Matryoshka FAQs

Question 1

How does Matryoshka achieve such high token efficiency?

Accepted Answer

It utilizes a unique handle-based storage architecture with an in-memory SQLite. Instead of sending full data, LLMs receive compact references (handles), leading to over 97% token savings by executing operations server-side on these handles.

Question 2

What problem does Matryoshka solve for LLMs?

Accepted Answer

Matryoshka enables Large Language Models to analyze documents and codebases significantly larger than their typical context windows. It bypasses limitations by using symbolic reasoning and a logic engine instead of traditional RAG or chunking methods.

Question 3

Is Matryoshka safe for LLMs to generate commands?

Accepted Answer

Yes, Matryoshka employs Nucleus DSL, a constrained S-expression language. LLM outputs are parsed, type-checked, and safely executed by the Lattice engine, preventing arbitrary or malicious code generation and execution.

Question 4

What is the Model Context Protocol (MCP) server integration?

Accepted Answer

Matryoshka includes a `lattice-mcp` server, which provides a direct interface for coding agents to perform efficient, chained document and code analysis. This enables agents to interact with the powerful Nucleus engine using compact handle-based results, further enhancing token efficiency for complex workflows.

Question 5

Can Matryoshka analyze code in addition to regular documents?

Accepted Answer

Absolutely. Matryoshka integrates Tree-sitter for structured code analysis across 20+ programming languages. This allows for code-aware queries, symbol listing, function body extraction, and efficient reference finding within large codebases.

Matryoshka

Matryoshka

Características Principales

Casos de Uso

Características Principales

Casos de Uso