Paperclip
Choose Paperclip when you want software to run a team of AI agents for you: roles, goals, budgets and schedules, with people approving and reviewing what the agents do.
Paperclip vs PAPI checked October 7, 2026
Paperclip runs your AI agents. PAPI keeps the people using them on one plan.
Paperclip is open-source software that runs AI agents like a company: roles, reporting lines, goals, budgets and schedules. PAPI runs no agents. It keeps one shared plan, its decisions and conventions, a handoff for each build and a review step, so people who each use their own AI tool keep building the same thing.
Quick answer
Choose by the part of the work you need to carry forward.
PAPI
Choose PAPI when people do the building, each in their own AI tool, and you need one shared plan, recorded decisions and a review step so their work stays aligned.
See what PAPI doesHow each product works
Two ways to give AI-assisted work more structure.
Paperclip
Paperclip is MIT-licensed, open-source software made of a Node.js server and a React dashboard. It describes itself as orchestration for teams of AI agents: you set a goal, add agents with roles and reporting lines, give them budgets and approve the plan. Agents wake when work is assigned, when someone follows up or on a schedule, and run in whichever agent runtime you connect.
Around that sit task threads, goal tracking, review and approval stages, cost tracking with hard limits, connections to services and MCP servers, and an activity log. Several organizations and several human users can share one self-hosted deployment.
PAPI
PAPI is a hosted project workflow that each person reaches from their own AI tool over MCP. It holds one plan for the project: the backlog, the current cycle, the decisions made and why, the conventions every build should follow, a handoff for each task and a review of each finished build.
PAPI does not run agents or models for your work. Planning and reviews happen in the AI session you already have open, on your own subscription. PAPI stores structured project records such as tasks, decisions and cycle notes, and never reads or stores your source code.
At a glance
Paperclip and PAPI side by side
A feature name rarely tells you how work feels day to day. This table summarizes the practical difference in each workflow.
| What to compare | Paperclip | PAPI |
|---|---|---|
| Core job | Runs a company of AI agents: an org chart with roles and reporting lines, goals, tasks, budgets and schedules. | Keeps one project plan that every person builds from: tasks, cycles, decisions, conventions, handoffs and reviews. |
| Who does the work | AI agents that Paperclip wakes and runs. People set goals, approve, review and can take tasks themselves. | People, each working in their own AI tool. PAPI gives them the context; it does not start or run any agent. |
| AI tools | Bring your own agents. Supported runtimes include Claude Code, Codex, Cursor, Gemini CLI, OpenCode and OpenClaw, plus HTTP endpoints and custom processes. Models are chosen per agent. | Any MCP-compatible AI tool, and each person can use a different one: for example Claude Code, Cursor, Codex, VS Code with GitHub Copilot or Windsurf. |
| How work starts | Agents wake for assigned work, follow-up messages or configured schedules. Routines create recurring tasks from cron, webhook or API triggers. | A person opens their AI tool and asks PAPI what is next. PAPI returns the current cycle, the next task and its build handoff. |
| Staying aligned | Tasks link to project and company goals, so each agent receives the goal behind its work. | Decisions stay on record, and settled conventions are written into every future build handoff, whichever AI tool picks up the task. |
| Oversight | Review and approval stages, hire approvals, budget hard stops, and pause, reassign or stop work at any time. | Each build is reviewed against its handoff before it is accepted, and what the review finds feeds the next plan. |
| Model spend | Tracks token and cost usage by company, agent and project, with alerts and pauses at limits you configure. | No model spend to manage: PAPI makes no model calls for your work. Each person’s AI tool bills them as it already does. |
| Where it runs | Self-hosted. Runs locally with an embedded Postgres database, or on your own Postgres or Docker setup in production. A hosted Paperclip Cloud has a waitlist. | Hosted. Each person connects their AI tool to the project; there is no server or agent runtime to operate. |
What the difference means
Look beyond the feature names.
01
Agents as the workforce versus people as the workforce
Paperclip
Paperclip treats AI agents as employees. You define a goal, add agents with roles, titles and reporting lines, set budgets and approve the plan. Agents then pick up tasks, delegate along the org chart and report back, while people supervise from the dashboard.
People are part of the org chart too: tasks can be assigned to people, and several human users can share a deployment. Most of the product is still built around the agents it runs.
PAPI
PAPI starts from the people. Each person keeps the AI tool they already chose and works in it as usual. PAPI is the shared project they all connect to: the plan for this cycle, the task in front of them and the handoff that says what is in scope and how the work will be judged.
PAPI does not start agents, choose models or schedule work. Work happens when a person picks up a task in their own tool.
Practical takeaway: If you want software to run the agents, look at Paperclip. If the people on your team run their own AI tools and you need them working from one plan, look at PAPI.
02
One plan when every person uses a different AI tool
Paperclip
Paperclip handles mixed tools at the agent level. Each agent can use a different runtime and model, and Paperclip keeps the team’s tasks, skills, permissions and history in one place.
That fits when agents do the work and Paperclip is where they are run from. The goal context each agent sees comes from its task, its project and the company goal above them.
PAPI
PAPI handles mixed tools at the person level. One developer might use Claude Code, another Cursor, another Codex. Each connects to the same PAPI project over MCP and gets the same plan, the same recorded decisions and the same conventions.
The handoff is stored with the project rather than in one person’s chat, so a task built in one tool can be reviewed by a teammate working in another, against the same scope.
Practical takeaway: Both let different AI tools work side by side. Paperclip does it by running each agent; PAPI does it by giving each person the same plan.
