Not just a chat window — an Agent console you can see and control.
01 A graphical Agent workbench — windows, tasks, steps, permissions, and processes on one
screen. Not a shell around a command line, but a console designed for long-term use.
02 It writes, edits, and verifies — every structured edit is bound to freshly read file state.
Before commit it checks again; if the file changed, the edit stops instead of overwriting it.
03 Permissions are not decoration — R0–R4 risk tiers, a kernel-level Hardline, three approval
modes. Pro mode means fewer prompts, not fewer boundaries.
04 Every step has a fact behind it — the Event Log is the single source of truth. A refresh
restores from facts, not from the wording of an answer.
06 Architected for long tasks — Task Lease, Checkpoints, Runtime Session rebuild, and Context
Compaction. Recovery reconciles current state before continuing on a fresh session and lease.
Distributed as source on GitHub. No invitation, no application, no allowlist — if you found it and want to take part, you can.
It is not a Beta. This round is source-only and limited to Apple Silicon Macs. Broad model compatibility is not certified, the full long-task public test is still gated, and long-term memory, an in-house Loop, and the plugin ecosystem remain later work.
The most useful thing you can send me right now is a bug report from a real long task: the Window ID, what you expected, and what happened.