Argus
A self-evolving multi-agent system for autonomous research and engineering, splitting planning, execution and review across separate roles so long-running work continues, pauses and resumes without a human driving every step.
Argus is a self-evolving multi-agent system for persistent, reviewed autonomous research and engineering, able to plan, execute, verify, pause and continue work beyond a single model turn.
It splits the work of directing an agent across four roles that cannot do each other's jobs: a Manager that owns stage transitions and where a lesson is kept, a Planner that owns the next task and the evidence it must produce, an Engineer that owns implementation and experiments but cannot declare its own work complete, and a read-only Reviewer that owns the verdict and may return the work as blocked. Credentials, payments, irreversible actions and publication always stop for a human.
It improves without retraining by admitting new skills and source-linked findings scoped from a single project up to a whole vertical or globally, with seven built-in verticals and seventeen more available in a community package. This repository is a development preview; the official source repository is maintained separately and receives changes through synchronization.