One dashboard for every agent you are running
Once a business has more than one AI agent, the problem stops being intelligence and becomes management: who is doing what, under whose authority, at what cost. Paperclip is the control plane for that, and self-hosted it keeps the task history and the credentials on your own machine.
- Goals, roles and budgets
- Coordinates several runtimes
- Task history stays yours
- One dashboard
- Goal
- Clear the support backlog
- Assigned to
- Two agents, one reviewer
- Budget
- Set per run, enforced
- Status
- Tracked with full history
Paperclip does not run the agents itself. It gives them goals and budgets, routes the work through adapters, and records what happened.
A control plane, not another agent
It coordinates, it does not execute
Paperclip drives other runtimes through adapters rather than running the model work itself.
Goals and roles
Define what should be achieved and which agent plays which part, instead of prompting each one by hand.
Budgets that are enforced
Set a spend per run so an autonomous agent cannot quietly empty your provider account.
A history you can audit
Every task and its outcome recorded, which is the part that matters once agents make decisions.
The machines it runs on
Every plan is a virtual machine of your own with root access, billed monthly in rupees. The application is installed before we hand it over.
The control plane itself is light. What decides the machine is how many agents you coordinate and whether their runtimes live on the same server.
Basic 8 suits a small team running several agents through the dashboard. Move to Pro if the agent runtimes also run on this machine rather than elsewhere.
Prices are for the server. Paperclip is open source; the agents it coordinates and the models behind them are billed by whoever provides them.
The credentials are the point
A control plane holds the keys to every agent and every system they touch. That is precisely the thing worth keeping on a machine you own.
Adapter credentials stay local
The keys your agents use to reach models and systems live on your server, not in someone's SaaS.
Task history is yours
Outputs and decisions are recorded on your disk, where you can keep or delete them on your own terms.
No per-seat or per-run pricing
The machine costs the same whether one agent runs or twenty.
Runs beside your other agents
Coordinate agents on this machine, on another server, or wherever they already run.
Updated the way you choose
Pull updates when you are ready, or run it in a container for a cleaner upgrade path.
Private by default
Nothing about your workflows leaves the machine except the calls your agents make on purpose.
From an empty dashboard to work being done
- 1
We build the machine
You pick a size and a region; we build the server with Paperclip installed and send you the details.
- 2
Set up your organisation
Open the dashboard, add the adapters you will use, and give each agent a role.
- 3
Set goals and budgets
Describe what should be achieved and what it may cost, per agent and per run.
- 4
Watch and adjust
Track progress, read the history, and change roles or budgets as you learn what works.
We support the server and the installation. What your agents do, and what they are allowed to spend, is yours to decide.
The runtimes it drives
Adapters connect Paperclip to the tools that do the work, including anything reachable over a webhook.
- Claude Code
- OpenClaw
- Hermes
- Codex
- Cursor
- Anthropic Claude
- OpenAI GPT
- Google Gemini
- Local models
- Webhook adapters
- Custom processes
Several adapters can run at once, so agents on different runtimes can work on the same goal.
Other things we host and build
Paperclip AI Hosting, answered
What does Paperclip actually do?
It coordinates AI agents. You give it goals, assign roles, set budgets and watch the history; it routes the work to agent runtimes through adapters. It is not itself the thing that runs the model.
What does Paperclip actually do?
It coordinates AI agents. You give it goals, assign roles, set budgets and watch the history; it routes the work to agent runtimes through adapters. It is not itself the thing that runs the model.
Put your agents under one roof
Tell us which runtimes you are using and we will size the machine and build it with Paperclip installed.