The search engine that never leaves your machine

Terraphim AI runs entirely on your hardware. Semantic search across every knowledge source you use -- private, fast, and built in Rust.

Capabilities

Everything you need, nothing in the cloud

Multi-algorithm search

Five relevance functions -- TitleScorer, BM25, BM25F, BM25Plus, and TerraphimGraph -- run concurrently. The best result surfaces regardless of the underlying algorithm.

Knowledge graph automata

Aho-Corasick automata match concepts using LeftmostLongest strategy. Build thesauri from documents or URLs and construct rolegraphs per user profile automatically.

Universal haystack connectors

Ripgrep, Confluence, Jira, Discourse, JMAP email, ClickUp, Logseq, Quickwit, and MCP -- all unified through a single middleware layer with consistent ranking.

Firecracker VM isolation

Run untrusted code in fully sandboxed microVMs with sub-2-second boot times. The pooling system pre-warms instances for under 500ms allocation when you need it.

29 Rust crates

A modular workspace of 29 specialised crates covering search, persistence, agent orchestration, knowledge graphs, and haystack integrations. Each crate does one thing well.

Local LLM support

Integrate Ollama for local inference or OpenRouter for cloud models -- your choice, your data. AI summarisation, chat, and document processing without mandatory cloud APIs.

How it works

Three steps to private search

01

Configure your haystacks

Point Terraphim at your knowledge sources: local folders, Confluence wikis, email, or any supported connector. Everything stays local.

02

Build your knowledge graph

The automata engine extracts concepts, builds thesauri, and constructs role-specific graphs that understand your domain vocabulary.

03

Search across everything

Query once, search everywhere. Five ranking algorithms compete to surface the most relevant results in under 200 milliseconds.

In action

See it running

terraphim-agent -- ~/projects
$ cargo run --release -- --config terraphim_engineer_config.json Terraphim Server v0.9.0 Role loaded: Terraphim Engineer Relevance function: TerraphimGraph Haystacks: ripgrep, confluence, discourse, quickwit Server listening on http://127.0.0.1:8080 --- $ curl -s localhost:8080/documents/search -d '{"query":"automata design"}' { "results": [ {"title": "Aho-Corasick Automata Design", "score": 0.94, "source": "ripgrep"}, {"title": "RoleGraph Construction Pipeline", "score": 0.87, "source": "confluence"}, {"title": "Thesaurus Building from URLs", "score": 0.83, "source": "discourse"} ], "elapsed_ms": 142, "haystacks_searched": 4 } $ terraphim-agent Terraphim Agent -- interactive REPL terraphim> /commands list /search Semantic search across haystacks /chat AI conversation with context /vm list Firecracker VM management /help Show all available commands
Architecture

Built on 29 specialised crates

terraphim_service

Core search, document management, and AI integration layer

terraphim_rolegraph

Knowledge graph with nodes, edges, and role-based relationships

terraphim_automata

Aho-Corasick text matching, autocomplete, and thesaurus building

terraphim_middleware

Haystack indexing, document processing, and search orchestration

terraphim_persistence

Multi-backend storage with transparent cache warm-up

terraphim_firecracker

Firecracker microVM integration for secure execution

terraphim_config

Role-based configuration with multi-backend support

terraphim_tui

Terminal UI with hierarchical REPL commands