# PuterVision — MCP Research & Development Labs

> PuterVision is a private engineering research laboratory pioneering deterministic state graph storage, perceptual visual caching, and context compression algorithms for AI developer agents and Model Context Protocol (MCP) architectures.

## Core Research Focus
- **Deterministic Context Stores**: SQLite-backed state graph memory for AI coding agents to eliminate LLM hallucination in task tracking and audit trails.
- **Perceptual Fast-Path Matching**: Sub-5ms screen state recognition using Hamming distance dHash algorithms to bypass expensive vision LLM calls.
- **Spec-Driven Development (SDD)**: Compiling Markdown PRDs and Gherkin BDD specs into graph execution nodes to verify requirement coverage.
- **Local Multimodal Perception**: Embedded zero-cloud CLIP vision models and LanceDB vector stores for private semantic retrieval.

## Published Open-Source Initiatives

### 1. State Memory MCP (`@putervision/state-memory-mcp`)
- **Website**: https://statememorymcp.com
- **Repository**: https://github.com/putervision/state-memory-mcp
- **Description**: Zero-infrastructure, deterministic Model Context Protocol (MCP) server providing AI agents with persistent SQLite state graph memory for tracking tasks, decisions, blockers, plans, and artifacts.
- **Primary Capabilities**: SQLite graph database, 3D WebGL graph visualizer, Spec-Driven Development (SDD), Git branch isolation.

### 2. Visual Memory MCP (`@putervision/vision-memory-mcp`)
- **Website**: https://visionmemorymcp.com
- **Repository**: https://github.com/putervision/vision-memory-mcp
- **Description**: Visual perception cache & perceptual hashing MCP server. Caches webpage and app screenshots locally using dHash, local CLIP vector search, and accessibility trees to cut LLM token costs by up to 90%.
- **Primary Capabilities**: L1/L2 dHash zero-token fast-path, L3 local CLIP vector search, visual layout drift verification.

### 3. ScreenChunk (`ScreenChunk`)
- **Website**: https://screenchunk.com
- **Repository**: https://github.com/putervision/screenchunk
- **Description**: Ultra-fast live screen capture studio, 2D region cropping, and WebRTC screen recorder with client-side zero-knowledge AES-256-GCM chunk protection.
- **Primary Capabilities**: In-browser desktop recording, 2D canvas region selector, WebCrypto AES-GCM chunking, zero-server recording, instant shareable decryption links.

### 4. BassMusic.AI (`BassMusic.AI`)
- **Website**: https://bassmusic.ai
- **Repository**: https://github.com/putervision/bassmusicai
- **Description**: Free, 100% client-side AI electronic bass music and vocal stem generator running in-browser via WebGPU and WebAssembly SIMD neural models (MusicGen & VITS/MMS).
- **Primary Capabilities**: Zero-server inference, 60+ electronic bass genres, procedural FX synth, 5-band parametric EQ, seamless looping, Playwright/agent automation API (`window.bassMusicAI`).

## Canonical URLs & References
- **PuterVision R&D Labs**: https://putervision.com
- **State Memory MCP**: https://statememorymcp.com
- **Visual Memory MCP**: https://visionmemorymcp.com
- **ScreenChunk**: https://screenchunk.com
- **BassMusic.AI**: https://bassmusic.ai
- **GitHub Organization**: https://github.com/putervision