What Google Jules does
Google Jules clones an authorized GitHub repository into a short-lived virtual machine, proposes a plan, changes files, runs checks, and returns work for review.
Google Jules is designed for asynchronous delegation rather than continuous pair programming. A developer chooses a GitHub repository and branch, describes a bounded task, and reviews a proposed plan before code changes begin. Jules then works in a short-lived cloud virtual machine, so the developer can leave the browser and return when a notification says the task has completed or needs help.
The no-cost Jules plan currently permits 15 tasks in a rolling 24-hour window and three concurrent tasks. Google AI Pro raises those limits to 100 and 15; Ultra to 300 and 60. Paid benefits are presently tied to individual Google AI subscriptions and Gmail accounts, not a general Workspace enterprise purchase. Every tier accesses the same core agent, while the paid tiers mainly raise capacity and access priority. Limits can change as Google evolves the service.
Reliability depends heavily on repository hygiene. Jules reads `AGENTS.md`, can use a validated setup script and saved environment snapshot, and works best when discrete lint and test commands expose success. Its Ubuntu VM has internet access, so committed secrets, unsafe dependencies, and unreviewed scripts are material risks. Google says private repositories are not used for model training and only explicitly authorized repositories are accessible, but users still need minimal GitHub scope, safe setup code, diff review, and branch protection.
How Google Jules works
Jules clones an explicitly authorized GitHub repository into a fresh Ubuntu virtual machine, reads project guidance, prepares dependencies, and drafts a plan for approval. After approval it edits files and runs discrete checks, then presents logs and changes for the developer to review and integrate.
Create an isolated task environment
Jules clones an explicitly authorized GitHub repository and chosen branch into a fresh short-lived Ubuntu virtual machine. A validated setup script, environment snapshot, README, and root AGENTS.md help it install dependencies and understand repository conventions.
Translate the issue into reviewable work
The agent reads code and non-code files, interprets the prompt and acceptance criteria, and proposes a sequence of changes. The developer reviews and approves that plan before Jules begins editing, which is the best point to narrow an unsafe or overly broad task.
Edit and check asynchronously
Jules changes files and runs discrete commands inside the internet-connected cloud VM, recording progress and retrying some transient failures. The developer may leave the application and return when notifications report completion, failure, or a request for intervention.
Verify before merging
Inspect task logs, dependencies, commands, and every changed line, then rerun lint, tests, and security checks in a trusted environment. Integrate through normal GitHub review and branch protection; isolation does not make generated code or downloaded packages safe.
How to set up Google Jules
Prepare a safe repository
Remove committed secrets, document build and test commands, add a current root AGENTS.md, and make sure a disposable branch can pass checks.
Connect only needed GitHub access
Sign in at Jules, accept the privacy notice, authorize GitHub, and select specific repositories instead of granting unnecessary organization-wide scope.
Validate the cloud environment
Add lightweight install, lint, and test commands to the setup script, run validation, and save a snapshot without long-running servers or watch processes.
Delegate one bounded task
Choose the repository and branch, state exact acceptance criteria and exclusions, then inspect and approve Jules's proposed plan before execution.
Review before integration
Read logs and the complete diff, rerun checks in a trusted environment, inspect dependency and security changes, and merge through normal branch protection.
Google Jules FAQs
Is Google Jules free?
Yes. The no-cost plan currently includes 15 tasks per rolling 24 hours and three concurrent tasks.
What do Google AI Pro and Ultra add?
Jules in Pro currently allows 100 daily and 15 concurrent tasks; Ultra allows 300 daily and 60 concurrent tasks, with higher access priority to newer models.
Can a Workspace account buy paid Jules limits?
Not currently through the documented self-serve route. Paid Jules benefits require an eligible individual Google AI subscription on a Gmail account.
Does Jules train on private repositories?
Google says Jules does not use private repositories to train models and can access only repositories explicitly authorized through GitHub.
Where does Jules run code?
Each task runs in a fresh, short-lived Ubuntu cloud virtual machine with internet access, cloned source, task-specific logs, and optional setup snapshots.
Listing reviewed 2026-07-15. Product details and pricing can change; verify important terms on the provider's website.
Related Coding AI tools
Related AI guides
Reviews
Tell the community what you made, what worked, and what you wish you knew before starting.