)
    db.save()

Every LLM application eventually needs a vector database. PistaDB gives you a production-grade one in a single file and a C library with zero dependencies. Ship it inside a desktop app. Bundle it on an edge device. Drop it next to your Python script.