Promptbook versus Paperclip
When to choose Promptbook
When you want to deploy a specific business operation and hand it over to the people who will run it. They get roles, rules and tasks in readable project files with Git history. An implementation partner can help.
When to choose Paperclip
When you want to run agents as a whole organization with reporting lines, tasks tied to company goals, and monthly budgets per agent.
| Feature | Promptbook | Paperclip |
|---|---|---|
| Tasks and chats in project Git | Yes.book files in one repository | PartialGit for working files, tasks in Paperclip |
| Self-hosting | YesYour computer or server | YesOwn instance |
| Runs with your computer off | ConditionalWhen hosted elsewhere | ConditionalWhen hosted elsewhere |
| Where your data lives | On your host: project files with Git change history. | On your host: task database and local files; cloud option also available |
| AI models | Claude, GPT, Gemini, Qwen, … including local. Automatic choice from available models for the task. | Claude, GPT, Gemini, Qwen, … depending on connected tool |
| Scheduled tasks | YesRecurring agent tasks | YesSchedules and events |
| Integrations | Universal browser. Web apps even without a connector. | Connectors, plugins and connected-agent tools |
| Passwords and credentials | Local encrypted Wallet; passwords outside the agent’s context. | Locally encrypted secrets; passed to the agent process at runtime |
| Trust model | Zero Trust: E2E App–engine. The relay cannot read working content. | Own server; authentication and HTTPS/VPN by deployment |
| Mobile and remote control | YesApp over E2E to your own engine | ConditionalWeb; configure HTTPS or Tailscale when self-hosting |
| AI tools and selection | Codex, Claude Code, Gemini CLI, Qwen Code, OpenCode, … Automatic choice by task and availability. | Claude Code, Codex, Gemini CLI, OpenCode, … by adapter |
E2E protects the App–engine connection. External AI and services receive authorized inputs. Git keeps file history; it does not encrypt files or undo external actions. Wallet stays outside Git. Data and security ↗ Connection architecture ↗
Feature details
Self-hosting
PromptbookThe engine that runs and schedules agent work lives on your computer, company server or VPS under your control.
PaperclipPaperclip can be self-hosted. Adapters can run agents locally or call them through a remote interface.
Runs with your computer off
PromptbookClosing the controls leaves work running. Shutting down the engine host stops it. An engine on a running server can continue.
PaperclipWork depends on Paperclip staying up and the specific agent runtime being available.
Where your data lives
PromptbookAgents, tasks and conversations stay in project files on your host, with an auditable Git change history. Git does not encrypt files or undo actions in external services; Wallet stays outside it.
PaperclipOn your host: task database and local files; cloud option also available
AI models
PromptbookClaude, GPT, Gemini, Qwen and other models through supported tools and extensible integrations, including local models depending on the adapter and task. You are not tied to one vendor. External models receive the inputs they need.
PaperclipClaude, GPT, Gemini, Qwen, … depending on connected tool
Scheduled tasks
PromptbookGive a team of agents an ongoing responsibility, goal and rules. The engine runs one-off and recurring tasks.
PaperclipScheduled or event-driven agent wakeups, assigned tasks and budgets.
Integrations
PromptbookA universal browser operates web apps much like a person, including services without a ready-made connector or API. Capabilities depend on access, permissions and each service’s restrictions.
PaperclipConnectors, plugins and connected-agent tools
Passwords and credentials
PromptbookWallet is an encrypted credential vault on your host, separate from Git. A protected integration fills verified sign-in forms without putting passwords in the agent’s context. Protection requires keeping Wallet isolated from agent tools.
PaperclipLocally encrypted secrets; passed to the agent process at runtime
Trust model
PromptbookZero Trust for the App–engine connection: working content is encrypted between endpoints, and neither the relay nor central backend decrypts it. External AI and tools receive authorized inputs; account and operational metadata remain separate.
PaperclipOwn server; authentication and HTTPS/VPN by deployment
Mobile and remote control
PromptbookUse the control App to assign work, follow chats and manage the project over an authenticated E2E connection to your own engine. The local Console has separate access on the engine host. Closing the App does not stop the operation.
PaperclipWeb; configure HTTPS or Tailscale when self-hosting
AI tools and selection
PromptbookCodex, Claude Code, Gemini CLI, Qwen Code, OpenCode, GitHub Copilot, … A shared interface supports further integrations. Promptbook automatically chooses a suitable available tool and model for the task, capabilities and configured accounts, respecting your choices and limits.
PaperclipClaude Code, Codex, Gemini CLI, OpenCode, … by adapter
Core features reviewed 10 November 2026. Integrations, models and access reviewed 11 November 2026.