Orunla

A local-first AI memory system that stores facts as a knowledge graph on your machine. Orunla extracts entities and relationships from text using on-device AI (GliNER/ONNX), applies Ebbinghaus forgetting curves to keep memories relevant, and provides hybrid retrieval (FTS5 + graph search) ranked by recency and confidence.

Features

• Knowledge graph storage -- entities, relationships, and facts in SQLite with FTS5 indexing
• Local AI extraction -- GliNER runs on-device via ONNX Runtime (no API keys, no cloud, works offline)
• Memory decay -- Ebbinghaus forgetting curve prunes stale facts automatically
• Hybrid search -- full-text keyword search + graph traversal, ranked by strength
• Multiple interfaces -- desktop app (Tauri), CLI, REST API, MCP server
• MCP integration -- works with Claude Code, Claude Desktop, Cursor, Cline, Windsurf, and any MCP client
• 100% private -- all data stays in ~/.orunla/memory.db

Quick Start

Download

Grab the latest release for your platform from the Releases page. Each zip contains the desktop app, CLI, MCP server, and ONNX runtime.

CLI

# Add a memory
orunla_cli ingest "Sarah works at Microsoft and lives in Seattle."

# Search memories
orunla_cli recall "Who is Sarah?"

# Start the REST API server
orunla_cli serve --port 8080

MCP Server (for AI agents)

Add to your MCP config (Claude Code, Cursor, etc.):

{
  "mcpServers": {
    "orunla": {
      "command": "/path/to/orunla_mcp"
    }
  }
}

See MCP.md for full setup instructions per IDE.

REST API

# Save a memory
curl -X POST http://localhost:8080/ingest \
  -H "Content-Type: application/json" \
  -d '{"text": "The deploy key rotates every 90 days."}'

# Search
curl -X POST http://localhost:8080/recall \
  -H "Content-Type: application/json" \
  -d '{"query": "deploy key"}'

See API_REFERENCE.md for all endpoints.

How It Works

Text Input --> GliNER Entity Extraction --> Knowledge Graph (SQLite)
                                                  |
                                           Hybrid Retrieval
                                        (FTS5 + Graph Search)
                                                  |
                                         Ebbinghaus Decay
                                        (rank by strength)

1. Ingestion -- GliNER extracts entities (people, orgs, locations, concepts) and relationships from text, producing subject-predicate-object triplets
2. Storage -- triplets are stored as nodes and edges in SQLite with FTS5 indexing on source text
3. Retrieval -- hybrid search combines keyword matching (FTS5) with graph traversal, then ranks by memory strength
4. Decay -- strength = e^(-days/30) * (1 + ln(1 + access_count)) * confidence -- unused memories fade, frequently accessed ones persist

Building from Source

Prerequisites

• Rust toolchain (stable) -- rustup.rs
• ONNX Runtime 1.20.0 -- required for GliNER entity extraction
• Node.js 22+ -- only needed if building the desktop app

1. Download ONNX Runtime

Download from github.com/microsoft/onnxruntime/releases/v1.20.0:

Platform	Download	Library file
Windows x64	onnxruntime-win-x64-1.20.0.zip	lib/onnxruntime.dll
macOS Apple Silicon	onnxruntime-osx-arm64-1.20.0.tgz	lib/libonnxruntime.1.20.0.dylib
macOS Intel	onnxruntime-osx-x86_64-1.20.0.tgz	lib/libonnxruntime.1.20.0.dylib
Linux x64	onnxruntime-linux-x64-1.20.0.tgz	lib/libonnxruntime.so.1.20.0

Extract the archive and place the library file where the binary can find it:

• Windows: put onnxruntime.dll in the same directory as the built .exe
• macOS: set ORT_DYLIB_PATH=/path/to/libonnxruntime.dylib
• Linux: set LD_LIBRARY_PATH to include the directory, or install to /usr/local/lib

2. Build

# CLI and MCP server
cargo build --release --bin orunla_cli
cargo build --release --bin orunla_mcp

# Binaries are in target/release/

3. (Optional) Build the desktop app

npm ci
npm run tauri build

4. Run

The GliNER model (~80MB) downloads automatically from HuggingFace on first run.

# Copy ONNX runtime next to the binary (Windows example)
cp onnxruntime.dll target/release/

# Test it
./target/release/orunla_cli ingest "Hello world"

Documentation

• Overview -- architecture and design
• AI Setup -- connecting to Claude, ChatGPT, n8n, etc.
• MCP Guide -- MCP server configuration per IDE
• API Reference -- REST endpoints
• CLI Guide -- command-line usage
• Developer Guide -- building, tunnels, advanced config

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

Apache License 2.0. See LICENSE for details.