Desktop App
DJL
An open-source, local-first desktop AI agent for macOS and Windows that plans work, runs tools visibly, and pauses for human approval.
- Role
- Full-Stack Engineer
- Timeline
Overview
DJL is an open-source desktop agent that runs on the user's own machine. It plans multi-step work, executes tools with every step visible, routes between cloud and local models, and pauses for human approval before anything consequential happens. Built by a two-person team, it ships for macOS and Windows through an automated release pipeline so every operator receives signed updates without manual installs.
Outcomes
- Co-built the agent's core loop — planning, visible tool execution, and approval gates — as one of two engineers on an open-source codebase.
- Implemented multi-model routing so the agent switches between cloud and local models without changing the user's workflow.
- Built the automated release pipeline that packages, validates, and publishes signed macOS and Windows builds from a single workflow.
Stack
- TypeScript
- React
- Electron
- Node.js
- Local AI Models