Loom / August 1, 2026

Loom public preview

Loom is now available as an open-source starting point for teams experimenting with persistent, reviewable agent work.

This first public preview opens Loom as a practical reference for work that outlives a single prompt. It is designed around a simple premise: an agent task should keep its objective, active context, and meaningful execution record together.

The preview focuses on the task surface rather than a new model interface. It is intended to work alongside the coding agents and developer tools a team already uses, while making the surrounding state easier to inspect and resume.

Future updates will describe concrete changes to Loom's runtime, recovery, review, and collaboration surfaces.

Open the Loom page

Recent updates

July 3, 2026

Loom

Run records and review checkpoints

A clearer model for connecting an agent's claim of progress with the checks and artifacts behind it.

June 5, 2026

Loom

Workspace recovery and handoff

A task should be inexpensive to resume after an interruption, whether the next operator is a person or another agent.