Local web UI to create and draft agent tasks without memorizing CLI commands.
tools\agent-gui\run-gui.bat ← double-click
pnpm agent:gui ← from repo root
Opens the browser at http://127.0.0.1:9477.
- Setup gate — blocks the UI until workspace, autostart, and agent CLI are ready (
ghis optional) - Sidebar (left) — task queue; status badge + linked PR when
ghis available - Center — task composer (title + Create), Program editor with Save / Verify / Run AI, provider & model footer
- Activity (right rail) — recent agent actions (tools, status, errors), scrollable
- More menu (···) — Complete with AI, Reload file, New task
Enterin title — create task (or start AI if a task is loaded)Ctrl/Cmd + Sin editor — save programCtrl + Enterin AI dialog — confirm requestEsc— close the more menu
The console follows the system theme automatically (light/dark) via prefers-color-scheme.
- Node.js 22+
- Valid Agent Loop workspace (
.agent-loop/,specs/agent-tasks/,scripts/agent/) - Autostart file (
.agent-loop/autostart) - Cursor CLI (
agent) or Claude Code CLI (claude) or OpenAI Codex CLI (codex), installed and authenticated - GitHub CLI (
gh, optional): PR badges in the task sidebar
On first open, Agent Console shows a setup gate until all required checks pass. Optional items (e.g. gh) do not block the UI.
With Cursor hooks installed (install.mjs --cursor), the ensure-gui hook on sessionStart checks whether Agent Console is up (/api/state on port 9477) and starts it if not (requires .agent-loop/autostart).
Manual ensure: pnpm agent:gui:ensure
| Variable | Default | Description |
|---|---|---|
AGENT_GUI_PORT |
9477 |
Local server port |
AGENT_GUI_NO_OPEN |
— | Set 1 to skip opening the browser |