Terraphim AI is a privacy-first assistant that searches your knowledge repositories locally, using semantic graphs and automata. Nothing leaves your machine.
Every search query, every document, every relationship in your knowledge graph exists solely on your hardware. Terraphim AI is built in Rust with no cloud dependencies. There are no analytics endpoints, no telemetry hooks, no API calls to external servers.
This is not a limitation. It is the point. When your data never leaves your machine, trust becomes a property of the system rather than a promise from a vendor.
Structured relationships between concepts, built from your documents. Role-specific graphs adapt search to your domain. The rolegraph system maintains nodes, edges, and document references for personalised ranking that evolves with your work.
Three relevance functions work in concert. TitleScorer for direct matches, BM25 for statistical relevance, TerraphimGraph for semantic expansion through thesaurus-driven concept networks. Find what you mean, across every source you own.
Blazing-fast text matching and thesaurus expansion using LeftmostLongest automata. Sub-millisecond autocomplete across thousands of concepts. The engine that connects typed queries to the full breadth of your knowledge graph.
Search across local files via ripgrep, Confluence, Jira, Discourse, email through JMAP, Logseq notes, and more. A single query traverses every haystack. Results ranked by a unified relevance score.
Async Tokio runtime. Memory-safe concurrency. 29 specialised crates, each with a single responsibility. Compiled to native speed on every platform. Desktop via Tauri. Terminal via REPL. Browser via WASM.
Untrusted code runs in Firecracker microVMs with sub-two-second boot. Knowledge graph validation gates execution. Secure web request sandboxing. The safety of a VM with the speed of a container.
Developed by Applied Knowledge Systems Ltd and released under a permissive licence. Every component is inspectable, replaceable, and yours to extend.