OpenWorker
Open-source desktop AI assistant that aims to finish everyday tasks, with approval gates and a sandbox option.
- Stars
- 18.9k
- Forks
- 2.7k
- Licence
- MIT
- Language
- Python
- Last update
- 4 days ago
- Type
- App
What it is
OpenWorker is a desktop app that runs an AI agent on your own machine to carry out tasks such as reviewing code for security issues, drafting documents, or triaging an inbox, rather than only chatting. You bring your own API key for a model provider, or run local models through Ollama, and it can connect to tools like GitHub, Slack, and Google Calendar. The README describes it as being in open beta.
- Scan a codebase and its dependencies for security risk, with proposed fixes re-scanned before you approve them
- Prepare a customer call brief from your CRM and inbox
- Draft a report or spreadsheet that lands as a file you can open
- Set up a morning brief or weekly report on a schedule
- Mention @OpenWorker in a Slack channel to start a session and get a thread reply
Great for
- Developers who want an AI agent to review code for security risks with approval before fixes are applied
- People who want to use their own API keys and choose between many model providers
- Users who want local models through Ollama and keep data on their machine
- Teams wanting scheduled automations such as recurring reports or channel watches
Not ideal for
- People who need a fully polished, stable product: the README calls it an open beta with rough edges
- Windows users who want a warning-free install: the README says the Windows build is not yet code-signed, so SmartScreen will warn
- Anyone who wants to hand the agent unsupervised control of consequential actions: the README says actions are approval-gated by default and reviewer verdicts are judgments, not guarantees
How to get it
- 1
Download the app
You'll need: macOS 12+ (Apple Silicon) or Windows 10/11 (x64) for the desktop app · A model API key, or Ollama for local models
Clone the repository
git clone https://github.com/andrewyng/openworker - 2
Run it
- Download the macOS or Windows app from the Download section of the README
- Open the app and add a model key, or point it at Ollama
- Go to Settings → Models & Keys to connect a provider
- Ask for a specific task and review the approval prompts
- 3
Get the most out of it
- Turn on the NVIDIA OpenShell sandbox so agent commands run in isolated containers, per the README's sandbox docs
- Use the macOS built-in sandbox, which the README says needs nothing extra to install
- Connect integrations such as GitHub, Slack, Jira, or Google Calendar for real workflows
- Set up scheduled automations for recurring work like a morning brief
- Add MCP-compatible tools with per-tool control
Downloads
Strengths
- Runs on your machine and supports many model providers, including fully local models via Ollama
- Actions are approval-gated by default, and every tool call is recorded in an audit trail
- Optional sandbox runs agent commands and file tools in an isolated container or sandbox
Trade-offs
- Described as open beta, so rough edges are expected
- The Windows installer is not yet code-signed and triggers a SmartScreen warning
- Model and integration use depends on your own API keys and any associated costs, which the README does not detail
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.