03
Budgets and approvals versus handoffs and reviews
Paperclip
Paperclip’s controls are built for agents that act on their own. Budgets can be set per company, agent and project, with threshold alerts and hard stops that pause work when a limit is reached.
Review and approval stages, hire approvals and an activity log let people see what agents did and step in: pause, reassign or stop work.
PAPI
PAPI’s controls are built for people working with AI. Before a build, the handoff sets the scope and the acceptance criteria. After it, a review records whether the work matched, what turned up along the way and what should change in the next plan.
There is no model spend to cap, because PAPI makes no model calls for your work. What gets checked is whether the build did what the team agreed.
Practical takeaway: Paperclip keeps autonomous agents within budget and approval rules. PAPI keeps people’s builds within the scope they agreed.
Cost and setup
Count the software, the model usage and the hosting
Paperclip is free, open-source software that you host, and your agents bill through their own model providers. PAPI is a hosted service with a free plan; each person’s AI tool is billed separately.
Paperclip
Open source under the MIT license: $0 for the software. You host it and pay for your agents’ model usage.
- Paperclip installs with one command and runs locally with an embedded Postgres database. In production you point it at your own Postgres or deploy it with Docker.
- Agents bring their own models and runtimes, so token costs come from the providers you connect. Paperclip tracks that spend and can pause work at budget limits you set.
- No Paperclip account is needed to self-host. Paperclip Cloud, a hosted version, has a waitlist.
PAPI
Free: €0 for up to three projects. Pro: €20/month founding price for the first 500 Pro builders; €29/month regular price shown.
- Pro adds unlimited projects and a deeper read on each one. On yearly billing the founding price is €16/month.
- Team adds shared projects, roles and a shared task pool, at custom pricing during early access. Viewer seats are free, so you pay only for people who build.
- PAPI has no credits or usage bills. Each person’s AI tool or model plan is billed separately, as it is today.
Paperclip license, install and hosting details checked October 7, 2026 against its official README and npm packages. PAPI prices are in EUR and checked the same day; the founding price applies to the first 500 Pro builders. Prices and plans can change.
Choose by your workflow
The right fit depends on what you want the system to own.
Paperclip is a better fit when…
- You want AI agents to carry out work on their own, on a schedule or when tasks are assigned, with people approving and reviewing.
- You run many agents across different runtimes and need roles, reporting lines and per-agent budgets to keep them in check.
- You want open-source software you host yourself and control where it runs.
PAPI is a better fit when…
- Your team is people, each building with their own AI tool, and you need everyone working from the same plan and decisions.
- You want every task to come with a handoff that sets scope and acceptance criteria, and a review before the work is accepted.
- You don’t want to host or pay for an agent runtime: PAPI is hosted and makes no model calls for your work.
Using both can make sense when…
- Paperclip runs agents for recurring or operational work, while the people building the product plan and review their own work in PAPI.
- You agree which system owns which tasks. There is no built-in sync between them, so track each piece of work in one place.
- You want one AI tool to reach both. Paperclip and PAPI each publish an MCP server, so a single MCP-compatible client can connect to both.
Questions people ask
Paperclip vs PAPI: FAQs
Does PAPI run AI agents?
No. PAPI does not start, schedule or pay for agents or models. Each person plans, builds and reviews in their own AI tool, and PAPI supplies the plan, the handoff and the place to record the result. The one optional exception: if you upload a brief file during onboarding, PAPI sends its text to Anthropic’s API once to pull out its structure, as the trust page explains.
Can I use Paperclip and PAPI together?
Yes, as long as each owns different work. One split is Paperclip running agents for recurring or operational jobs while the people building the product plan and review in PAPI. There is no built-in sync, so agree which system tracks which tasks. Both publish MCP servers, so one MCP-compatible AI tool can connect to both.
What does each cost?
Paperclip’s software is free under the MIT license; you pay for wherever you host it and for the model usage of the agents you run. A hosted Paperclip Cloud has a waitlist. PAPI is free for up to three projects. Pro is €20/month at the founding price for the first 500 Pro builders (€29/month regular), and Team, which adds shared projects and roles, has custom pricing during early access. PAPI bills no model usage; each person’s AI tool is paid for separately.
Is either open source?
Paperclip is open source under the MIT license and self-hosted. PAPI is a hosted service. The MCP server PAPI installs on your machine is source-available under the Elastic License 2.0 in the public getpapi/papi repository, so you can read what runs locally before you connect it. Elastic License 2.0 is not an open-source license.
Does PAPI see my code?
No. PAPI stores structured project records: task titles, cycle notes, decisions and the other things you see on your board. It never reads or stores your source code. If you register a document, only Markdown files are stored, and that is opt-in. Paperclip, being self-hosted, keeps its own data in the database you run it on.
Which suits a team where everyone uses a different AI tool?
It depends on who does the work. If agents should run on their own with different runtimes, Paperclip gives each agent its runtime and model and coordinates them. If people do the building and each prefers a different AI tool, PAPI connects everyone to the same project, so the plan, decisions and conventions are the same whichever tool picks up a task.
Compare other options
Still weighing up your workflow?
Evidence
Sources checked October 7, 2026
Product details and pricing notes reflect the linked official pages at the time checked. Vendor plans and features can change.
Keep your work moving
Give your next build a clear brief.
Bring your project context into the AI client you already use and start with a plan you can review.
Connect PAPI
Free on three projects · no card · 14-day refund on Pro