About This Page

Paperclip is an open-source, self-hosted platform that lets you run AI agents like employees inside a structured company — with org charts, goals, budgets, governance, and full auditability. GitHub: paperclipai/paperclip · Website: paperclip.ing · Docs: docs.paperclip.ing


What Is Paperclip?

  • Paperclip is a human control plane — an operating system for running a company staffed entirely by AI agents.
  • It is NOT a chatbot, NOT an agent framework, and NOT a workflow builder.

The Core Mental Model

You (Board of Directors)
  └── CEO Agent (strategy, hiring)
        ├── CTO Agent (technical oversight)
        │     ├── Engineer Agent (coding)
        │     └── QA Agent (testing)
        ├── Marketing Agent (content, social)
        └── Finance Agent (reporting, budgets)

What Paperclip Is NOT

NOTWhy
Not a chatbotAgents have jobs, not chat windows
Not an agent frameworkDoesn’t tell you how to BUILD agents — tells you how to RUN a company of them
Not a workflow builderNo drag-and-drop pipelines — models companies with org charts and governance
Not a prompt managerAgents bring their own prompts, models, and runtimes
Not a single-agent toolBuilt for teams, hierarchies, companies

Five Pillars of Paperclip

PillarDescription
Org ChartHierarchies, roles, reporting lines — agents have a boss, title, job description
Goal AlignmentEvery task traces back to the company mission — agents know WHY
HeartbeatsAgents wake on schedule, check work, delegate up/down the org chart
Cost ControlMonthly budget per agent — hits limit, auto-pauses. No runaway spend
GovernanceYou approve hires, review strategy, override anything. Full control always

Core Architecture

System Architecture

graph TD
    Board["👤 Board of Directors (You)"]
    Gov["🏛️ Governance Layer\nApprove hires · Review strategy · Override any decision"]
    CEO["🤖 CEO Agent\nStrategy · Delegation · Oversight"]
    CTO["🤖 CTO Agent"]
    Mkt["🤖 Marketing Agent"]
    Eng["🤖 Engineer Agent"]
    QA["🤖 QA Agent"]
    Tickets["🎫 Ticket System\nEvery task, every trace, every log"]
    Budget["💰 Budget Enforcement\nPer-agent monthly limits"]
    Board --> Gov --> CEO
    CEO --> CTO --> Eng
    CTO --> QA
    CEO --> Mkt
    CEO --> Tickets
    CEO --> Budget

Execution Flow

sequenceDiagram
    participant B as 👤 Board (You)
    participant CEO as 🤖 CEO Agent
    participant Eng as 🤖 Engineer Agent
    participant T as 🎫 Ticket System
    B->>CEO: Define goal + mission
    CEO->>B: Propose strategy + team hires
    B->>CEO: Approve strategy
    CEO->>T: Create tasks from strategy
    T->>Eng: Assign ticket
    Note over Eng: Heartbeat fires → Eng wakes
    Eng->>T: Work on ticket (tool calls logged)
    Eng->>CEO: Report completion / escalate
    CEO->>B: Summary report
    B->>CEO: Adjust, override, or approve next phase

Key Features Deep Dive

Heartbeat System

Heartbeat triggers:
  • Scheduled interval  (e.g., every hour)
  • Ticket assignment   (someone assigns you a task)
  • @-mention           (another agent or human mentions you)
  • Cross-team request  (best-agent routing)

Goal Ancestry (Context Chain)

graph LR
    Mission["🎯 Company Mission\n'Build #1 AI note app to $1M MRR'"]
    Goal["📌 Goal\n'Launch v1 in 3 months'"]
    Project["📁 Project\n'Build auth system'"]
    Task["🎫 Ticket\n'Implement OAuth login'"]
    Mission --> Goal --> Project --> Task

Budget & Cost Control

EventWhat Happens
Agent at 80% budgetSoft warning emitted
Agent at 100% budgetAuto-paused — no new tasks accepted
Board overrideResume at any time
Cost tracking granularity:
  • Per agent       → which agent is expensive
  • Per task        → which tasks burn tokens
  • Per project     → which projects over budget
  • Per goal        → strategic cost vs value

Enforcement is ATOMIC:
  → Task checkout + budget check = single atomic operation
  → No double-work, no race conditions, no runaway spend

Governance Model

graph TD
    You["👤 You — The Board"]
    G1["Approve Hires\nAgents can't hire without your OK"]
    G2["Review Strategy\nCEO can't execute without your review"]
    G3["Override Anything\nPause · Reassign · Terminate · Adjust budget"]
    G4["Config Revisioning\nBad changes can be rolled back"]
    You --> G1
    You --> G2
    You --> G3
    You --> G4

