Paperclip AI Pricing: Is It Free? The Real Cost of Running an Agent Company

Paperclip AI costs nothing to download and something every single day you run it. The software is MIT-licensed, self-hosted and has no account, no seat fee and no paid tier as of September 2026. The real bill is hosting, model API usage and your own time. This post puts numbers on all three, shows how Paperclip's per-agent budgets keep the model bill from surprising you, and compares the total against hiring help. For background on what the platform actually does, start with our Paperclip AI guide.

What is free

All of it, as far as the software goes. The full Paperclip server, dashboard, CLI, adapters, plugin system and documentation are open source under the MIT license from Paperclip Labs, Inc. There is no community edition versus enterprise edition split, no feature gate and no telemetry requirement. You do not create an account with Paperclip to run it. The install command is one line and needs only a model API key to do anything useful.

Third-party marketplaces sell company templates and skill packs for it, and hosting companies sell VPS images with it preinstalled. Those are optional add-ons from other vendors, not Paperclip pricing.

The three real costs

Cost	What drives it	Typical range
Hosting	A VPS with Node and Postgres, sized by agent count	$0 on a laptop to $150 per month for a large instance
Model usage	Tokens consumed on every heartbeat, tool call and ticket. Scales with agent count, heartbeat frequency and model choice	Tens to thousands of dollars per month
Your time	Setup, writing job descriptions and skills, reviewing approvals, debugging stuck agents	10 to 40 hours in the first month, then a few hours a week

Hosting is covered in detail in our hosting guide. Model usage is where the money goes, so the rest of this post focuses there.

Worked monthly scenarios

These are estimates to size your expectations, not quotes. Assumptions are stated so you can swap in your own numbers.

Shared assumptions: agents run 12 hours a day, 22 working days a month. Blended model pricing of roughly $3 per million input tokens and $15 per million output tokens for a frontier-class model, and roughly $0.80 in and $4 out per million for a smaller model. A "light" heartbeat that finds no work costs about 2,000 tokens. A working ticket costs 40,000 to 200,000 tokens depending on how many tool calls it takes.

Setup	Agents and roles	Heartbeat	Hosting	Model usage	Total per month
Starter	3 agents: research, content drafting, reporting. Small model for two, frontier for one. About 8 real tickets a day total.	Every 15 minutes	$10	$60 to $180	$70 to $190
Working team	10 agents: a manager, two coding agents, two content agents, QA, SEO, outreach, monitoring, reporting. Mixed models. About 40 tickets a day.	Every 10 minutes	$30	$400 to $1,400	$430 to $1,430
Agent company	25 agents across three departments with approval gates. Frontier models on lead roles. About 120 tickets a day.	Every 5 to 10 minutes	$110	$1,800 to $6,000	$1,900 to $6,100

Three observations from running numbers like these:

• Idle heartbeats are cheap; work is not. A 10-agent team on 10-minute heartbeats makes about 15,800 wake-ups a month. At 2,000 tokens each that is 32 million input tokens, which is under $30 on a small model. The tickets are what cost money.
• Model choice is the biggest lever. Moving routine roles to a smaller model cuts the middle scenario roughly in half. Paperclip is model-agnostic, so mix them by role.
• Output tokens dominate coding and writing agents. Budget those roles higher than monitoring or triage roles.

How budgets and auto-pause control spend

This is the feature that makes the numbers above manageable. Every agent in Paperclip has a monthly budget in US dollars. Paperclip tracks cost per company, per agent, per project, per provider and per model, and when an agent reaches 100 percent of its budget it is paused automatically. It does not finish "one more ticket." It stops until a human raises the limit or the month rolls over.

Practical use:

1. Start every new agent at a small budget, $20 to $50, for its first month.
2. Watch the cost view after a week. Agents that pause early are either doing more than you expected or doing it inefficiently. Both are worth knowing.
3. Raise budgets by role, not across the board. A manager agent that delegates should have a small budget. Its reports do the expensive work.
4. Set approval gates on tickets that trigger large jobs so a human signs off before a big spend.

