Paperclip explained: an operating layer for teams of AI agents
Paperclip quickstart: one company, one agent, one review gate
Try the documented onboarding flow on loopback with a disposable task.
What you will learn
- Prepare a local trial
- Create the smallest operating loop
- Check an approval and a stop
Before you start
- Node.js and agent runtime basics
- A test company with non-sensitive tasks
Turn the source review into a bounded experiment with stop conditions.
Key takeaways
- Use a disposable first company.
- Connect one agent before designing a fleet.
- Verify the run and its approval record separately.
Prepare a local trial
The README documents a CLI onboarding path and a default trusted-local loopback mode. Use a disposable data directory and read the prompts before accepting them. A test machine is safer than treating a first run as a public service.
A connected agent may need its own credentials and filesystem access. Keep those permissions narrow and use a task that cannot alter production data.
Create the smallest operating loop
Define a company, a goal and an issue. Attach one agent that can receive a heartbeat, then pause to inspect the proposed work before allowing further execution. Save the run ID and the resulting work item.
The success condition is a traceable issue-to-run-to-result path, not simply a green dashboard. If the agent never wakes, inspect schedule, adapter and queued work separately.
Check an approval and a stop
Put a review gate before an externally visible change. Set a small budget and observe what happens when the allowance is exhausted in a safe test. Do not assume a warning threshold is identical to a hard stop.
This article gives a test plan, not a report of a completed Paperclip run. The commands should be checked against the pinned release you install.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Choose the local loopback onboarding mode.
- 2
Create one synthetic issue and record its run ID.
- 3
Exercise an approval and a budget stop in the test instance.
Copy-ready example
pnpm paperclipai onboard
# choose local_trusted and loopback for a local trialFrequently asked questions
Can I expose the default mode to colleagues?
The documented local_trusted mode is loopback-only and has no login; use authenticated mode for shared access.
Was this quickstart executed here?
No. It is derived from the fixed README and deployment guide.
Sources
- Paperclip / README.mdSource checked 2026-09-26
- Paperclip / docs/deploy/deployment-modes.mdSource checked 2026-09-26