AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
A desktop app that starts coding agents from your tickets. santree writes each agent's prompt from the ticket, runs it in a worktree of its own, and turns review comments into the next prompt.
Developer toolsTypeScript · RustGitHub @santree-aiOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
Your backlog, shipped in parallel — a desktop app that runs Claude agents across your repo's tickets, each in its own git worktree. Triage in, PRs out.
Taken from the repository README
Loading the full README. The saved excerpt is shown below.
Your backlog, shipped in parallel.
santree is a desktop app that runs Codex and Claude Code across your repo's tickets.
Each agent gets an isolated git worktree you can watch, steer, and merge.
Triage in, PRs out.
Running one coding agent is easy. Running five is a workflow problem: they step
on each other's diffs, you lose track of who's blocked on you, and the actual
job (tickets in, reviewed PRs out) lives scattered across a tracker, a
terminal multiplexer, and a browser.Read the full README ↗Your backlog, shipped in parallel.
santree is a desktop app that runs Codex and Claude Code across your repo's tickets. Each agent gets an isolated git worktree you can watch, steer, and merge. Triage in, PRs out.
Website · Docs · Download for macOS · Changelog
Running one coding agent is easy. Running five is a workflow problem: they step on each other's diffs, you lose track of who's blocked on you, and the actual job (tickets in, reviewed PRs out) lives scattered across a tracker, a terminal multiplexer, and a browser.
santree puts the whole loop in one window. Every task gets an isolated git worktree with a real terminal running the real Codex or Claude Code CLI: your login, your subscription, no credentials handled by santree. Around the terminals it builds the workflow: a permanent sidebar that shows every agent's state across every project, your team's triage queue with its SLA clock, your tickets as a list or a dependency graph, a review inbox with an AI co-reviewer, and a work queue that turns failing checks and review comments into an agent's next task. Nothing is mocked: every view is backed by live data, and an unconnected view shows its honest empty state.
The sidebar is the app's index. It never leaves: search, your tickets, the triage queue, and every project with its worktrees and the agents running on them, each with a dot that says working, just finished, or needs you. The area beside it is whatever you picked.
santree runs on macOS: download the DMG. It's signed and notarized, and keeps itself up to date (pick the stable or beta channel in Settings → General).
You'll also want:
gh), signed in. Optional; it
powers Reviews, the PR panes and the checks.You need Rust (the pinned toolchain installs
automatically), Node 22+, pnpm, and the Xcode Command Line Tools
(xcode-select --install).
pnpm install
pnpm dev # first run compiles the Rust side, a few minutes, then fast
Architecture, data flow, testing, the release pipeline, and the terminal internals live in DEVELOPMENT.md. The ground rules for how santree is allowed to interact with agent CLIs are in COMPLIANCE.md; they're load-bearing.
The screenshots above come from the app's screenshot fixture mode, a fake
company (Mallard Labs, makers of QuackStack) served to the real views; see
src/dev/fixtures/README.md to retake them.
santree is pre-release and moving fast. Bug reports and small PRs are very welcome; for anything bigger, open an issue first so we can talk it through.
MIT © Santiago Toscanini