TERRAPHIM AI v1.5.2
Applied Knowledge Systems Ltd -- Privacy-First Semantic Search
================================================================
Initialising kernel ........ [OK]
Loading automata engine .... [OK]
Building knowledge graph ... [OK]
Indexing haystacks ......... [OK]
Starting MCP server ........ [OK]
----------------------------------------------------------------
Role: Terraphim Engineer
Graph: 12,847 nodes | 38,291 edges
Haystacks: ripgrep, confluence, jira, logseq
Relevance: TerraphimGraph (BM25 fallback)
Privacy: ALL DATA LOCAL -- NO CLOUD CONNECTIONS
================================================================
System ready. All operations run on your hardware.
Type > help for available commands.
================================================================================
> help
AVAILABLE COMMANDS
> search <query>
Semantic search across all indexed haystacks. Uses knowledge graph expansion to find related concepts. BM25, TitleScorer, and TerraphimGraph relevance functions rank results.
> graph --build <source>
Build or update the knowledge graph from documents. Aho-Corasick automata extract concepts. LeftmostLongest matching ensures precise term identification across your corpus.
> haystack --add <type>
Connect a new data source. Supported: ripgrep (local files), confluence, jira, discourse, jmap (email), logseq, mcp.
> chat <message>
AI conversation with role-based context. Uses local Ollama or OpenRouter. Knowledge graph provides grounding. Summarisation, Q&A, and document analysis.
> autocomplete <prefix>
Sub-millisecond term completion powered by automata. Searches across thesaurus entries and document titles. Available via MCP, HTTP API, and WASM in the browser.
> vm --exec <command>
Execute untrusted code in Firecracker microVMs. Sub-two-second boot. Knowledge graph validation gates execution. Full isolation with VM-level sandboxing.
================================================================================
> search "semantic search architecture"
Searching 4 haystacks... expanding 3 concepts via knowledge graph...
Found 23 results in 47ms
1. Knowledge Graph Architecture
source: confluence | score: 0.94 | matched: 5 concepts
Describes the rolegraph design and node/edge relationships
2. Search Pipeline Orchestration
source: ripgrep | score: 0.87 | matched: 3 concepts
Middleware layer coordinating queries across haystacks
3. Automata-Based Term Expansion
source: logseq | score: 0.81 | matched: 4 concepts
Thesaurus builder using Aho-Corasick for concept extraction
----------------------------------------------------------------
> graph --stats
Knowledge Graph Status:
Nodes: 12,847
Edges: 38,291
Automata: 4 active (LeftmostLongest)
Thesaurus: 6,432 entries
Last built: 2m 14s ago
>
================================================================================

Architecture

Language Rust (edition 2024)
Runtime Tokio async
Crates 29 workspace members
Desktop Svelte + Tauri
WASM Browser autocomplete

Search Engine

BM25 Active
TitleScorer Active
TerraphimGraph Active (primary)
Autocomplete < 1ms response
Automata Aho-Corasick LML

Privacy Status

Cloud calls NONE
Telemetry DISABLED
Data location LOCAL ONLY
Sandboxing Firecracker VM
Licence Open source
================================================================================

> cat README.md

Terraphim AI is developed by Applied Knowledge Systems Ltd and released as open source. A Cargo workspace of 29 specialised crates, each with a single responsibility: from automata engines to agent supervisors, from haystack integrations to knowledge graph orchestration.

Every component is inspectable, replaceable, and yours to extend. Run it as a desktop app (Tauri + Svelte), a terminal REPL, an HTTP API server, or integrate it into your AI workflow via the MCP server.

git clone man terraphim
================================================================================
READY.