Helix

Instant embedded multi-model engine: Relational tables + Vector similarity search + Key-Value cache with auto-generated REST API.

Designed by Atlas Research Labs as a zero-Docker, static single-runtime alternative to Supabase, Pinecone, and Redis.

Architectural Principles

• Zero-Docker Simplicity: Eliminates the operational complexity of running separate Postgres, pgvector, and Redis containers for modern AI backends. One static runtime embeds all three paradigms.
• Relational Document Storage: ACID append-only write-ahead journals (.helix/tables/*.jsonl) with primary keys, equality, and range filter operators ($eq, $gt, $gte, $lt, $lte, $in).
• High-Performance Vector Index: Exact Cosine Similarity and Euclidean Distance (L2) nearest-neighbor indexing for RAG pipelines, text embeddings, and semantic search.
• In-Memory Key-Value Cache with TTL: Microsecond-latency cache with active expiration, wildcard pattern lookups (session:*), and persistence snapshots.
• Auto-Generated REST API: Instant endpoints for all tables, vectors, and cache keys with zero configuration or schema boilerplate.
• Embedded OLED Studio: Pure OLED black local Web UI (http://127.0.0.1:6060) with Table Explorer, Vector Similarity Inspector, and KV Manager.
• Zero External Dependencies: Implemented entirely with Node.js standard libraries (http, crypto, fs, path). Package size under 100 KB, idle memory under 15 MB.

Installation

Global CLI installation via NPM:

npm install -g helixdb

Or execute directly via npx:

npx helixdb --help

Both binary commands helix and helixdb are exposed.

CLI Reference

1. Starting Database Engine & Auto-REST Server

Start the REST API server:

# Default port 6060
helix server

# Custom port and interface
helix server -p 8080 -h 0.0.0.0

2. Embedded OLED Web Studio

Launch the visual management dashboard:

helix studio

Open http://127.0.0.1:6060 in any browser to inspect tables, execute vector searches, and manage cache keys.

3. Relational Table Operations

# Insert a record
helix insert users '{"name":"Eren Yeager","role":"admin","level":50}'

# Query records with filtering
helix select users --where '{"role":"admin"}'

# Query with comparative operators
helix select users --where '{"level":{"$gt":25}}'

4. Vector Similarity Search (RAG)

# Insert vector embedding
helix vector insert documents --vector '[0.25, 0.50, 0.75, 1.0]' --metadata '{"title":"System Architecture"}'

# Search top-3 nearest neighbors (Cosine similarity)
helix vector search documents --vector '[0.25, 0.50, 0.75, 1.0]' -k 3 --metric cosine

5. In-Memory Key-Value Cache

# Set key with optional 300 second (5 min) TTL
helix set session:user_9988 '{"auth":true}' --ttl 300

# Get key
helix get session:user_9988

# Delete key
helix del session:user_9988

# Display database metrics
helix stats

Auto-Generated REST API Reference

When helix server or helix studio is running:

Relational Tables

• GET /api/v1/tables: List all tables and row counts.
• GET /api/v1/tables/:table: Query records (?limit=10&offset=0).
• POST /api/v1/tables/:table: Insert new JSON record.
• GET /api/v1/tables/:table/:id: Get document by ID.
• PUT /api/v1/tables/:table/:id: Update document fields.
• DELETE /api/v1/tables/:table/:id: Delete document.

Vectors

• GET /api/v1/vectors: List vector collections.
• POST /api/v1/vectors/:collection: Insert { vector: number[], metadata?: object }.
• POST /api/v1/vectors/:collection/search: Execute search { vector: number[], topK?: 5, metric?: "cosine" }.
• DELETE /api/v1/vectors/:collection/:id: Delete vector.

Key-Value Cache

• GET /api/v1/kv?pattern=*: List keys.
• GET /api/v1/kv/:key: Get key value and remaining TTL.
• PUT /api/v1/kv/:key: Set { value: any, ttl?: number }.
• DELETE /api/v1/kv/:key: Delete key.

Programmatic Usage

Helix can be embedded directly as a library:

const { TableEngine, VectorEngine, CacheEngine } = require('helixdb');

// Tables
const tables = new TableEngine();
const user = tables.insert('users', { name: 'Eren', tier: 'enterprise' });

// Vectors
const vectors = new VectorEngine();
vectors.insert('articles', { vector: [0.1, 0.2, 0.3], metadata: { title: 'AI' } });
const results = vectors.search('articles', [0.1, 0.2, 0.3], { topK: 1 });

// Cache
const cache = new CacheEngine();
cache.set('active_token', 'jwt_xyz', 60);

Quality & Compliance

• Zero Emojis: Strictly pure ASCII/ANSI logging across CLI, UI, and documentation.
• Zero Mock Data: Authentic mathematical vector distance and filesystem persistence.
• Zero Docker: Runs as a single process with minimal resource footprint.

License

MIT License. Copyright (c) 2026 Atlas Research Labs.