Agents
An M1 agent is proposed as a saved configuration around the AI development tool you already use—not a new model and not an autonomous employee.
01
Set stable operating instructions, reusable skills, permitted repositories, plain and secret environment inputs, runtime arguments, timeout, concurrency, and working-directory mode.
02
For larger work, a lead agent can interpret the request and involve specialized agent roles. People remain participants and reviewers; each sub-result returns to one parent task.
03
Teams should be able to inspect what was requested, which configuration ran, where it ran, what changed, what checks passed, and what remains uncertain before accepting the result.
Pre-release · No instant signup or payment