Paperclip (paperclip.ing, GitHub) is the open-source app for managing AI agents at work - an MIT-licensed Node.js server and React dashboard that organizes agents you bring (Claude Code, Codex, Gemini CLI, Cursor, OpenCode, OpenClaw) into an org chart with budgets, tickets, and heartbeat scheduling. It launched in March 2026 and passed 50,000 GitHub stars within weeks. This page compares it honestly with cto AI Business - the managed multi-agent platform where the team, the models, and the infrastructure come with the product.
TL;DR. Paperclip is a manager for agents you already run: self-hosted, bring-your-own agents and API keys, total control and auditability, costs whatever your infra + LLM subscriptions cost. cto AI Business is the managed version of the same idea: Team Lead + Team Members out of the box, 10+ frontier models with no API keys, MCP integrations pre-configured, free forever to pilot. Engineers who want to self-host pick Paperclip; everyone who wants the team running today picks the managed platform.
What the two products actually are
Paperclip is infrastructure you deploy. npx paperclipai onboard --yes on Node.js 20+, PostgreSQL for production. You define an org chart - hierarchies, roles, reporting lines - and connect agents via adapters. Every task is a ticket with an owner, status, and append-only audit trail of every tool call and decision. Per-agent monthly spending caps (with an 80% warning) keep costs hard-bounded. A heartbeat loop wakes agents on schedules. It does not include agents - you bring Claude Code, Codex, Gemini CLI, or another supported adapter, each with its own subscription or API key.
cto AI Business is the running team. Concrete concepts:
- Headquarters - command center. Team overview, approvals, billing, pause/resume token usage.
- Plan - your business objective in 1–3 sentences. Drives Team Lead decisions.
- Team Lead - primary chat interface. Hires/fires Team Members, assigns + evaluates work, interacts with MCPs.
- Team Members - specialist agents you hire via "Hire employee." Each has a role, an assigned model, an activity log.
- Tasks - kanban view of work in progress, with outputs attached per task.
- Approvals - actions you want to gate behind human sign-off, surfaced in chat + page header.
You don't write code, manage API keys, or run a server. You configure via Headquarters and chat with the Team Lead.
The comparison, honestly
| cto AI Business | Paperclip | |
|---|---|---|
| Hosting | Managed - nothing to deploy | Self-hosted (Node.js 20+, PostgreSQL) |
| Agents | Included - hire Team Members in the UI | Bring your own (Claude Code, Codex, Gemini CLI, Cursor, OpenCode, OpenClaw…) |
| Models | Auto model + 10+ frontier LLMs, no API keys | Whatever your agents' keys/subscriptions cover |
| Price | Free forever pilot (ad-supported, rolling 24h+7d limits); Premium for raised limits | Free (MIT) + your infra (~$10–20/mo VPS) + LLM costs per agent |
| Source | Proprietary platform | Open source, fork/audit/contribute |
| Cost control | Pause/resume token usage from Headquarters | Hard per-agent monthly caps, 80% warnings |
| Audit trail | Per-member activity logs, task outputs | Append-only ticket log of every tool call |
| Integrations | 10 pre-configured MCPs + any custom MCP | MCP via @paperclipai/mcp-server, npm adapters |
| Setup time | Minutes (sign in, write a Plan) | Hours (deploy, configure DB, wire agents + keys) |
Where Paperclip genuinely wins: control. It's MIT-licensed, runs on your hardware, and the audit trail is exhaustive. If your data can't leave your network, or you want to read the source of the thing managing your agents, that's decisive.
Where cto AI Business wins: time-to-running-team and total cost honesty. Paperclip is free software, but the agents aren't free - a serious multi-agent setup means multiple LLM subscriptions or API bills (heavy Claude Code usage alone is a $200/mo Max plan), plus a server, plus you as the operator. cto's free tier includes the models.
