Terraphim AI is a privacy-first search assistant that runs on your machine. It builds knowledge graphs from your documents and finds what matters -- without ever sending your data elsewhere.
Terraphim does not just search text -- it understands structure. Every document you index gets processed through Aho-Corasick automata that extract concepts and build a graph of relationships. The result is a personalised knowledge network that improves with every file you add. Each role you configure gets its own graph, its own thesaurus, and its own relevance function.
A haystack is any source of knowledge. Local folders indexed with ripgrep. Confluence wikis and Jira tickets through the Atlassian connector. Discourse forums. Email via JMAP. Even Quickwit for log analysis. Each haystack plugs into the same interface, so adding a new source is straightforward. All indexing happens on your hardware.
The pattern matching engine uses Aho-Corasick automata with LeftmostLongest matching. Thousands of terms matched simultaneously in a single pass. Autocomplete returns in under 5 milliseconds. The automata compile to WebAssembly at roughly 200KB compressed, so the same engine runs in your browser, your terminal, and your desktop application.
We believe the best tools feel like workshops, not factories. Terraphim is designed for people who think carefully about their work and want their tools to respect that care.
-- Applied Knowledge Systems LtdEvery line of Terraphim is open source. The workspace compiles to native binaries, WebAssembly, and Tauri desktop applications. Async throughout with Tokio. Memory safety guaranteed at compile time. Built by Applied Knowledge Systems Ltd for people who believe privacy is a right, not a premium feature.