At a Glance:
OpenWorker is a local-first, open-source AI coworker for macOS and Windows that produces files, works across 25+ integrations and MCP tools, supports hosted or local models, and approval-gates consequential actions.
Overview:
OpenWorker is a desktop AI coworker for completing everyday tasks and returning finished documents, spreadsheets, reports, web pages, or app updates instead of only chat responses. Its local agent works across desktop files, the terminal, and connected applications, with support for provider API keys or fully local Ollama models. Users can launch work from the app or Slack, schedule recurring runs, and approve writes, messages, calendar changes, and shell commands before execution. The project is currently in open beta.
Key Decision Points:
Supported desktops: Downloads target Apple Silicon Macs running macOS 12 or later and x64 systems running Windows 10 or 11; Windows builds are not yet code-signed and may trigger SmartScreen warnings.
Local-first architecture: The agent loop, conversations, connector tokens, and model keys reside on the user's machine; a small cloud service is used only to broker OAuth handshakes for connectors.
Model choice: Users provide keys for supported hosted providers, can enter other model strings at their own risk, or run fully local open-weight models through Ollama.
Approval boundaries: Sending messages, changing calendars, writing through tools, and executing shell commands require approval; unattended requests wait in an inbox.
Integration scope: The README lists 25+ connectors, local files, terminal access, and additional tools reachable through MCP with per-tool control.
Core Features:
Finished deliverables: OpenWorker creates documents, spreadsheets, reports, and web pages as files that can be opened and shared.
Cross-application task execution: The agent can work across local files, the terminal, GitHub, Slack, Jira, Notion, Linear, HubSpot, Outlook, monday.com, Gmail, Google Calendar, and other documented connectors.
Slack-triggered sessions: Mentioning
@OpenWorkeropens a desktop session and returns the result to the originating Slack thread.Scheduled automations: Recurring briefs, reports, and channel watches run on a schedule and retain full transcripts in the app.
Approval-gated actions: Consequential writes, sends, calendar changes, and shell commands pause for user approval or redirection.
Provider and MCP support: The aisuite-based agent engine supports multiple LLM providers, local Ollama models, tools, toolkits, and MCP connections.
Use Cases:
Prepare customer briefs, reports, documents, spreadsheets, or web pages from information available to the desktop agent.
Triage an inbox or reorganize a calendar while reviewing consequential actions before they execute.
Check release status across connected services such as Jira and GitHub, then return a finished summary.
Schedule a morning brief, weekly report, or standing watch over a connected channel.
Open-Source Alternative Value:
OpenWorker provides an MIT-licensed implementation of a local-first desktop AI coworker. The repository includes its Python agent server, React and Tauri desktop interface, MCP client, connectors, memory, automations, speech-to-text sidecar, and packaging tools. Developers can run the standalone server with a browser UI or launch the full desktop shell from source, while users can select hosted model providers with their own keys or use local Ollama models.
