The Unofficial Guide to OpenWorker

Everything you need to master Andrew Ng's open-source AI agent that delivers finished work — setup guides, model comparisons, integration tips, and real-world workflows.

Read the Guide Setup Tutorial

What Is OpenWorker?

OpenWorker is an open-source AI agent built by Andrew Ng and Rohit Prasad that fundamentally changes how you interact with AI on your desktop. While most AI tools generate text in a chat window and leave you to copy-paste the results, OpenWorker works directly inside your files, apps, and services to produce finished deliverables. It can prepare a customer brief by pulling data from your CRM, draft a report by aggregating metrics from multiple dashboards, or triage a Slack alert by cross-referencing your runbook.

The architecture behind OpenWorker combines a Tauri 2 desktop shell with a local Python FastAPI server powered by the aisuite library. This means the agent runs natively on your Mac, processes requests locally, and only sends data to the LLM provider you choose. You bring your own API key — whether that is OpenAI, Anthropic, Google, DeepSeek, or a fully local Ollama instance — and OpenWorker handles the orchestration between your model and your connected tools.

What sets OpenWorker apart from other agent frameworks is its permission-based safety model. Every action the agent takes is classified by risk level, and consequential operations require your explicit approval before execution. This makes OpenWorker suitable for professional workflows where a misplaced message or overwritten file could cause real damage. The MIT-licensed codebase is fully auditable, giving teams and enterprises the transparency they need to deploy AI automation with confidence.

Key Features of OpenWorker

Six capabilities that make OpenWorker the most flexible open-source AI coworker available today.

Delivers Finished Work

OpenWorker goes beyond generating text. It carries tasks through from start to finish — producing polished documents, sending messages, updating calendar entries, and filing reports without you touching the keyboard.

Bring Your Own Model

Run OpenWorker with GPT, Claude, Gemini, DeepSeek, Kimi, GLM, Inkling, or any Ollama model. You choose the provider and keep full control over cost, speed, and data routing.

Local-First Privacy

Your files and data never leave your machine unless you explicitly send them through a chosen LLM provider. OpenWorker stores credentials locally and processes everything on your desktop.

Permission-Based Safety

A typed risk engine classifies every action as read, write_local, exec, or external. OpenWorker checks in before doing anything consequential, so you stay in charge of what happens.

Deep Integrations

OpenWorker connects to Slack, Gmail, Outlook, Google Calendar, Notion, HubSpot, GitHub, Jira, Linear, Asana, Google Drive, Dropbox, Box, and more through one-click or manual setups.

Event-Driven Triggers

Set a trigger once and OpenWorker runs tasks automatically. A Slack mention can triage an alert, an incoming email can draft a reply, and a calendar event can generate a report before standup.

How OpenWorker Works

Three phases from request to result. You describe the outcome, OpenWorker handles the execution, and you approve before anything goes live.

01

Describe the Outcome

Tell OpenWorker what you need — a customer brief, a reorganized calendar, or a weekly report. Frame it as the end result, not a step-by-step instruction.

02

OpenWorker Plans & Acts

The agent breaks your request into steps, selects the right tools and integrations, and executes each step using the AI model you configured.

03

Review & Approve

Before any consequential action — sending a message, modifying a file, or calling an external API — OpenWorker pauses and asks for your confirmation.

Supported Models in OpenWorker

OpenWorker supports 30+ AI models across cloud and local providers. Here are the most popular choices.

Model Provider Best For Cost
GPT-5.6 Sol OpenAI Strongest reasoning $$$$
Claude Fable Anthropic Nuanced writing $$$
Gemini 3.6 Google Multimodal tasks $$$
DeepSeek R2 DeepSeek Cost-effective coding $
Kimi K3 Moonshot Long context window $$
Ollama (Local) Self-hosted Full privacy, no API cost Free

See the full model comparison guide →

Real-World Use Cases for OpenWorker

Practical workflows where OpenWorker saves hours every week by turning multi-step processes into single requests.

Customer Brief Preparation

Ask OpenWorker to prepare for a renewal call. It pulls account history from your CRM, recent tickets from Linear, and last quarter's usage data, then assembles a one-page brief with talking points.

Calendar Optimization

Tell OpenWorker to untangle your calendar. It identifies overlapping meetings, suggests reschedule windows, drafts polite decline messages for low-priority invites, and consolidates scattered 1:1s into focused blocks.

