Google Is Shutting Down Jules on Dec 31: Key Dates and How to Export Your Coding Agent Data

Google is shutting down Jules, the asynchronous AI coding agent it first showed in December 2024. In a notice dated October 29, 2026, the Jules team said the service will wind down on December 31, 2026, as part of an effort to "unify our agentic coding efforts." Users can keep finishing sessions until November 11, 2026. After that, Jules turns read-only until the end of the year, and then it is gone.

Code that Jules wrote is safe, because it already lives in GitHub as branches and pull requests. What is at risk is everything Jules kept on its own side: settings, environment variables, scheduled tasks, repository memories, setup scripts and session history. Google has published an export method for that data, and the useful window for it is the next two weeks, not late December.

Key facts at a glance

What Jules was, and why it is ending

Jules worked differently from an autocomplete tool in an editor. You gave it a task against a GitHub repository, it cloned the code into a cloud environment, made a plan, did the work in the background and handed back a pull request. It entered public beta in May 2025 and left beta on August 6, 2025, when Google said it had already produced more than 140,000 public commits. Through 2026 it gained MCP support, automatic fixes for failing CI runs and the Gemini 3.1 Pro model, the last of which arrived in a changelog entry on March 9.

Google frames the shutdown as consolidation rather than failure. The notice calls Jules an experiment in how asynchronous agents could help developers clear their backlog, and says the lessons are shaping products across Google, including its agent-first platform, Antigravity. In a post from the Jules account, the team said it is bringing those learnings into Antigravity so developers get "autonomous remote background task execution and an agent-first development environment" in one place.

In an email to users, Google Labs added that Google AI Pro and Ultra subscribers will no longer have Jules in their plan after December 31, while their other subscription benefits stay the same. Google has not announced refunds or a dedicated migration tool.

The three dates that matter

1. Now to November 11: Jules works normally. Finish or close any sessions you still care about, and run the export while every feature still works.
2. November 11 to December 31: Jules is read-only. You should still be able to look at past work, but you cannot start new tasks.
3. December 31: the service shuts down. Anything you did not export is lost.

If you rely on Jules scheduled tasks, such as a weekly dependency bump or a nightly test fix, note that they stop being useful on November 11. Move them before then rather than discovering a silent gap later.

How to export your Jules data

Google's export method is unusual. Instead of a download button, it gives you a command-line tool and a long prompt, and expects another coding agent to do the work. The steps from the official notice are:

1. Install Jules Tools: npm install -g @google/jules.
2. Sign in yourself: jules login. This opens a browser sign-in, so run it by hand rather than asking an agent to do it.
3. Connect GitHub if you used repositories. If the export fails with a 401 error asking about GitHub, connect your account to the Google Labs Jules GitHub app and try again. If you never used Jules with repositories, you can skip this.
4. Run the export prompt. Open a CLI coding agent from your home directory. Google names Claude Code as an example. Paste the export prompt from the notice. The agent checks your setup, saves and runs a Python script, and reports what it saved.

The script writes everything to a jules-export folder in your home directory:

Pitfalls to avoid during the export

Google's own notes flag several problems people have already hit:

• Secrets in plain text. env-variables.json contains your real environment variable values. Keep the folder private, never commit it to a repository, and consider rotating any credentials you stored in Jules once you have moved them.
• Short-lived login. The CLI access token expires about an hour after sign-in. If every call starts returning 401, run any jules command or sign in again.
• Undocumented endpoints. Settings, environment variables, repositories and memories come from an internal endpoint, and sessions from REST endpoints the CLI uses. Google calls both undocumented, which is another reason to export soon: if they change or start returning errors near the end of the year, there may be no fix.
• Missing repository data. If GitHub is not connected, the export runs but contains no repository details.
• Hidden setup scripts. The repository list does not include setup scripts or test commands; the script recovers them from each session's code output. Empty memory files simply mean there were none.

Where to move your workflows

Google points users to Antigravity, its platform for agent-based development. It includes a desktop app for running several agents in parallel, a command-line tool with background subagents, an IDE and an SDK. Google lists a $0 individual plan with basic weekly rate limits, higher limits for Google AI Pro and Ultra subscribers, and business seats starting at $30 per month. It runs on macOS, Windows and Linux.

Antigravity is not the only option. Several command-line and cloud coding agents now run long tasks in the background and open pull requests. Before you choose, compare a few practical points:

• Where the agent runs: on your machine, in the vendor's cloud, or in your own CI.
• Scheduling: whether it can repeat tasks on a timetable, as Jules could.
• Repository instructions: how it reads project rules and setup steps, so your exported memories and scripts can be reused.
• Secrets handling: where environment variables are stored and who can read them.
• Pricing limits: weekly caps and per-seat costs, which vary widely between plans.

What to keep from the export

The most valuable parts of the export are not the logs. They are the memories, setup scripts and test commands that capture what Jules learned about each repository. Copy them into files inside the repositories themselves, such as a contributor guide, an agent instructions file or a setup script under version control. That way the next agent, whichever one you choose, starts with the same knowledge, and you do not lose it again if that tool changes.

Session patches and plans are worth keeping as a record, especially for changes that were never merged. Review the patch-N.diff files for unfinished work before you archive the folder.

Takeaways

• Jules stops taking new work on November 11, 2026 and shuts down on December 31, 2026.
• Code and pull requests in GitHub are unaffected; settings, memories, scheduled tasks and history are not.
• Export early with Jules Tools and the official prompt, and keep env-variables.json private.
• Move scheduled tasks before November 11 so nothing silently stops.
• Store repository knowledge in your own version-controlled files, whatever agent you pick next.