you need to try Paperclip RIGHT NOW!

Summary

NetworkChuck demonstrates Paperclip, an open-source coordination layer for managing multiple AI agents as an organization. Instead of having agents chat freely in a shared “conference room,” Paperclip coordinates them through assigned tasks, reporting lines, and shared work. It can bring together agents running on different systems—including Claude Code, Hermes, Codex, and local-model tools—so teams can use different models and capabilities in one workflow.

Setup and team management

The video walks through installing Paperclip on an Ubuntu virtual machine. The presenter notes that a local Mac or Linux system, or Windows running WSL 2, can also be used for experimentation. For a more persistent setup, he suggests using a separate VM or cloud server.

Paperclip is accessed in a browser at the server’s address on port 3100. The presenter configures a Claude Code agent as CEO, then installs Claude Code when the initial setup fails because it is missing.

Remote Hermes agents are added through an onboarding and approval process. The presenter builds a hierarchy with a CEO overseeing a CTO and specialist agents, then adds workers running other agent systems.

Paperclip provides an organizational chart, task views, timelines, and activity tracking. People can approve hires and decisions, while agents can assign tasks to one another—including through Paperclip skills used outside the main interface.

Tasks can have parent and blocking relationships, reviewers, approvers, monitors, and supervisors. Agents can escalate obstacles to their managers rather than immediately asking a human to intervene.

Investigation example: The studio network

The team is assigned to investigate a recurring studio problem: flushing the toilet appears to disconnect people from the NAS. The agents divide the work among tasks such as mapping the network, checking switches and storage, reviewing logs, and examining security.

Paperclip’s artifacts section stores outputs such as reports and network maps. The presenter also highlights experimental “cases” for organizing an investigation, while noting that agents and workflows are not always flawless and that some tasks encounter timeouts.

The team’s initial findings did not prove that the toilet caused the outages. A later investigation pointed to a likely combination of environmental disturbance and unreliable third-party SFP optical transceivers. Four fiber links appeared unusually prone to losing synchronization, and the affected modules reportedly came from a batch with a high failure rate.

The presenter replaced them with MikroTik-branded modules but said he would monitor the system for a week or two before declaring the problem resolved. He still suspected that activity around the restroom might be contributing to the disturbance.

Other Paperclip features

• Routines: Scheduled or manually triggered tasks. The presenter creates a daily storage-health check that reports remaining space and asks for a proposed response if capacity gets low.
• Secrets and variables: Credentials can be stored and granted to specific agents. In a separate demonstration, a Flare API key and tenant ID are made available to an agent, which then launches investigative tasks. The presenter pauses work when an agent’s actions need to be stopped.
• Import and export: An organization—including agents, routines, tasks, and attachments—can be exported and moved to another Paperclip instance.
• Stand-ups: A scheduled routine has agents report on their work and ask one another follow-up questions.

Review and practical takeaway

The video presents Paperclip as a way to give separate AI agents shared goals, responsibilities, oversight, and a record of their work. Its strengths in the demonstration include cross-system coordination, task delegation, traceability, and recurring workflows.

The presenter also acknowledges that agents may make mistakes, need supervision, or require tasks to be paused. The network investigation illustrates both the potential of coordinating multiple specialists and the importance of treating their conclusions as evidence to review—not as automatic proof.

Guides and demonstrations

1. Paperclip installation and initial setup on Ubuntu, with notes for Mac, Linux, and WSL 2.
2. Connecting and hiring agents, including Claude Code and remote Hermes agents, and assigning them organizational roles.
3. Delegating and monitoring a multi-agent investigation using tasks, projects, artifacts, and timelines.
4. Creating scheduled routines for recurring checks and stand-ups.
5. Managing agent access and work with secrets, approvals, and pause controls.
6. Integrating Flare by providing an agent with API credentials and asking it to run checks.

Main speakers and sources

• NetworkChuck — presenter and demonstrator.
• Dotta — creator of Paperclip, briefly featured in the video.
• Paperclip agents — their tasks, reports, and recommendations are shown as examples of the product’s workflow.
• Flare — sponsor, demonstrated as an external threat-analysis service integrated with an agent.