# CoDynamics Lab — LATCH

## What is LATCH?
LATCH is a proprietary LLM inference layer that compiles document sets into persistent model-level memory. It is not RAG. It is not prompt compression. It is not a KV cache optimization. It is a new category: compiled document memory at the model level.

## How it works
1. Documents are processed through a proprietary compilation step.
2. A persistent binary representation is saved to disk as a .latch or .latchdoc file.
3. Subsequent queries run against the compiled memory — no raw document re-processing.
4. Compilation cost is paid once and amortized across all future queries.

## Key performance claims (benchmarked on NVIDIA H100 80GB, vLLM)
- Time-to-first-token: 0.11s (vs 23.1s baseline cold start) — 210× faster
- Cache reload from .latch file: 1.6ms
- Multi-document benchmark pass rate: 91.7% (11/12 gates)
- Cost reduction: 97% amortized after 25 queries
- VRAM reduction: 50%
- End-to-end speedup: 5.2×
- Supported model families: Qwen, Mistral, Llama, DeepSeek

## Portable file formats
- .latch — privacy-first binary containing only compiled model memory (no source text)
- .latchdoc — full package with embedded raw text for full-text search and quality fallback

## Deployment
- Self-hosted Docker image (40GB+ initial pull)
- Runs on NVIDIA H100/A100 80GB GPUs
- OpenAI-format API compatible
- Local browser-based console UI included

## Pricing
- Evaluation / Personal: $79 one-time (up to 3 activations, one user)
- Commercial Deployment: contact sales (annual license)
- Enterprise / OEM: custom agreement

## Links
- Website: https://www.codynamicslab.com
- Documentation: https://www.codynamicslab.com/documentation/
- Purchase: https://codynamicslab.gumroad.com/
- Hugging Face: https://huggingface.co/CoDynamicsLab/LATCH-Qwen2.5-14B
- Contact: mike@codynamicslab.com

## Company
CoDynamics Lab Corporation, Delaware C-Corp, Gilbert, AZ.
Built and operated by a solo founder with 20+ years of product development engineering experience and 14 granted US patents.