The agent that does what you
want, how you want.
Hand Argus an objective. It spins up a fleet of child agents, steers them while they work, approves the routine steps, and escalates the decisions that are actually yours to make. You set the intent; it runs the orchestra.
One intent.
A fleet that carries it out.
Argus holds the goal and the guardrails. Each child agent inherits exactly the capabilities it needs — no more — and reports back as it works. When a session stalls or a step needs a human, it surfaces; everything else just keeps moving. You stay in the loop without being the bottleneck.
Orchestrates
A fleet, not a single thread.
Argus spins up child agents, hands each a slice of the work, and keeps them moving in parallel — the same parallelism you'd run by hand, now driven by one objective.
Approvals with lineage
Safe steps clear themselves.
Routine, in-policy actions get auto-approved against the capabilities you granted. Anything outside that lineage stops and waits for you — so nothing dangerous slips through, and nothing trivial slows down.
Always watching
A pulse on every session.
A live pulse tracks every child agent, logs what they do, and reaps the ones that finish. When something needs you, it escalates — the name Argus, the hundred-eyed watcher, is the whole idea.
Further reading
Automating adversarial reviews for Claude Code and Codex
Put an Argus in charge of a group of agents and tell it to run an adversarial review on their work before it reaches you. Here's what that review is, what it catches, and how to run it — in Scape or anywhere else.
Read it →Point Argus at the work.
Set the objective and the guardrails. Argus runs the fleet — you stay in the loop. Available now for macOS.
