How-to
How to put every AI agent you run behind one dashboard in about two hours
Paperclip is an open-source server and UI that puts agent activity, month spend, pending approvals and failed run transcripts on one screen. The operations detail is the reason to install it, not the orchestration branding. Budget about two hours for the first agent, and skip it if nobody will open the dashboard.
Five point automations in five different tools means five places to look when one of them stops. Paperclip is an open-source attempt at a single place: a Node.js server and React UI that orchestrates a team of AI agents, with one dashboard for work, cost and failures.
What it actually gives you
The README says you bring your own agents, assign goals, and track work and costs from one dashboard. The dashboard docs are more concrete: overview cards for Agents Enabled, Tasks In Progress, Month Spend and Pending Approvals, live agent activity below them, and an agent detail page whose Runs tab shows the transcript of what went wrong.
That last one is the reason to care. Runs produce structured logs, cost events, session state and audit trails. OpenTelemetry auto-instrumentation for the server and Sentry error monitoring are both opt-in.
The two hour plan
I have not run this setup end to end, so treat every minute below as my estimate from the documentation, not a measurement.
- Read the README and the dashboard guide: about 20 minutes.
- Install the server and UI on a machine that stays awake: about 30 minutes.
- Bring in one existing agent, the noisiest one: about 30 minutes.
- Set its goal and a month spend cap: about 15 minutes.
- Turn on approvals for anything that sends or spends: about 15 minutes.
- Write a short runbook for whoever reads the dashboard: about 10 minutes.
Total: about 2 hours. The second and third agents go faster, because the server is already up.
Decide before you install
| If this is true | Do this |
|---|---|
| Three or more automations, no shared log | Install it, one agent first |
| Agents spend money or send messages | Install it, approvals are the point |
| One automation, one owner, one tool | Skip it, that tool's own log is enough |
| Nobody will open a dashboard daily | Skip it, you need alerts, not a screen |
The arithmetic on your own inputs
Count the tools you open to confirm your automations ran. Multiply by the minutes each check takes, then by working days. Four tools at five minutes each is twenty minutes a day, about seven hours a month. A dashboard does not delete that time, it concentrates it into one screen, so assume you keep a third of it back. More on where automations quietly fail.
Limitations
- I have not tested the install, the agent adapters or the spend accounting. The steps come from the primary sources.
- The docs do not promise it fits every agent you already run. If your lead bot lives inside a closed builder, expect glue work, or no path at all.
- Self-hosting means you own the uptime. A monitoring tool that is down is worse than none.
- Telemetry and error monitoring are opt-in, so a default install watches less than you might assume.
- Community sentiment around agent control planes is warm. Sentiment is not evidence.
My take
Install it as an operations dashboard, not as an orchestrator. The parts I trust from the docs are boring and useful: spend caps, pending approvals, run transcripts, audit trails. My guess is that most small teams will use it to answer one question, which agent broke and when, and never touch the goal assignment. If nobody on your team will open it on a normal Wednesday, do not install it.
Questions people ask
Do I have to rebuild my automations to use Paperclip?
The README says you bring your own agents, so the intent is to point existing agents at it rather than rewrite them. It does not promise that every builder or closed platform has a path in, and I have not tested any adapter, so check your noisiest automation first before you plan a second one.
Will it tell me when an agent fails, or do I have to look?
The dashboard shows live agent activity and the agent detail page has a Runs tab with the transcript of what went wrong, which is a screen you read. Error monitoring through Sentry is opt-in. If you need to be told rather than to look, turn that on and treat the dashboard as the place you investigate, not the place you get alerted.
Is two hours realistic?
That is my estimate from the documentation, not a measured install, and the project does not publish a setup time. Reading and the first agent are the two steps most likely to run long. If you have never deployed a Node.js server before, double the install block and do it on a day when nothing is on fire.
Have a process like this leaking time or leads?Thirty minutes, no deck, and you leave with a plan.
Book a call