TyKolt/kremis

What it does

TyKolt/kremis is a Model Context Protocol (MCP) server that provides a deterministic knowledge graph implementation. It enables AI applications to store, query, and manage structured knowledge locally without requiring embedding calls or external LLM dependencies. The tool supports canonical KREX export format for byte-identical audit trails and uses BLAKE3 hashing for cryptographic state verification.

How to set up

As an MCP server, kremis runs as a single binary compatible with Rust, macOS, Windows, and Linux environments. Download the binary from the GitHub repository, configure it in your MCP client (Claude, ChatGPT, or compatible applications), and connect via the MCP protocol. No additional dependencies or external services required beyond the binary executable.

Key features

• Deterministic knowledge graph with zero LLM/embedding calls in bridge layer
• BLAKE3 cryptographic state hashing for verifiable audit trails
• Canonical KREX export format for byte-identical reproducibility
• Local-first architecture with redb ACID compliance
• Single binary deployment across multiple platforms
• Alpha status with active development