Leverler sits in your system tray and watches for triggers — keywords, emails, anything — then launches local AI agents automatically. No cloud. No API keys. No data leaves your machine.
What is it
Leverler is an Electron desktop app that runs silently in your system tray. Define triggers. Assign agents. Walk away.
Runs entirely on your machine via Ollama. Your clipboard, your emails, your data — never touches a cloud server.
Watches clipboard for keywords and polls your inbox for matching emails. Agents fire automatically when triggers match.
MIT licensed. Fork it, extend it, add your own trigger types and agent templates. The whole stack is yours.
Installation
Leverler requires Node.js 18+ and Ollama. Both are free and install in minutes.
Download Ollama — the local model runtime that powers Leverler's agents.
Download Ollama →We recommend qwen2.5:7b — fast, smart, and runs well on Apple Silicon and modern Windows machines.
Clone the repo and install dependencies. Node.js 18 or higher required.
Leverler opens your dashboard and drops into your system tray. Hit Start Listening and define your first trigger.
# Start Ollama (if not running)
ollama serve# Pull the recommended model (~4 GB)
ollama pull qwen2.5:7b
# Or the smarter 14B variant
ollama pull qwen2.5:14bgit clone https://github.com/leveleragentic/leverler
cd leverler
npm installnpm start
# Opens dashboard + drops to system tray
# Go to Settings → Test Connection
# Then add your first triggerHow it works
Leverler is built around a simple loop. You define the conditions; it handles the rest.
Runs in the background watching your clipboard and inbox continuously. Zero CPU at idle.
A trigger fires when clipboard text matches a keyword, or a new email hits your inbox with matching content.
An agent spins up with the trigger context. Qwen2.5 processes the task locally, streaming output in real time.
Results appear in the dashboard. Review output, inspect logs, or chain the result into another trigger.
Supported models
Configure the model in Settings. Switch any time without restarting.
| Model | RAM | Speed | Best for |
|---|---|---|---|
| qwen2.5:7b recommended | ~5 GB | Fast | Daily use, email, summarization |
| qwen2.5:14b | ~9 GB | Medium | Complex reasoning, research agents |
| qwen2.5:3b | ~2 GB | Very fast | Low-resource machines |
| llama3.2:3b alt | ~2 GB | Fast | Lightweight alternative |
| mistral:7b alt | ~4 GB | Fast | Strong instruction following |
Open source
Leverler is MIT licensed. Add trigger types, build custom agent templates, integrate with any local model. Pull requests welcome.