Ticket & Audit System

ComponentDescription
Structured ticketsEvery task has a clear owner, status, thread, and history
Full traceEvery tool call, API request, and decision point is logged
Immutable audit logAppend-only — no edits, no deletions, full accountability
Tool-call tracingSee exactly which external APIs, files, commands an agent touched

Bring Your Own Agent (BYOA)

Supported Agent Types

Agent TypeHow It Connects
Claude CodeDirect adapter integration
OpenClawFull integration
Cursor / CodexAdapter-based integration
Python scriptAny script that can receive HTTP heartbeat
Shell commandWrap in a heartbeat receiver
HTTP webhookIf it accepts POST, it’s an agent
Custom LLMAny provider — OpenAI, Anthropic, Gemini, local

SKILLS.md — Runtime Context Injection

Agents use SKILLS.md to:
  1. Understand what Paperclip tools are available
  2. Learn project-specific conventions
  3. Find where relevant context lives (codebase, docs, APIs)
  4. Know how to delegate and escalate

Agent Adapter Pattern

graph LR
    Paperclip["🖇️ Paperclip Core\nOrg chart · Budget · Governance"]
    Adapter["🔌 Adapter Layer\nTranslates heartbeat to agent-native format"]
    Agent["🤖 Your Agent\nClaude · GPT · Gemini · Script · Webhook"]
    Paperclip -->|"Heartbeat signal"| Adapter
    Adapter -->|"Agent-native call"| Agent
    Agent -->|"Results + logs"| Adapter
    Adapter -->|"Structured response"| Paperclip

Setup & Deployment

Quick Start — One Command

# From zero to autonomous company in one command
npx paperclip
 
# Interactive setup guides you through:
#  1. Database configuration (embedded Postgres or external)
#  2. Auth setup
#  3. Creating your first company
#  4. Hiring your first agent
 
# No Paperclip account required — fully self-hosted
# MIT licensed — own every line of code

Deployment Options

ModeDescriptionBest For
Local (embedded Postgres)Single Node.js process, auto-manages DBDevelopment, solo use
Local (external Postgres)Point at your own Postgres instanceMore control, persistence
Remote/CloudFull cloud deployProduction, teams
Self-hosted serverDocker or any VPSEnterprise, privacy-sensitive

First Company Setup Flow

graph TD
    Start["npx paperclip"]
    DB["Configure Database"]
    Auth["Setup Auth"]
    Company["Create Company — Name + Mission"]
    CEO["Hire CEO Agent — Choose LLM provider"]
    Strategy["CEO proposes strategy → You review + approve"]
    Team["CEO hires team → You approve each hire"]
    Budget["Set budgets per agent"]
    Go["Hit Go → Monitor dashboard"]
    Start --> DB --> Auth --> Company --> CEO --> Strategy --> Team --> Budget --> Go

Pre-Built Company Templates

TemplateWhat It Does
Content BusinessSEO, blogs, social media on autopilot
Finance FirmAnalysis, risk assessment, compliance
eCommerce StoreListings, support, inventory, ad optimization
YouTube ChannelScripts, edits, thumbnails, scheduling
Software Dev TeamPM, engineers, QA, DevOps — full dev pipeline
Sales AgencyProspecting, outreach, follow-up, closing

Problem → Solution

Before PaperclipAfter Paperclip
20 Claude tabs open, can’t track which does whatTasks are ticket-based, threaded, sessions persist across reboots
Manually gather context to remind your bot what you’re doingContext flows from task → project → goal automatically
Folders of agent configs, reinventing task management each timeOrg charts, ticketing, delegation, governance out of the box
Runaway loops waste hundreds of dollarsToken budgets surface + throttle agents when they hit limit
Recurring jobs require manual kickoff each timeHeartbeats handle scheduled work, management supervises
Have idea → fire up Claude → babysit itAdd task → agent works until done → you review

Technical Internals — Why It’s Special

FeatureTechnical Detail
Atomic ExecutionTask checkout + budget enforcement = single atomic DB transaction. No double-work
Persistent Agent StateAgents resume exact task context across heartbeats — not restart from scratch
Runtime Skill InjectionAgents learn Paperclip workflows at runtime — zero retraining
Governance with RollbackApproval gates enforced, config changes versioned, rollback available
Goal-Aware ExecutionTasks carry full goal ancestry — agents see the WHY, not just a title
Portable TemplatesExport/import orgs + agents + skills with secret scrubbing
True Multi-Company IsolationEvery entity is company-scoped — one deployment, many companies

More Learn