OpenClaw

What it does

OpenClaw is a fleet of named, persistent AI agents ("claws") that each own a slice of someone's professional or personal workflow. A single agent might triage email, capture transcripts, manage a project board, or run a recurring report — and pass results to its sibling claws when needed.

Agents are scheduled, monitored, and audited centrally. Every action is logged, replayable, and traceable to the human approver who authorised it.

Architecture

• Per-claw repo + secrets: each claw has its own folder, encrypted secrets via age + sops, and a Telegram identity.
• Orchestrator: a separate openclaw_orchestrator project handles deployment + audit plus loop scheduling across the fleet.
• Long-lived loops: the always-on loop scales agent count to fit Anthropic API budgets and pacing.

Status

Telegram is live; WhatsApp interface and shared vector memory are pending. The fleet runs on a Hetzner workhorse VPS plus Vercel for any web surface.

Why it matters as a use case

Most "agent" demos are single-shot. OpenClaw is a working illustration of agents that survive across days, hand work to each other, and stay safe through explicit human approval gates.