🪴 Brain

Paperclip

4 min read

Paperclip

paperclipai/paperclip — open-source orchestration for zero-human companies. If OpenClaw is an employee, Paperclip is the company. A Node.js server + React UI that orchestrates a team of AI agents: you define a goal, hire agents (any model, any runtime), approve the strategy, set budgets, and ship.

It looks like a task manager — under the hood it has org charts, budgets, governance, goal alignment, and agent coordination. “Manage business goals, not pull requests.”

🔗 Links

Description

Download or use

# Quickstart (trusted local loopback, embedded Postgres)
npx paperclipai onboard --yes
 
# Authenticated mode
npx paperclipai onboard --yes --bind lan
npx paperclipai onboard --yes --bind tailnet
 
# Manually
git clone https://github.com/paperclipai/paperclip.git
cd paperclip && pnpm install && pnpm dev
# API server: http://localhost:3100

Requirements: Node 20+, pnpm 9.15+. Tailscale as the solo-deploy mobile path; then production with your own Postgres + Vercel.

🗒️ Description

🧩 Three steps of use

#StepExample
01Define the goal”Build the #1 AI note-taking app to $1M MRR.”
02Hire the teamCEO, CTO, engineers, designers, marketers — any bot, any provider
03Approve and runReview strategy, set budgets, hit go, monitor from dashboard

Adapters (as of 2026-05): OpenClaw, Claude Code, Codex, Cursor, Bash, HTTP. “If it can receive a heartbeat, it’s hired.”

🧩 What Paperclip solves

  • 20 open Claude Code tabs → ticket-based tasks, threaded conversations, persistence across reboots
  • Manual context-gathering → context flows task → project → company goal automatically
  • Folders of disorganized agent configs → org chart + ticketing + delegation + governance out of the box
  • Runaway loops burning $$$ in tokens → cost tracking, throttling, budget hard-stops
  • Recurring jobs you have to remember → heartbeats + cron + management supervision

🧩 Hard orchestration details (things you can’t do well by hand)

  • Atomic execution — task checkout + budget enforcement are atomic (no double-work, no runaway spend)
  • Persistent agent state — agents resume task context between heartbeats, they don’t restart from scratch
  • Runtime skill injection — agents learn Paperclip workflows + project context at runtime
  • Governance with rollback — approval gates, revisioned config changes, safe rollback
  • Goal-aware execution — each task carries the full goal ancestry — the agent sees why, not just the title
  • Portable company templates — export/import orgs (with secret scrubbing + collision handling)
  • True multi-company isolation — one instance, many companies, separate data and audit trail

🧩 12 systems under the hood

Identity & Access · Org Chart & Agents · Work & Tasks · Heartbeat Execution · Workspaces & Runtime · Governance & Approvals · Budget & Cost Control · Routines & Schedules · Plugins · Secrets & Storage · Activity & Events · Company Portability.

🧩 What Paperclip is not

  • Not a chatbot — agents have jobs, not chat windows.
  • Not an agent framework — it doesn’t tell you how to build an agent, it tells you how to run a company of them.
  • Not a workflow builder — no drag-and-drop; it models companies (org charts, goals, budgets, governance).
  • Not a prompt manager — agents bring their own prompts/models/runtimes.
  • Not a single-agent tool — it’s for teams. 1 agent → unnecessary. 20 agents → critical.
  • Not code review — work orchestration, not PRs.

✍️ Reasoning for

The strongest things for me:

  1. Mobile-first management — Tailscale + phone = realistically controllable autonomous businesses. Fits my Principles philosophy (work that brings income while you sleep).
  2. Budget hard-stops — at PLSoft I’ve chased runaway loops a few times; here that wouldn’t happen.
  3. Multi-company isolation with one deployment — fits my portfolio: Qamera AI / PLSoft / Value Builders / Brain = 4 companies on one instance.
  4. Heartbeats vs continuous loop — cheaper, auditable, easy to govern.

Risks:

  • A fresh project with a small user base; production deploy = early-adopter pain.
  • Only makes sense at >5 agents. For 1-2 agents it’s overhead without benefit.
  • Bring-your-own-ticket-system is on the roadmap — today you have to live in Paperclip’s UI, not Linear/ClickUp.

Alternatives considered

  • Claude Code on its own — no governance, no budgets, context disappears on restart
  • n8n / Make / Zapier — workflow engines, they don’t model a company/goals/budgets
  • Archon — YAML workflows per-repo, not a cross-company control plane
  • Hermes Agent — single self-improving agent, Paperclip orchestrates many
  • Agentic Systems (my own architecture sketch) — Paperclip is a ready-made implementation of a similar idea

🔗 Resources

  • Roadmap: bring-your-own-ticket-system, Clipmart (one-click company templates)
  • Claude Code / Cursor / OpenClaw — supported adapters
  • Agentic Systems — my earlier thinking on multi-agent architectures
  • Agentic Coding — the paradigm Paperclip layers a “company” on top of
  • Autonomous Sales Agent Playbook — the economics Paperclip enables (governance + budgets instead of a SaaS stack)

Template: tool

Graph View

Backlinks