paperclipai/paperclip

What it is. Paperclip is a self-hostable control plane (a Node.js server with a React web UI) that coordinates multiple AI coding agents — Claude Code, Codex, OpenClaw, Cursor, Gemini CLI, and others — toward business goals. Instead of running 20 disjoint agent terminals, you define a goal (e.g., "build a notes app to $1M MRR"), "hire" agents into an org chart with roles and budgets, and monitor progress, costs, and approvals from one dashboard.

How it works. Agents plug in through adapters and receive "heartbeats" — wake-ups for assigned tasks, replies, or scheduled routines. Paperclip adds the plumbing that a lone agent CLI lacks:

• Task manager — ticket-style issues with threads, comments, files, atomic checkout (one agent owns a task at a time), reviews, and approval gates.
• Org chart — agents get titles, reporting lines, permissions, and scoped budgets; humans and agents share one hierarchy.
• Governance — approval workflows, pause/resume/terminate controls, config change history with rollback, and per-connection action rules (Allowed / Ask first / Off).
• Cost control — token/cost tracking by company, project, agent, and model, with warning thresholds and hard budget stops.
• Routines — cron/webhook/API-triggered recurring jobs that spawn tracked tasks.
• Skills & training — shared skills with version history, saved eval/test inputs, and quality metrics; agents receive skill context at runtime without retraining.
• Multi-organization — one deployment can isolate separate companies, agents, and activity histories.

Notable design choices. Tasks carry "goal ancestry" so agents see the business rationale behind work. Human access, agent eligibility, and gateway action permissions are separate controls, enabling accountable connections (e.g., GitHub operations attributed to the person directing the work). Adapters range from CLI agents like Cursor and Gemini to HTTP/webhook bots; the README's motto: "If it can receive a heartbeat, it's hired."

Tech shape. The core modules shown in the architecture diagram cover identity/access, work/tasks, heartbeat execution, org chart, workspaces/runtime, governance, budgets, routines, secrets/storage, activity/events, and company portability (export/import with secret scrubbing). A plugin system with out-of-process workers, capability-gated host services, and tool exposure extends the roster.