Terraphim AI Type Specimen
Weight Scale
font-size: clamp(1.5rem, 4vw, 2rem); font-weight: 700; line-height: 1.15Glyph Showcase
Capabilities
font-size: clamp(1.5rem, 4vw, 2rem); font-weight: 700Terraphim AI is a Rust-based privacy-first AI assistant. It operates entirely on your device, building knowledge graphs from your documents and data sources. No cloud dependency, no tracking, no telemetry.
-
Knowledge Graph Search
Aho-Corasick automata with LeftmostLongest matching. Concept extraction, graph construction, and semantic indexing across 29 specialised crates.
-
Five Relevance Algorithms
TitleScorer, BM25, BM25F, BM25Plus, and TerraphimGraph. Switch per role, per query, per context. Statistical and semantic ranking combined.
-
Eight+ Data Sources
Ripgrep, Confluence, Jira, Discourse, JMAP email, ClickUp, Logseq, and MCP. Each becomes a searchable haystack node in your graph.
-
Role-Based Profiles
Distinct search contexts with custom relevance functions, knowledge domains, and source configurations. Switch without losing state.
-
Firecracker Isolation
Sub-2 second microVM boot for secure command execution. Knowledge graph validation before execution. VM pooling reduces overhead.
-
Local AI Integration
Ollama for local models, OpenRouter for cloud inference. Document summarisation, chat completion, and intelligent descriptions.
Monospace Specimen
font-family: JetBrains Mono; font-size: 0.8125remgit clone https://github.com/terraphim/terraphim-ai
cd terraphim-ai
cargo build --release
cargo run -- --config terraphim_engineer_config.json
POST /documents/search
Content-Type: application/json
{
"search_term": "knowledge graph",
"role": "Terraphim Engineer"
}
cd desktop
yarn install
yarn run tauri dev
Architecture
font-size: clamp(1.5rem, 4vw, 2rem); font-weight: 700The workspace comprises 29 crates spanning core services, agent systems, haystack integrations, and supporting utilities. The persistence layer supports memory, DashMap, SQLite, and S3 backends with transparent cache warm-up. Objects exceeding 1 MB are compressed with zstd before caching.
The frontend is a Svelte application with TypeScript, Vite, and Tauri for native desktop integration. The TUI provides an interactive REPL with hierarchical commands and ASCII graph rendering. The MCP server exposes all automata and rolegraph functions as standard AI tools.
Start Searching
Open source. Rust-powered. Privacy-first. Your knowledge, your device, your control.
View on GitHub