Technology RadarTechnology Radar

Paperclip

agentopen-sourcemulti-agentframework
Trial

Paperclip is an open-source Node.js + React platform that wraps a team of AI agents in a full company structure — org charts, job titles, goal alignment, token budgets, ticketing, audit logs, and approval workflows. It orchestrates agents (Claude Code, Codex, Cursor, bash scripts, HTTP webhooks) as if they were employees inside an organization rather than individual tools you prompt ad hoc.

The Core Idea

The mental model is deliberately organizational: you hire agents into roles with reporting lines, define the company's mission, and let Paperclip manage scheduling, budget enforcement, and task handoffs. Individual agent capability is taken for granted; the platform solves the coordination layer that's missing when you try to run more than one agent at a time.

This differs from workflow engines like Archon (which encode process as DAGs) and from frameworks like CrewAI or LangGraph (which define agent behavior in code). Paperclip's abstraction is the company — roles, approvals, audit history — not the workflow or the graph.

Architecture

React Dashboard (org chart · ticket board · budget monitor)
         │
    Node.js Server (REST API, scheduling, event bus)
         │
    ┌────┴──────────────────────────────────────────┐
    │                    │                          │
Heartbeat Engine   Governance Layer          Plugin System
(scheduled wakeups,  (approval workflows,    (out-of-process workers,
 budget check,        role-based permissions, tool exposure,
 workspace mgmt)      immutable audit log)    UI contributions)
         │
Agent Adapters (Claude Code · Codex · OpenClaw · Cursor · bash · HTTP)
         │
Encrypted Secrets + Object Storage (scoped per company)

Each agent receives a heartbeat on a configured schedule. The heartbeat delivers context (goals, active tickets, budget remaining) and collects work products. Atomic task checkout prevents two agents from pulling the same ticket; hard budget stops prevent runaway API spend — the cost-control problem Paperclip explicitly targets.

Key Features

  • Org chart & governance: Hierarchical agent roles, reporting lines, and approval gates before high-stakes actions
  • Goal alignment: Tasks trace back to the company's stated mission; agents know both what and why
  • Heartbeat execution: Cron-style agent wake-ups with budget checks and workspace handoff
  • Cost control: Per-agent token budgets with hard stops and cost event logging
  • Multi-company isolation: Complete data separation for multiple organizations in a single deployment
  • Portable templates: Export/import entire company structures (agents, goals, configs) as JSON
  • Plugin system: Extend with custom tool exposure and background workers

Why Trial

  • 59k+ GitHub stars in under eight weeks (launched March 4, 2026) — among the fastest-growing open-source AI repositories of 2026
  • 10k+ forks signals genuine experimentation, not just bookmarking
  • MIT licensed, TypeScript (97.7% of codebase) — auditable and low lock-in risk
  • Fills a genuine gap: the coordination layer between agents is consistently cited as the unsolved problem in multi-agent systems
  • Budget enforcement and audit logs address the most common production concern with autonomous agents (runaway costs and no accountability)

Caveats: no managed cloud offering exists as of April 2026 — self-hosted only, requiring Node.js 20+, pnpm 9.15+, and comfort with CLI tooling. Enterprise auth (SSO, RBAC beyond built-in roles) is not yet documented. The project is only weeks old; operational best practices are still forming in the community.

When Paperclip Makes Sense

  • You want multiple agents (coding, writing, testing, operations) working toward shared goals without constant human intervention
  • Cost control and audit trails are non-negotiable — Paperclip's budget enforcement is first-class, not an afterthought
  • You're building or operating what might be called an "AI-native business unit" where agent output is the primary work product
  • You want organizational primitives (roles, approvals, ticketing) rather than workflow primitives (DAGs, chains)

Not Right For

  • Single-agent workflows — the overhead of company structure isn't justified
  • Teams that need a managed SaaS offering (not available yet)
  • Encoding complex conditional logic across agents (use LangGraph, CrewAI, or Archon's DAG model instead)
  • Organizations that need enterprise SSO or compliance-grade audit exports today

Key Characteristics

Property Value
Language TypeScript (97.7%)
License MIT
GitHub paperclipai/paperclip
Stars ~59k (April 2026)
Launched March 4, 2026
Runtime Node.js 20+, pnpm 9.15+
Supported agents Claude Code, Codex, OpenClaw, Cursor, bash, HTTP webhooks
Hosted option Self-hosted only

See Also

  • Archon — DAG-based workflow engine for AI coding; complementary rather than competing
  • Background Coding Agents — the broader practice of asynchronous AI coding that Paperclip operationalizes at scale
  • AgentOps — observability layer that pairs well with Paperclip's audit log

Sources