yogthos/Matryoshka
View on GitHub ↗MCP server for token-efficient large document analysis via the use of REPL state
146 ★18 forksTypeScriptUpdated 4mo ago
What you need to know
Approach for processing documents far larger than an LLM context window using recursive language models and a constrained symbolic execution engine, without vector DBs or chunking.
Install
Based on the Recursive Language Models paper (arXiv 2512.24601).
Usage
- •LLM outputs constrained symbolic Nucleus commands (S-expressions) executed by the Lattice logic engine.
Key features
- ✓contexts up to 100x the model window
- ✓no vector DB or chunking
- ✓Nucleus command execution via the Lattice engine
Reviewed 2026-08-11
Topics
- Stars
- 146★
- Forks
- 18
- Language
- TypeScript
- License
- Apache-2.0
- Created
- 2026-01-12
- Last push
- 2026-05-17