$ show paperclipai/paperclip
Paperclip is a full-stack, self-hostable platform for orchestrating AI agents in a work management context. The server (Node.js/Express + PostgreSQL) provides a control plane for hiring agents (Claude Code, Codex, Cursor, Gemini, etc.), assigning tasks (issues), enforcing budgets and approvals, and managing workspaces. A Rust native runner executes agent sessions with a rich protocol, while adapters bridge to each AI provider's specific APIs. The React UI offers dashboards, task boards, agent management, and administrative features. The monorepo also includes a CLI, plugin SDK, MCP servers, extensive E2E/unit test infrastructure, and a skills catalog. While primarily an application, the repository contains data-intensive subsystems like telemetry collection, cost tracking, and a decision-training example store that blur into data engineering.
The LLM identified 3 files worth highlighting for reuse:
Implements a frame-based codec for duplex streams, handling variable-length frames with length-prefix encoding. Well-tested (vector tests, round-trip fuzzing), zero internal dependencies beyond Node.js buffers, and could be extracted into any Node.js project needing structured stream framing.
Provides a type-safe factory for defining plugins with manifest capabilities (UI slots, API routes, environment drivers, sandbox providers). Demonstrates a clean inversion-of-control pattern where the plugin declares its surface area and the host injects services. Could be reused in any extensible application needing a plugin architecture.
Defines a canonical semantic action for fetching approval context within the agent execution protocol. Each action is a self-contained module with its own schema, handler, and test fixtures. The pattern (type-safe, versioned, transport-agnostic) is reusable for building any domain-specific RPC protocol with strong contracts.