Deutsch
Paperclip AI coordinates work across multiple AI agents. It gives you a place to assign tasks, define roles, review progress and track costs, instead of managing each agent in a separate conversation.
It becomes useful when coordination is taking more effort than the individual tasks. A research workflow, for example, might need one agent to gather sources, another to compare findings and a reviewer to check the report. Paperclip supplies the management layer; the agents and their tools still perform the work.
For installation, see the Paperclip self-hosting guide.
What Paperclip runs
The Paperclip project is open source. Its main repository describes a Node.js server and React interface for coordinating agents, goals, tasks and costs.
It uses roles and reporting relationships to organise work. Those labels help describe responsibilities, but each task still needs a concrete deliverable and acceptance criteria. Calling an agent a reviewer does not establish that its review is independent or correct.
The server retains coordination state and connects to agent runtimes through adapters. Check the supported adapter and authentication method for each tool you intend to use, rather than assuming that every agent can be connected in the same way.
How a workflow fits together
Start with a goal that can be evaluated, then divide it into tasks with clear ownership and dependencies. Assign an appropriate agent and define what it should produce, what it may access and when a human decision is needed.
As agents work, the dashboard provides a shared view of assignments, results and costs. Review the actual artifacts and evidence before accepting a task. A status marked complete is useful coordination information, but it cannot replace checking the output.
For shared repositories or documents, agree how agents will avoid conflicting edits. Separate workspaces, branches or task-owned files can help. Plan how results will be combined and who resolves disagreements or failures.
Roles, budgets and review
Paperclip combines several management functions:
- Task and goal organisation.
- Agent roles, assignments and reporting relationships.
- Connections to supported agent runtimes.
- Scheduled work and activity tracking.
- Cost visibility and budget controls.
- Approval and intervention points.
Test those controls with your chosen adapters. Confirm which spending is reported, when limits take effect and what happens to a run already in progress. Keep provider-side usage limits where appropriate; an orchestration budget should not be assumed to cover every external service charge.
Use review gates for actions such as publication, customer messages, financial changes or production deployments. An agent can prepare the result and supporting evidence while the responsible person retains the final decision.
Use cases for a coordinated agent team
Software development
Separate investigation, implementation, testing and documentation into tasks that can run independently where practical. Give each agent the repository instructions and a known starting state. Review code, regression evidence and the affected user flow before merging.
Avoid parallel work merely to increase the agent count. Two agents changing the same module can create more integration work than they save.
Research and reporting
Assign source gathering, comparison and fact checking to distinct tasks. Require citations and preserve contradictory evidence. A final report should explain what the sources support, including gaps, rather than simply combine the agents' summaries.
Marketing drafts
Coordinate briefs, copy drafts, SEO suggestions and performance summaries. Keep brand claims, factual checks and publication approval with an editor. A generated social post is a draft until someone authorises sending it.
Support triage
Use agents to classify tickets, draft replies and identify documentation gaps. Define escalation rules for uncertain cases and limit access to the customer data each task needs. Check the destination and account before any outbound response.
A solo operator's task queue
One person can use the dashboard to track several specialised workflows. Start with a small number of roles and clear deliverables. Extra layers of agent management are only useful if they make the work easier to understand and verify.
How it differs from a personal assistant
OpenClaw and Hermes Agent focus on assistant workflows with tools and persistent context. Paperclip focuses on organising work across agents. The products can overlap or integrate, so choose according to the coordination problem you actually have.
For one occasional assistant task, a single runtime may be simpler. For several agents with dependencies, recurring work and spending to track, a shared task system can make ownership and progress easier to follow.
When to host Paperclip on a VPS
A VPS can keep the server and dashboard available independently of a laptop. It provides a consistent place for configuration, task state, adapters and logs, with remote access for the people operating the workflows.
Size the full deployment. Agent processes, browser sessions, builds, repositories and supporting services may consume much more capacity than the dashboard. If agents run elsewhere, include their hosting, connectivity and model usage in the cost estimate.
Protect the administrative interface, use individual accounts where supported and restrict credentials by role. Keep secrets out of task descriptions and source control. On a self-managed VPS, your team maintains Paperclip, its dependencies, the host and the connected agent software.
Follow the VPS setup guide, then test the actual adapters and workload before relying on the deployment.
Prepare for failed or repeated work
Record what should happen when an agent stops, exhausts its budget or returns an unusable result. Decide which tasks can be retried safely and which need a human check before another attempt.
Keep task history and relevant logs, with retention rules for private data. Back up persistent application state and required configuration, then test restoration. Verify that restarting the server does not duplicate external actions or lose track of an active task.
A useful first trial is one bounded workflow from assignment through review. Confirm the result, recorded cost, approval path and recovery behaviour before expanding the team.
Frequently asked questions
Is Paperclip a chatbot?
Its main role is coordinating agent work through tasks, roles and a shared interface. It connects to agents that perform the underlying tasks.
Does it replace code or editorial review?
No. It can organise review assignments and evidence, but the result still needs to be checked against the original requirement.
Why run it on a VPS?
A compatible VPS can host the persistent server, dashboard and supporting processes. The benefit is continuing access and coordination; the tradeoff is maintaining another service.
Does adding more agents always help?
No. Parallel work helps when tasks can be separated and their results combined. Overlapping assignments, repeated research and conflicting edits can increase cost and delay.
