Dimension
Paperclip
agent-swarm.dev
Core bet
PaperclipOrchestration and governance are the durable layer.
agent-swarm.devAccumulation is the moat — the learning loop on top of the agents.
What persists
PaperclipPersistent task state, runtime-injected skills, portable company templates and an append-only audit log.
agent-swarm.devCross-session memory, journals and agent identities that compound into institutional context.
How it gets better
PaperclipSwap in stronger agents or shared templates; Paperclip coordinates and governs them.
agent-swarm.devPrivate evals on your workflows — roadmap to private RL — on your definition of done.
Human role
PaperclipYou act as the board: set goals and approve structural changes; agents run autonomously between gates.
agent-swarm.devYou direct the Lead; workers execute, reviewers challenge — a human stays in command of the work.
Mental model
PaperclipA company of AI employees — org chart, reporting lines, budgets, a CEO agent (Zeus).
agent-swarm.devA Lead that decomposes and fans work across many workers and reviewers.
Governance
PaperclipFirst-class — budget caps, atomic task checkout, approval gates, full append-only audit trail.
agent-swarm.devHuman-in-command review and challenge; traces and journals stay inside your perimeter.
Agent / model choice
PaperclipAgent-agnostic — Claude, Codex, Gemini, Cursor, Pi, OpenCode, scripts and webhooks under one org chart.
agent-swarm.devModel- and harness-agnostic — Claude, Codex, opencode, pi; the model is replaceable compute.
Deployment
PaperclipOpen-source (MIT), self-hosted — Node.js + PostgreSQL + React; managed cloud on the roadmap.
agent-swarm.devSelf-hosted (Docker Compose) — or our cloud.