Rove Terminal Agent Workspace
Run coding tasks in parallel across isolated Git worktrees, with sessions that survive terminal disconnects.
- Source repo
- Sma1lboy/rove
- Stars
- ★ 166
- Last updated
- today
- License
- MIT
- Primary language
- TypeScript
- FA score
- 69/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Software engineers handling a refactor, bug fix, and code review at the same time can put each task in its own worktree and branch.
- Not a fit if
- Teams that do not use command-line coding agents
- Windows users unable to install Git for Windows/Git Bash
- Source review
- 69/100 · Some gaps
What does this agent do, and when should you use it?
Rove is a terminal-native workspace for running multiple coding tasks in parallel. Each managed task gets a Git worktree and branch, with terminal tabs for Claude Code, Codex, Copilot, Kimi, other command-line agents, or a CLI you register. You can manage work through the TUI or `rove api` to inspect output and diffs, send follow-ups, and land branches. Agent and shell sessions run on the host, so they continue after you quit the TUI or disconnect from SSH and can be reattached later. Rove requires Bun, Git, and at least one supported agent CLI; it runs on macOS, Linux, and Windows, with Node.js and Git for Windows/Git Bash also required on Windows.
Start rove in a repository and create a task; Rove sets up a Git worktree and branch for managed tasks, then launches the chosen agent CLI or a shell tab. Tabs within a task share its files. In the TUI, users can switch tasks, read session output, inspect worktree diffs, run checks, and send follow-up prompts. Scripts and coding agents can use rove api add, list, read-output, send, and land to create tasks, inspect output, send instructions, and land branches. Sessions remain on the host after the terminal disconnects and can be reattached later.
- Software engineers handling a refactor, bug fix, and code review at the same time can put each task in its own worktree and branch.
- Developers working over SSH or in terminals that may close can leave agent sessions running on the host and reconnect later.
- Developers comparing Claude Code, Codex, Copilot, or other CLI agents in one terminal workspace can choose an engine for each task.
- Scripts or coding agents orchestrating development work can use
rove apito create tasks, read output, send follow-ups, and land branches. - Developers working on a VPS, devbox, or narrow mobile SSH session can manage parallel tasks through the terminal interface.
How do you install or deploy this agent?
The install script sets up the required Bun runtime. Rove can also be installed or run through npm, Bun, or npx.
curl -fsSL https://rove.run/install.sh | shnpm install -g @sma1lboy/rovebun install -g @sma1lboy/rovenpx @sma1lboy/roveRove requires Bun ≥ 1.3.11, Git, and at least one supported agent CLI on PATH. Windows also requires Node.js and Git for Windows/Git Bash. If Bun is installed in an unusual location, set ROVE_BUN to its path.
How do you use this agent?
Change into a Git repository, launch Rove, and press n to create your first task:
cd your-repo
roveTo let a coding agent operate Rove, install the companion skill:
rove skill installScripts can use rove api to create tasks, read output, send follow-ups, and land branches:
rove api add --repo "$PWD" --prompt "Fix the flaky auth test."
rove api list
rove api read-output --task-id <id>
rove api send --task-id <id> --prompt "Run the integration suite too."
rove api land --task-id <id>What are this agent's strengths and limitations?
- Managed tasks get separate Git worktrees and branches, reducing file conflicts between parallel tasks.
- Works with multiple command-line agents and can use registered CLIs or plugin-provided engines.
- Sessions keep running on the host after the TUI exits or SSH disconnects, and can be reattached.
- Offers both a TUI and
rove apifor developers, scripts, and coding agents. - Includes diff inspection, inline review notes, branch landing, and task notifications.
- Requires Bun ≥ 1.3.11, Git, and at least one supported agent CLI; Rove does not provide a coding model itself.
- Windows additionally requires Node.js and Git for Windows/Git Bash.
- Each agent uses its own authentication, permissions, and models; the source does not state their service costs or account requirements.
