Jules
NewGoogle's autonomous coding agent that clones your GitHub repo into a cloud VM, drafts a plan for approval, makes the changes and opens a pull request, with scheduled tasks, a CLI and a REST API.
Jules
Explore features, practical uses and pricing below.
Jules is an autonomous coding agent from Google Labs that works on your GitHub repositories in the cloud. You pick a repository and branch and describe the task. Jules clones the code into a virtual machine, drafts a plan, makes the changes and shows you a diff, and when you're satisfied it opens a pull request. Because the work runs on Google's VMs, you can close the tab and come back when a notification says the task is done.
From prompt to pull request
The getting started guide sets out the flow. Sign in with a Google account, connect GitHub and choose which repositories Jules may access. Write a specific prompt and click "Give me a plan." Jules produces a plan you can review and approve before any code changes. After approval it works in a short-lived Ubuntu VM, installing dependencies, editing files and running tests, and logs each step in an activity feed with mini diffs. You can also assign work from GitHub by adding the "jules" label to an issue.
Getting the VM environment right
Google's FAQ names broken setup scripts as a common cause of failed tasks. The environment docs list what each VM ships with, including Node.js, Bun, Python, Go, Java and Rust. For simple projects Jules works out the setup from your README or AGENTS.md, which it looks for in the repository root. For anything more involved, add a setup script with explicit install and test commands. Long-running processes such as npm run dev aren't supported in setup scripts, so a task that needs a live dev server to verify itself will struggle.
Scheduled tasks, suggestions, CLI and API
- Scheduled tasks repeat a prompt on a cadence you set, which suits dependency bumps or routine checks. Editing a scheduled task isn't supported yet, so you delete and recreate it.
- Suggested tasks are experimental. With Proactivity switched on for a repository, Jules scans for #TODO comments and proposes fixes. The docs cap Proactivity at five repositories at launch.
- Jules Tools is a CLI installed with npm install -g @google/jules. It lists repos and sessions, starts remote sessions, including parallel ones, and pulls results back to your machine.
- The REST API is in alpha. It lets you create sessions, approve plans and message the agent from your own tools.
Plans, models and eligibility
The usage and limits page lists three tiers. The base Jules plan is free, allows 15 tasks per rolling 24 hours with 3 running at once, and uses Gemini 2.5 Pro. Jules in Pro raises that to 100 daily and 15 concurrent tasks, and Jules in Ultra to 300 and 60, both with access to newer models starting with Gemini 3 Pro. The paid tiers come with Google AI Pro and Google AI Ultra subscriptions, which for now are limited to individual @gmail.com accounts. Google Workspace users don't have an upgrade path yet, and you must be 18 or older.
Good fits and poor fits
Jules suits well-defined chores in GitHub repositories: version bumps, test coverage, small bug fixes and documentation you'd rather review than write. It's a poor fit for code hosted elsewhere or work that needs a running dev server to check. Google's FAQ says Jules does not train on private repository content. If you want an agent that also works in your editor, GitHub Copilot and OpenAI Codex offer local options alongside cloud tasks.
Example: a framework upgrade
Connect a Next.js repository, choose main and write: "Upgrade next to the latest 15.x release, migrate any deprecated config and make sure npm test passes." Add a setup script that runs npm ci. Read the plan closely, especially the files it intends to change, then approve it. Watch the activity feed for test failures, review the full diff and publish the branch for your usual CI and code review.
Things people want to know about Jules
Which languages work best?
The FAQ says Jules is language agnostic but works best with JavaScript or TypeScript, Python, Go, Java and Rust.
What happens if a task fails?
Jules retries automatically, then marks the task as failed and notifies you. Broken setup scripts and vague prompts are the usual causes.