Terraphim has been introduced, a private AI assistant and knowledge graph search engine that operates entirely on the user's device. The project is written in Rust and compiled in WebAssembly: indexing, searching, and traversing the graph are done locally, with no cloud, no telemetry, and no data sent outside. The code is open under the Apache 2.0 license. Development has been ongoing since 2023, with the last year involving AI integration.
Instead of neural network vectors, deterministic graph embeddings and pattern matching based on the Aho-Corasick algorithm are used on role-based thesauri — each match is traced back to a concept in the knowledge graph and takes less than a millisecond. One query covers multiple sources: local files, Confluence, Jira, Discourse, email, and grepapp.
For AI agents, there is an MCP server (connecting local knowledge graphs to assistants), a CLI terraphim-agent (searching through Claude Code and Cursor sessions, capturing errors and lessons learned), as well as hooks before and after calling tools. The Terraphim editor is also available, providing alternative phrasings based on the knowledge graph.
Source: opennet.ru