Combined, budgets and approvals mean your worst month is bounded by the sum of your budgets, which is a sentence no other agent framework we have used can offer. The orchestration explainer goes deeper on how budgets, heartbeats and approvals interact.

Subscription harnesses vs API keys

Some adapters, notably Claude Code, can run on a flat-rate subscription plan rather than a metered API key. That changes the math:

	Metered API key	Subscription harness
Cost model	Pay per token, unbounded except by budgets	Flat monthly fee with usage limits and rate windows
Best for	Many agents with predictable, modest workloads; HTTP and OpenCode adapters	One or two heavy coding agents that would otherwise cost more than the plan
Risk	Surprise bills if budgets are not set	Hitting rate limits mid-task; terms of service may restrict automated use, so check them
Paperclip cost tracking	Accurate per token	Estimated, since the provider is not billing per call

Many teams do both: a subscription for the lead coding agent and API keys through OpenRouter for everything else. Our Claude Code integration guide covers the setup.

Hidden costs

• Learning curve. Reviews of the platform consistently praise the vision and dashboard and consistently mention bugs and a real ramp-up period. Expect the first two weeks to be mostly configuration.
• Debugging stuck agents. An agent that loops on a ticket burns tokens and your attention. Budgets cap the tokens; nothing caps the attention except good job descriptions and skills.
• Frequent updates. Releases are date-stamped and frequent. Each one is a small maintenance task.
• Integration work. Getting agents into Slack, Linear, your CRM or your data warehouse is where value comes from and where hours go.
• Review time. Approval gates only save money if someone actually reviews them promptly.

Hosted cloud and third-party options

Paperclip Labs has a hosted cloud version on its roadmap, but it was not generally available as of mid-2026 and no first-party pricing has been published. In the meantime, hosting companies including Hostinger, Contabo and xCloud sell VPS plans with Paperclip preinstalled, typically at their normal VPS rates plus nothing for the software. Third-party marketplaces sell company templates and skill packs for one-off fees. None of these change the core picture: the platform is free, and usage is what you pay for.

If you are comparing against other tools, our Paperclip AI alternatives roundup lists which competitors are also free and which charge per seat.

Paperclip vs paying an agency

The honest comparison is not "free software versus paid service." It is "your hours plus model spend versus someone else's hours plus model spend." A solo operator with technical comfort and a clear use case can run the starter scenario for under $200 a month and a few hours a week. A business without that person will spend more in staff time in the first month than a setup engagement costs, and will end up with agents that nobody trusts enough to leave running.

A reasonable middle path is to pay for setup and the first month of tuning, then run it in-house with a monthly check-in. That gets you working job descriptions, sane budgets, Slack alerts and a backup plan, without a permanent retainer.

Frequently asked questions

Is Paperclip AI free?

Yes. It is open source under the MIT license, self-hosted, and requires no account or subscription. You pay for hosting and for the model API usage your agents generate.

How much does Paperclip AI cost per month?

For a 3-agent starter setup, roughly $70 to $190 a month including hosting. A 10-agent team typically runs $430 to $1,430. A 25-agent company can reach $1,900 to $6,100. Model choice and heartbeat frequency drive most of the range.

Does Paperclip have a paid or enterprise plan?

Not as of September 2026. A hosted cloud version is on the roadmap. Third-party hosts and marketplaces sell related services, but the software itself has one edition and it is free.

Can Paperclip stop an agent from overspending?

Yes. Each agent has a monthly USD budget and is paused automatically at 100 percent. Cost is tracked per company, agent, project, provider and model.

Is it cheaper to use a Claude Code subscription instead of an API key?

For one or two heavy coding agents, often yes. For many light agents, API keys through a provider such as OpenRouter are usually cheaper and easier to track. Check the subscription terms for automated use.