Paperclip: Open Source Human Control Plane for AI Labor — Dotta Bippa

Key Concepts

• Paperclip: An open-source agent orchestrator designed as a "human control plane" for AI labor.
• Agentic Surface: The capability of an agent to perform tasks, hire other agents, install skills, and manage workflows.
• Skills Manager: A system for installing and managing specific capabilities (e.g., Remotion for video, Agent Browser for web interaction).
• Org Chart Architecture: A hierarchical structure where a CEO agent delegates tasks to executive and individual contributor agents.
• Vendor-Neutral Harness: The ability to integrate models from various providers (Claude, Codex, Gemini, etc.) into a single workflow.
• Routines: Reusable, scheduled, or manual task sequences that automate repetitive business processes.

1. Introduction to Paperclip

Paperclip is positioned as an open-source orchestration tool for "zero-human companies." Rather than fully autonomous systems that operate in a black box, Paperclip acts as a human control plane, allowing users to maintain oversight, set goals, and inject personal taste into AI-driven workflows. The project has seen rapid growth, recently surpassing 50,000 GitHub stars.

2. Operational Framework: The Org Chart

The core of Paperclip is the creation of an organizational structure:

• CEO Agent: The primary interface for the human user. It breaks down high-level goals into actionable tasks for subordinates.
• Executive/Individual Contributors: Specialized agents (e.g., coders, video writers, marketing strategists) hired by the CEO to execute specific functions.
• Agentic Hiring: The CEO agent has the authority to hire new agents and install necessary "skills" to complete assigned projects.
• Context Management: Unlike standalone agents, Paperclip agents share a unified memory and context (e.g., brand guides, project dashboards), ensuring consistency across tasks.

3. Workflow and Methodology

• Task Execution: Users define a project, and the CEO agent generates a plan. The human user reviews and provides feedback on this plan before execution begins.
• Quality Control (QA & Approval): Paperclip implements a "human-in-the-loop" or "agent-in-the-loop" verification process.
  • QA Agent: Uses the "Agent Browser" skill to verify work (e.g., testing code in a browser).
  • Review/Approval: Workflows can be configured to require a reviewer (QA) and an approver (Manager) before a task is considered complete.
• Routines: These are automated workflows that handle recurring tasks, such as generating release logs, merging PRs, or posting to Discord. They support template variables for dynamic execution.

4. Technical Implementation & Advanced Tips

• Installation: Users can initialize the system via terminal using npx paperclip-ai onboard.
• Model Flexibility: Paperclip is vendor-neutral. Users can mix and match models based on cost and intelligence requirements. For example, using high-intelligence models (Claude/Codex) for the CEO and cheaper models (Qwen 3.6+) for routine tasks via OpenRouter.
• Instruction Tuning: Users are encouraged to provide ongoing feedback to agents. By refining instructions (e.g., "If blocked, provide a tutorial on how to fix it"), users create "meta-agents" that improve over time.
• Skill Consultants: A specialized agent role designed to diagnose and optimize how other agents utilize their installed skills.

5. Real-World Application: The 40k Stars Celebration

The creator demonstrated the platform's utility by automating a video production task:

1. Goal: Create a Remotion-based video celebrating 40,000 GitHub stars.
2. Process: The CEO agent was instructed to hire a video writer and install the "Remotion best practices" skill.
3. Execution: The agent generated a plan, which the human refined (e.g., adjusting animation pacing).
4. Result: A professional-grade video was produced in approximately 5 minutes, leveraging existing brand guides and stats dashboards already stored within the Paperclip environment.

6. Roadmap and Future Development

The project is in its early stages (launched March 2024) with active development on:

• Multi-user support: Enabling team-based collaboration.
• Cloud/Sandboxing: Deploying agents in isolated environments (E2B, dev.exe).
• Maximizer Mode: A high-intensity mode for agents to work continuously toward a goal.
• Desktop App: A native interface for managing the orchestrator.

Synthesis

Paperclip shifts the paradigm of AI usage from "prompting a chatbot" to "managing an organization." By providing a structured, hierarchical, and skill-based environment, it allows non-technical users to scale their output by delegating work to a fleet of specialized agents. The platform’s strength lies in its ability to maintain brand consistency and operational reliability through human-defined workflows and rigorous QA processes.