- Worktree isolation is per managed task; tabs in one task share files, and project-main or
rove .directory tasks reuse an existing directory.
How does this agent compare with similar options?
The README positions Rove as a terminal-native workspace and distinguishes managed tasks, which use separate Git worktrees and branches, from running multiple agents in a single directory. It does not provide feature or pricing comparisons with specific products.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Rove Terminal Agent Workspace This agent | 69 · Some gaps | CLI | ★ 166 | today | TypeScript | Codex · Claude Code |
| Runner Multi-Agent Terminal | 82 · Some gaps | Desktop appFree | ★ 275 | today | Rust | Codex · Claude Code |
| zerostack | 57 · Major gaps | CLIFree + model costs | ★ 1.7k | today | Rust | OpenAI API · Claude API |
| Loop Engineering | 69 · Some gaps | CLIFree + model costs | ★ 11k | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The README says tasks use separate worktrees and branches and that each CLI retains its own authentication and permissions, which supports some isolation and data-flow transparency. The product still runs arbitrary registered CLIs, shells, and plugins on the host. The supplied material does not establish permission sandboxing, detailed sensitive-credential handling, or confirmation for external effects, so least privilege, sensitive-data handling, and external-effects scores remain partial. The README mentions diff review, landing branches, and lifecycle recovery, but does not fully explain rollback after mistakes. Repository files and the license provide attribution clues; publisher provenance is unverified and maintenance responsibility is unclear, so no identity-based credit is warranted.
The README, package scripts, and CI configuration are broadly consistent about Bun, Git, additional Windows requirements, and build, typecheck, and test paths. CI describes multi-platform gates and a diagnostic command, supporting ordinary availability and failure handling. The supplied material does not include the full implementation or dependency-availability commitments, and static configuration cannot establish actual run results, so these scores are not full.
The README clearly targets terminal users who want parallel coding-agent work and covers SSH, laptops, devboxes, VPS, scripts/API, plugins, and multiple CLIs. It explains that tabs within a task share files, that directory tasks reuse an existing directory, and lists Bun, Git, and Windows requirements. Plugins and arbitrary CLI registration broaden use cases, but permission boundaries, trigger policy, and constraints across all scenarios are not fully specified.
The README is organized around quick start, installation, operation, API, plugins, diagnostics, and development, with command examples and documentation links. It states prerequisites, supported systems, the MIT license, and a changelog path. CI and Changesets describe a release process, supporting strong scores. The supplied files do not provide a complete FAQ, limitation list, or clear maintenance owner; aliases and multiple installation routes add some naming and entry-point complexity.
The task list, embedded sessions, diff review, scripting API, and plugin catalog address concrete needs in parallel-agent work. The README's workflows and commands make the intended value clear. Cost-benefit is only partially established: the material does not quantify resource use, deployment and maintenance costs, or comparative benefits, and multiple persistent sessions may consume host resources.
README claims can be mapped to installation commands, API examples, CI workflows, test files, and a changelog path; CI also names specific check categories, giving limited cross-file corroboration. The implementation, complete documentation, and test results are not supplied, so many behaviors remain repository-text assertions. This review treats documentation and static configuration as evidence and does not infer that they were executed or passed.
- Rove runs coding agents, shells, and plugins on the host; review their permissions, authentication, and data access before use.
- The README describes worktree isolation, but terminal tabs within one task share files. Use separate tasks when stronger isolation is needed.
- This assessment uses only the supplied static files and does not confirm that the described features, CI, or tests actually ran successfully at this revision.
FAQ
Do I need to pay for a model or service to use Rove?
Are files isolated between agent tasks?
rove . directory tasks reuse an existing directory.Will sessions stop if I close my terminal or disconnect from SSH?
Can I manage tasks with scripts instead of the TUI?
rove api supports adding and listing tasks, reading output, sending follow-up prompts, and landing branches. It also covers task inspection, notifications, and other task operations.