#OpenWorker #DesktopAgents #LocalFirstAI #Automation
OpenWorker is interesting not as another chatbot, but as the return of the agent to the desktop. The README states the position sharply: the agent should deliver finished work - a document, Slack reply, calendar change, or triaged inbox - rather than a task list for the user.
The architecture is local-first: a desktop app, a local Python agent server, your files, terminal, 25+ connectors, and the model you choose. You can bring keys for OpenAI, Anthropic, Google, or open-weight providers, or go local through Ollama. The important boundary is that data leaves the machine only through the selected models and integrations.
Running from source:
git clone https://github.com/andrewyng/openworker
cd openworker
bash packaging/setup_dev_env.sh
.venv/bin/openworker-server --cwd ~/some/project --port 8765
cd surfaces/gui
npm install
npm run dev
The main engineering detail is the approval gate. OpenWorker asks for confirmation before sends, calendar changes, and shell commands. That is exactly the level where an “agentic OS” stops being magic and becomes a local runtime with permissions, state, and receipts.
