Sorted by what needs you
Home lists the approvals and questions from every running task. Allow or deny a command right there, or open the task that asked.
Hand off the task, then review the work.
Agent V is a desktop coding agent. Write what you want and how you will know it is done; it does the work and shows you every step.
Works in your repository with a terminal, a browser and git.
Checks its own result against your Done when list before it finishes.
Runs on the models you choose
- OpenAI
- Anthropic
- Gemini
- Ollama
- DeepSeek
- Groq
- OpenRouter
- xAI
- Mistral
- OpenCode Go
- Bedrock
- Vertex AI
Built to be handed work
Each task gets a brief, a plan, and a record of what it read, changed and ran.
See what it cost
Usage shows tasks, tokens and estimated spend per day, how much changed work you kept, which models did the work, which tools failed, and which edits never had a passing check.
Checks its own work
Write what done looks like. The agent marks each line met or not met with the evidence it saw, and the receipt shows the time, tokens and cost.
Uses the model you trust
Eleven providers on your own keys or cloud credentials, any OpenAI-compatible endpoint including Azure OpenAI, or local models through Ollama. Pick the mode and the effort for each task.
From brief to merged branch,
you only step in where it counts.
01
Write the brief
Say what you want and how you will know it is done. Agent mode plans, edits files and runs commands; Ask mode reads and answers.
02
It does the work
It reads code, edits files, runs your commands and tests, and drives a browser when the job needs one. Bigger jobs go to instances that report back to the task that started them; an instance that writes gets its own worktree branch.
03
You review it
Changes shows every file the task touched. Keep it, undo it, or rewind to before any instruction. Edits you made since are left alone.
04
Ship it
Commit and open the pull request from the task, then watch its checks in the PR tab. A failing check goes back to the agent in one click.
Connect it once,
use it in every task
- GitHub pull requests, reviews and issues from the task
- A real browser it drives to check its own UI work
- Screenshots and images it can read, to see what it built
- New tools it writes itself when a job needs one
Recently
shipped
1.4.0Runs on your machine,
with the keys and limits
you set
- Open source under GPL-3.0-or-later
- Your own model keys, or local models through Ollama
- You choose which actions wait for your approval
- Each task can run in its own git worktree
- Every file edit is checkpointed and can be rewound
- The code index runs on your machine, and dictation can too
Stop putting it off.
Hand it to Agent V.
A desktop coding agent that works in your repository and checks its own result.
Bring your own model keys, or run local models through Ollama.
Read the source- Windows · x64 installer
- macOS · Apple silicon and Intel
- Linux · AppImage, deb, rpm