A review-first desktop workspace for seventeen coding-agent CLIs. Every task gets its own Git worktree, a separate branch and checkout, so parallel agents never collide. See which runs need you, then review the diff before you merge.
A working model of the Daintree interface, running on sample project data.
Delegate. Monitor. Review. Merge.
That loop is the whole product, and it is the same loop whichever agent CLI you run. Generation is fast; review is the bottleneck.
- 01
Delegate
Point Daintree at a repo and create a worktree per task, by hand or straight from a GitHub issue. Each gets its own branch and checkout.
Creating worktrees - 02
Monitor
Launch any mix of agents across any number of projects. Pilot sorts every run by what is blocked, waiting and ready, so you check one list instead of twenty panels.
Pilot - 03
Review
An agent reporting done is a claim. Read the diff, check merge readiness, and read CI state from your connected forge before you decide anything.
Review Hub - 04
Merge
Commit, push and merge from inside Daintree, or hand the branch to your normal review process. The worktree cleans up behind you.
Code forges
| Agentic IDE | Daintree |
|---|---|
| An editor with agents added | A review layer above agent CLIs |
| One workspace, one checkout | One isolated worktree per task |
| Generation happens in the editor | Attention routing and evidence across runs |
| The vendor’s agent, the vendor’s way | The real CLI, your config, your subscription |
Daintree is not another VS Code fork. It sits above the agent CLIs you already have, and it is built for delegating work rather than for typing it.
Works with the CLI you already run
- Claude Code
- Codex CLI
- Antigravity
- OpenCode
- Cursor Agent
- GitHub Copilot
Plus eleven more built in, and any other agent a plugin adds. Daintree launches the real binaries, so your config and your existing subscription carry over, and live state detection works the same way for every built-in agent. See all 17 built-in agents
Pilot
Twenty agents running. Three of them need you.
Checking twenty sessions one by one becomes the job. Pilot puts every run across every project in one view, grouped by project and ordered by what is blocked, what is waiting on an approval, and what is ready to read. Park anything that can wait and it drops out of the list until you want it back.
The ordering is computed from live agent state, and nothing about the view is saved. It opens on what is true now, not on a filter you set last week and forgot.
Pilot documentationAn illustration of the Pilot view: three projects, six agent runs. Two runs in daintree-website are blocked, one waiting on an approval and one on a question. Two runs in api-gateway are working. In mobile-client one run is blocked on a rate limit and one is parked.
Evidence before merge
Done is a claim. The diff is the evidence.
An agent that says it finished has told you what it believes. Review Hub puts what actually changed next to whether it can land: the full diff per worktree, merge readiness against your base branch, and the check state your code forge is reporting right now.
Read it, stage what you want, and commit and push without leaving the worktree. Or discard the branch and brief the agent again. Both are one decision, made with the same information a human reviewer would have.
Review HubAn illustration of Review Hub for the branch fix/auth-retry: four changed files totalling 217 added and 39 removed lines, with merge readiness clean against main, four forge checks passed and one still running, three commits ahead, and nothing left uncommitted in the worktree.
One window for delegation, review, and merge
Every Agent, One Window
Any mix of agents, running in parallel, each in its own isolated workspace
Parallel Worktrees
Every task gets its own branch, no collisions
Unified Input
One set of @files, /commands, and shortcuts, working with every agent
Chat Portal
Ask any agent a question without switching context
GitHub Integration
Pick issues in bulk. Daintree spawns worktrees and agents for each, instantly.
Voice input, an MCP server, terminal recipes, plugins, themes and remote compute are all in here too. Explore feature guides
Blast radius and trust
Every boundary is explicit
Nobody wants to leave an agent running unattended with permission checks off, and not deciding is how you end up babysitting it instead of working. Here is exactly what each boundary holds, and exactly where it stops.
Changes
Every task gets its own branch and checkout
A worktree keeps agents working at the same time from overwriting each other, and keeps their work out of the tree you are in.
NotIt is Git isolation, not a sandbox. The agent process still runs as your user.
WorktreesProcess
Run the work somewhere that is not your laptop
A worktree can carry its own environment: a Docker stack, a cloud VM, a remote sandbox. You supply the provision, pause and teardown commands and Daintree runs them on the worktree lifecycle.
NotThis is where real process isolation comes from, and it is the level to reach for before turning permission checks off.
Remote computeDaintree itself
No relay, no control plane, no account
Your source, your diffs and your terminal output have no Daintree backend to pass through, because there is not one. Telemetry ships off. The MCP server ships off, and binds to localhost behind four authorization tiers when you turn it on.
NotWhat an agent CLI sends to its own provider is between you and that tool. Plugins execute code, and a manifest declares intent rather than confining it. The Daintree Assistant, still in testing, will be the one component of ours with a service in that path.
Trust and security
- Latest release
- See the changelog
- License
- Apache 2.0
- Built-in agents
- 17
- Contributors
- 5
- Source
- daintreehq/daintree
Give every agent its own worktree
Run the real CLI, see what needs you, and review the diff before you merge.
No account for the app · Apache 2.0 · View source