Rover v1.7: Interactive iterations and multi-agent parallel tasks

Rover is a manager that runs on top of AI coding agents such as Claude Code, Codex, Gemini, Qwen, and Cursor. You can define multiple tasks and assign them to your preferred coding agents, each working in the background autonomously in isolated environments while you focus on higher-priority work.

You can learn more about Rover here.

These are the major changes in version 1.7:

• Full task context during iterations: When running rover iterate, agents now receive the complete context of the task, including the original request and all iteration changes. Previously, the reduced context sometimes caused unrelated or low-quality updates.
• Interactive iterations for quick adjustments: Use rover iterate ID --interactive to open an interactive session with an agent inside a sandbox. It’s perfect for applying small fixes, polishing details, or adding last-minute adjustments before merging a task.
• Complete the same task with multiple agents: Run a task through several agents simultaneously by passing the --agent flag multiple times. This allows you to compare different approaches and pick the solution that best fits your needs. While you may not want to do this for every task, it comes really handy when you are evaluating whether to add a new agent to your toolbox or simply want to explore a second (or third or fourth!) opinion for tricky issues.

You can update Rover to this new version using the following command:

npm install -g @endorhq/rover@latest

Iterating over tasks

Rover uses multi-step workflows to complete tasks. These workflows help agents produce consistent and reliable results, but they also take time since agents must go through all individual steps sequentially.

Often, agents finish 90-95% of your request, leaving some minor follow-up changes. For example, a small refactor, a missing requirement, or a new idea you came up with. Running a full workflow process might be unnecessary, as agents can complete these little tasks in a single prompt.

With v1.7, you can now use the --interactive, -i flag in rover iterate to skip the full workflow and open a direct interactive session with your selected agent. Rover will prepare a sandbox with the current task context so you can add new instructions or requirements.

rover iterate TASK_ID --interactive

For more complex changes, you can still trigger the full workflow by omitting the flag. This release also improves how Rover passes context to agents, ensuring they always receive the full task history (including previous iterations) resulting in more accurate changes.

Comparing agents

Coding agents from multiple vendors behave differently even when following the same workflow. We found that coding agents have their own strengths and weaknesses, or simply their own coding style.

Claude tends to adhere better to existing implementations, while Codex often seeks more optimal solutions. Depending on the task, either approach might work better for a specific agent.

Rover lets you take advantage of all of them. By using the --agent flag multiple times in rover task, Rover spawns one task per agent so you can compare their solutions.

rover task --agent codex --agent claude 

Each agent gets its own sandbox environment and workspace. Once completed, their task summaries make it easy to understand how each agent approached the same problem.

Next steps

We’re getting feedback continuously of how developers integrate Rover into their development process. While the .rover directory and rover.json configuration file keep all the information in the project, they introduce a commitment: you must adopt Rover files in your project from the very beginning.

We’re exploring ways to make this more flexible so you can use Rover without committing configuration files into your repository.

We’ve also identified opportunities to expand Rover’s workflows. While we continue adding new built-in workflows, this approach doesn’t scale indefinitely. It’s time to give you the power to tailor Rover to your needs. In upcoming releases, we’ll add support for creating and loading custom workflows.

Join our community to stay connected and share your feedback:

• GitHub: github.com/endorhq/rover
• Social: Follow us on X, Mastodon, or Bluesky to stay up to date and share your experience
• Discord: Join the discussion in our Spaceship server!

Happy coding with Rover! 🚀