Slack Alert Triage

Configure a Slack mention trigger so OpenWorker automatically checks the runbook, gathers relevant logs and metrics, and drafts an in-thread response for your review before posting.

Weekly Report Generation

Schedule OpenWorker to run before Monday standup. It aggregates commits from GitHub, completed tasks from Asana, and key metrics from your dashboard, then delivers a formatted PDF or Markdown report.

Tips for Getting the Most from OpenWorker

Practical advice from experienced users to help you build efficient workflows with OpenWorker.

Ask for Outcomes, Not Steps

Instead of telling OpenWorker to open a spreadsheet, copy data, and format it, simply say 'Summarize last month's sales by region in a table.' The agent figures out the steps.

Start with Read-Only Mode

When you first set up OpenWorker, use the most restrictive permission mode. Gradually expand permissions as you build trust with how the agent handles your data and tools.

Use Triggers for Recurring Tasks

Any task you do more than twice a week deserves a trigger. Set up email-based, schedule-based, or Slack-based triggers to let OpenWorker handle repetitive work automatically.

Pick the Right Model for the Job

Use GPT or Claude for complex writing tasks, DeepSeek for code-heavy workflows, and local Ollama models for sensitive documents that should never leave your machine.

Chain Tasks with Context

OpenWorker maintains context across a conversation. After it generates a report, you can say 'Now send the summary to the #team-updates Slack channel' without re-explaining the report.

Review the Execution Log

After every completed task, check the execution timeline. It shows exactly what OpenWorker read, wrote, and called. This audit trail helps you refine prompts and catch unintended actions early.

OpenWorker FAQ

What is OpenWorker?
OpenWorker is an open-source desktop AI agent created by Andrew Ng and Rohit Prasad. Unlike traditional chatbots, OpenWorker delivers finished work — polished documents, sent messages, updated calendars — by acting across your files and everyday tools.
Is OpenWorker free to use?
Yes. OpenWorker is released under the MIT license and the desktop app is completely free. You bring your own API key from any supported provider like OpenAI, Anthropic, or Google, and pay the model provider directly. Running a local model through Ollama costs nothing.
What platforms does OpenWorker support?
OpenWorker currently runs on macOS with a native desktop application built on Tauri 2. Windows support is actively in development and expected soon. Linux support may follow based on community demand.
Which AI models work with OpenWorker?
OpenWorker supports over 30 tool-calling models including GPT-5.6 Sol, Claude Fable, Gemini 3.6, DeepSeek R2, Kimi K3, GLM-5, Inkling, and fully local models through Ollama. You can switch models per task without restarting.
How does OpenWorker protect my data?
OpenWorker follows a local-first architecture. Your files, credentials, and work data stay on your machine. The only outbound traffic goes to the LLM provider and integrations you explicitly connect. API keys and tokens are stored locally, never on remote servers.
What integrations does OpenWorker support?
OpenWorker integrates with Slack, Gmail, Outlook, Google Calendar, Notion, HubSpot, GitHub, Attio, Google Drive, Jira, Linear, Asana, Dropbox, Box, and more. Connections are set up through one-click OAuth or manual configuration.
How is OpenWorker different from Claude Cowork?
OpenWorker is open-source, model-independent, and runs entirely on your local machine. Claude Cowork is a closed-source Anthropic product that only works with Claude models. OpenWorker lets you choose any provider and keeps your data local by default.
Can OpenWorker run tasks automatically?
Yes. OpenWorker supports event-driven triggers. You can set it to activate on a Slack mention, incoming email, or calendar schedule. It runs the task in the background and checks in before taking any consequential action.
What is the permission system in OpenWorker?
OpenWorker uses a typed risk engine that classifies actions into four categories: read, write_local, exec, and external. Five permission modes let you control how much autonomy the agent has, from fully supervised to more autonomous operation.
How do I get started with OpenWorker?
Download OpenWorker from the official site, install it on your Mac, add an API key from any supported provider, connect your first integration, and describe an outcome. The entire setup takes under 10 minutes. Visit our setup guide for detailed steps.

Start Using OpenWorker Today

Download the desktop app, connect your preferred AI model, and let OpenWorker handle the tasks that eat your day. Open-source, private, and always under your control.

Setup Guide Compare Models