Running multiple AI agents on the same project easily blows your API budget and creates chaos: agents duplicate work, lose context, and erase all progress if you restart them. Paperclip, an open-source Node.js server with a React dashboard, tackles this by wrapping agents in a company structure complete with org charts, budgets, goals and audit logs. Since its March 2026 launch by a pseudonymous developer (@dotta), it’s racked up 43,000 GitHub stars—proof that the coordination gap matters as much as raw agent smarts. After a one-line install, you name a company, write a mission statement and spin up agents with roles—CEO, engineers, marketers. Each “heartbeat” interval, agents check their task queues, see the full context chain from mission down to specific tickets, and act within assigned budgets and approval gates. All decisions, tool calls and conversations are logged. Agents can’t hire or spend without human sign-off, and they stop when their budget cap hits. The platform works with any HTTP-friendly model—Claude Code, Codex, OpenClaw, Gemini CLI—and stays fully self-hosted under an MIT license, so you pay only your LLM provider, not Paperclip. Paperclip’s heartbeat system confronts what @dotta calls the “Memento Man” problem: agents start each session with zero memory. Instead of retraining models, you tweak persona prompts and add rules to guide future sessions. Over time, each agent picks up constraints and preferences like a new hire taking written feedback. Compared with tools like LangChain or CrewAI, which focus on building individual agents or pipelines, Paperclip sits above them—managing agents as an organization rather than supplying chat windows or drag-and-drop workflows.