Orchestrate AI Agents into Org Charts with Paperclip
Use Paperclip's open-source orchestrator to build AI org charts where a CEO agent delegates tasks to specialized employees (coders, marketers) for reliable business automation, starting with 'npx paperclip-ai onboard'.
Build AI Organizations for Real Business Output
Paperclip lets you create an org chart of AI agents as employees (CEO, CTO, coders, marketers) that handle accountable work like coding, video production, and reports. Install via npx paperclip-ai onboard to scaffold a company; bring your own models like Claude, Codex, Gemini, or cheaper ones via OpenRouter (e.g., free Qwen 2.5 Coder). Agents share memory, brand guides, and context automatically, so a video writer can access stats dashboards and branding to produce on-brand Remotion animations in 5 minutes—tasks that would take a human a week.
You act as the human CEO: assign issues to your AI CEO, who breaks them down, hires agents dynamically (e.g., video writer with Remotion skill), creates plans, and iterates based on your feedback. For a 40k GitHub stars celebration (hit 50k during demo), the CEO hired a video writer, installed Remotion best practices skill, planned a stats-animated video, and refined cuts to 2 seconds after feedback. This preserves your taste without manual context pasting across tools.
Track monthly spend per agent/project (uses API subscriptions initially), set budgets, and run one task at a time by default (configurable concurrency). Released March 4; by April 8 (34 days), gained 40k+ stars via community PRs improving reliability.
Enforce Reliability with Workflows and QA
Agents drift without structure—Paperclip adds vendor-neutral workflows: require QA reviewer (with agent-browser skill for browser testing: open sites, fill forms, click buttons) before completion, then manager approval. This iterates coder-QA loops reliably across models, unlike one-way hooks in Claude/Cursor.
Routines automate repetitive tasks: schedule or run manually with templates (e.g., "Create Discord message for merged PRs, write changelog, use Greptile for first-pass code review"). Group by project/agent; embed skills. Non-coders use for marketing/sales/finance—e.g., process Twitter bookmarks into reports on execution adapters or memory strategies.
Future: Action buttons on reports to spawn issues/plans; organizational learning from feedback (e.g., coders learn 2-second cuts, skill consultant diagnoses skill underuse).
Customize Agents Iteratively for High Quality
Start small: Hire CEO first (Claude/Codex recommended), approve hires/plans agent-by-agent to ensure quality before scaling. Edit instructions constantly—e.g., Codex coder: "On blockers, suggest fixes with tutorials; write partial tests only." Use meta-agents like skill consultants for diagnosis.
Avoid huge templates (130+ agents); craft preferences manually for results. Cheaper models for routine tasks, frontier for high-intelligence. Agents negotiate/communicate with shared memory.
Roadmap (30 days): CEO chat, maximizer mode (burn tokens relentlessly), multi-human users/cloud deploy, workspaces for PRs, sandboxing (E2B/devbox), desktop app, artifacts/deployments, better memory/knowledge base. Deploy to cloud for teams; empowers non-technical users to manage AI labor without coding tools like Cursor/GitHub.