The real cost of "free and open source"
Stack up a three-agent Paperclip team: a VPS ($10–20/mo), an LLM subscription or API budget per agent, and your time deploying, upgrading, and debugging adapters. The software costs nothing; the system doesn't.
cto AI Business inverts that: the platform bundles the models (Auto picks per task across 10+ frontier LLMs), so the free tier is genuinely free - ad-supported with rolling 24h+7d limits, enough to validate a workflow end-to-end. Premium raises the limits when you go to production. No keys to provision, no server to patch.
How a cto AI Business team works
Set up:
- Create the business. Sign in at cto.new, write a Plan.
- Talk to the Team Lead. Describe what you want the team to handle.
- Hire Team Members. Click "Hire employee" → name, role, model. Or let the Lead hire within your approval rules.
- Wire MCP integrations. Sentry, Vercel, Supabase, Cloudflare Observability, Notion, Neon, Linear, Prisma, Render, Webflow are pre-configured; custom MCP via the same UI.
- Set approvals. Decide which actions need human sign-off.
- Run it. Tasks flow into kanban; intervene from Headquarters.
Both products share a conviction: agents should be managed like a team - roles, oversight, budgets - not fired off as one-shot prompts. They disagree on who runs the machinery.
Models - Auto + frontier menu
cto's gateway covers: Claude Opus 4.8, Sonnet 4.6, GPT-5.5, GPT-5.3 Codex, Gemini 3.5 Flash, Gemini 3.1 Pro, GLM 5.1, Grok 4.20, Kimi K2.6, MiniMax M2.7.
Each Team Member gets an assigned model - Auto by default, or pin a specific one. In Paperclip, model access is per-agent: whatever each connected agent's own subscription or key provides.
Other AI agent platforms in the space
| Platform | Best for |
|---|---|
| cto AI Business | Free-forever pilot, Team Lead + multi-agent, MCP-native, managed |
| Paperclip | Open-source, self-hosted agent management, exhaustive audit trails |
| Polsia | Hands-off autonomous execution (compared here) |
| Lindy | Largest native-integration catalog (3,000+), polished SMB UX |
| CrewAI | Code-first multi-agent framework |
| AutoGen | Python multi-agent prototyping (MIT) |
When Paperclip is still right
Honest cases:
- Data can't leave your network. Self-hosted is the requirement, not a preference.
- You already run the agents. If your team lives in Claude Code or Codex with existing subscriptions, Paperclip adds management without new model costs.
- You want to audit or extend the source. MIT license, npm adapter ecosystem, active repo.
- Compliance needs the append-only trail of every tool call and decision.
If none of those apply, the managed platform gets you a working team this afternoon.
FAQ
What is Paperclip AI?
An open-source (MIT) application for managing teams of AI agents at work - org chart, goals, budgets, tickets, and scheduling for agents you connect. It launched in March 2026, built by the pseudonymous developer @dotta, and passed 50,000 GitHub stars within a month. (Not the "paperclip maximizer" thought experiment - that's a 2003 AI-alignment hypothetical by Nick Bostrom.)
Is Paperclip AI free?
The software is free and MIT-licensed. Running it isn't: budget a VPS (~$10–20/mo) plus each connected agent's LLM subscription or API costs. A hosted plan is on the roadmap but hasn't launched.
What agents does Paperclip support?
Claude Code, Codex, Gemini CLI, Cursor, Pi, OpenCode, OpenClaw, and Hermes via official adapters, plus third-party adapters installable as npm packages and MCP support through @paperclipai/mcp-server.
Do I need my own API keys with cto AI Business?
No. The cto.new gateway provides 10+ frontier models; each Team Member is assigned a model (Auto by default). No keys, no per-provider billing.
Is cto AI Business open source?
No - it's a managed platform. If self-hosting is a hard requirement, Paperclip is the better fit; the free cto tier is the faster way to validate whether a multi-agent team works for your business at all.
Can I use both?
Reasonably, yes. Some teams pilot the workflow on cto AI Business (free, no setup), then self-host Paperclip once they know exactly which agents and roles they need - or keep regulated workloads on Paperclip and everything else managed.