TERRAPHIM

2026-03-24 · General Ledger

Privacy-First AI Assistant — Semantic Search via Knowledge Graphs

Feature Register

Core capabilities of Terraphim AI, with current implementation status
# Feature Description Status
01 Knowledge Graphs Custom graph-based semantic search using Aho-Corasick automata for fast text matching
02 Privacy-First All processing runs locally on your device. No data ever leaves your machine
03 Multi-Source Search Search across Confluence, Jira, Discourse, email (JMAP), local files, and more
04 Role-Based Profiles Configure different knowledge domains and search preferences per role
05 BM25 Relevance Advanced text ranking algorithms including BM25, BM25F, BM25Plus, and TitleScorer
06 MCP Server Model Context Protocol integration for AI development tools and autocomplete
07 Desktop Application Native desktop app built with Svelte, TypeScript, and Tauri
08 WASM Autocomplete Browser-based autocomplete via WebAssembly (~200KB compressed)
09 Firecracker VMs Secure command execution in microVMs with sub-2-second boot times
10 LLM Integration OpenRouter and Ollama support for summarisation and chat
11 Agent System Multi-agent coordination, supervision, evolution, and goal alignment
12 Quickwit Search Cloud-native search engine integration for log and observability data
Total Features 12 registered · 11 complete · 1 in progress

Architecture Breakdown

Core Components

terraphim_service
Main service layer -- search orchestration, document management, AI integration
terraphim_middleware
Haystack indexing, document processing, multi-source search coordination
terraphim_rolegraph
Knowledge graph implementation with node/edge relationships and thesaurus
terraphim_automata
Text matching engine: autocomplete, thesaurus building, Aho-Corasick automata
terraphim_persistence
Multi-backend storage abstraction with cache warm-up (memory, dashmap, SQLite, S3)
terraphim_config
Role-based configuration management with JSON configs and sensible defaults

Haystack Integrations

Data source connectors for multi-source semantic search
# Source Protocol Status
01RipgrepLocal filesystem via rg
02AtlassianConfluence + Jira REST API
03DiscourseForum REST API
04JMAPEmail via JMAP protocol
05QuickwitCloud-native search engine
06MCPModel Context Protocol (SSE/HTTP)
Total Integrations 6 registered · 6 complete

Quick Start Journal

Journal Entry: Installation 
git clone https://github.com/terraphim/terraphim-ai
cd terraphim-ai
cargo build --release

# Start the backend server
cargo run --release -- --config terraphim_engineer_config.json

# In a separate terminal, start the frontend
cd desktop
yarn install
yarn dev
Journal Entry: MCP Server 
# Start the MCP server for AI tool integration
cd crates/terraphim_mcp_server
./start_local_dev.sh

# Available MCP tools:
# autocomplete_terms, find_matches, replace_matches,
# load_thesaurus, is_all_terms_connected_by_path,
# fuzzy_autocomplete_search_jaro_winkler
Journal Entry: TUI Interface 
# Build and run the terminal UI
cargo build -p terraphim_tui --features repl-full --release
./target/release/terraphim-agent

# REPL commands: /search, /chat, /commands, /vm, /help

Summary Totals

29 Crates
6 Haystacks
5 Relevance Functions
3 Platforms
<2s VM Boot Time
200KB WASM Size

Open the Ledger

Your data. Your device. Your knowledge graph.

View on GitHub Documentation