Google Jules is Google’s asynchronous coding agent, developed by the company’s labs area. The core difference is the execution model: instead of running in your terminal, each task runs on a fresh virtual machine in the cloud from the repository and returns a pull request for review.
What it’s for
- Dispatching coding tasks and walking away; the work continues in the cloud
- Running several tasks in parallel, according to the plan limit
- Generating reviewable plans, diffs and pull requests from a description
- Fixing continuous integration failures and following review feedback
How it works
- You start the task from the web interface, a label on a repository issue, the command line or the API
- Each task becomes a session with recorded, auditable activities
- Execution clones the repository, installs dependencies and runs build and tests
- The result arrives as a pull request, which can receive fixes from feedback
Where it runs
- The product’s web interface
- Virtual machines in Google’s cloud
- Its own command line
- REST API for integration with other flows
Availability and licensing
- Proprietary product, with no open repository for the agent
- Access follows plans with different daily task and concurrency limits
- Repository integration requires per-repository authorization
How it differs
- It is asynchronous and cloud-based, not a local terminal agent
- It is pull-request oriented, which brings it close to the code review flow
- It is a separate product from Gemini CLI and Google Antigravity, even though it shares the Google origin
Things to note
- It is not meant for long-running processes, such as a development server
- The agent changes repositories; reviewing the generated pull requests is part of the flow
- Limits and available models may vary by plan