srvdb 0.1.7

Zero-Gravity Embedded Vector Database - Offline-first, RAM-efficient vector search

SrvDB: High-Performance Embedded Vector Database

SrvDB is a production-grade, serverless vector database built in Rust. It is designed for edge computing, local AI applications, and high-concurrency environments where low memory footprint and zero-latency persistence are critical.

Unlike client-server vector databases that introduce network overhead, SrvDB runs directly in your application process, utilizing memory-mapped files (mmap) for near-instant access to billion-scale datasets without heavy RAM requirements.

Core Capabilities

• Zero-Copy Architecture: Leveraging OS-level memory mapping to serve vectors directly from disk, bypassing the need for massive heap allocations.
• SIMD Acceleration: Hardware-optimized cosine similarity kernels (AVX-512, NEON) automatically selected at runtime.
• Acid-Compliant Persistence: Buffered Write-Ahead Logging (WAL) strategy ensures zero data loss even during process termination.
• Concurrency: Thread-safe architecture demonstrating linear scalability up to 16 cores (beating ChromaDB in high-load benchmarks).

Performance Benchmarks

Benchmarks were conducted on standard consumer hardware (NVMe SSD, 16GB RAM).

Metric	SrvDB v0.1.6	Target	Status
Ingestion Throughput	797 vectors/sec	>500	✅ Exceeded
Search Latency (P99)	3.8ms	<15ms	✅ Exceeded
Recall Accuracy	100.0%	100%	✅ Exact
Storage Efficiency	~192 bytes/vec	-	-

Benchmark methodology: 10k vectors (1536-dim), k=10, single-node ingestion.

Installation

Python

pip install srvdb

Rust

Add this to your Cargo.toml:

[dependencies]
srvdb = "0.1.6"

Quick Start

Python Example

import srvdb

# Initialize (creates ./search_index if not exists)
db = srvdb.SvDBPython("./search_index")

# Add Data (Auto-persisted via WAL)
db.add(
    ids=["vec_1", "vec_2"],
    embeddings=[[0.1] * 1536, [0.2] * 1536],
    metadatas=['{"source": "prod"}', '{"source": "dev"}']
)

# Search
results = db.search(
    query=[0.1] * 1536,
    k=5
)

print(f"Found ID: {results[0][0]}, Score: {results[0][1]}")

Architecture

SrvDB utilizes a hybrid storage engine:

1. Index Layer: A flat, memory-mapped binary file (vectors.bin) containing raw Float32 vectors.
2. Metadata Layer: A durable embedded Key-Value store (metadata.db) mapping string IDs to internal offsets.
3. Buffer Layer: A 1MB write-buffer (BufWriter) that batches I/O syscalls, providing a 5x improvement in ingestion speed over unbuffered IO.

Licensing

SrvDB is open-source software licensed under the AGPL v3.0.

• Open Source Use: Free to use in any GPL/AGPL compatible open-source project.
• Commercial Use: If you wish to embed SrvDB in a proprietary, closed-source application without releasing your source code, you must purchase a Commercial License.

Contact [[e-mail]] for commercial licensing inquiries.

### 5. Push to GitHub

Now, execute these commands to push your professional repo:

```bash
# 1. Initialize Git (if not done)
git init

# 2. Add the new files
git add README.md LICENSE CONTRIBUTING.md BENCHMARK.md

# 3. Commit
git commit -m "docs: release preparation v0.1.6"

# 4. Rename branch to main
git branch -M main

# 5. Connect to your repo (Change URL to yours)
git remote add origin https://github.com/Srinivas26k/srvdb.git

# 6. Push
git push -u origin main