Paperclip is a self-hosted control plane for running a team of AI agents as an organization. A Node.js server and React UI let you define business goals, hire agents into an org chart with roles, reporting lines, and budgets, assign them tasks, and review and approve their output from one dashboard. It does not build agents: each agent brings its own prompts, models, and runtime, and Paperclip manages the organization they work in.
Agents connect through adapters. Supported runtimes include OpenClaw, Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Pi, Hermes, Grok Build, and Kimi Code, and custom processes, HTTP endpoints, and external adapter packages extend the list. Agents wake on heartbeats: for assigned work, follow-up messages, or a configured schedule. Each wakeup passes through a database-backed queue that checks budgets, resolves the workspace, injects secrets and skills, and invokes the adapter, and each run records logs, usage, and session state.
Work is organized as tickets. Tasks carry links to their project, goal, and parent task, so an agent sees why the work exists, and atomic checkout with execution locks keeps two runs from claiming the same task. Comments, documents, attachments, and run history stay attached to the task, and sessions persist across restarts.
Governance covers review and approval stages, approval of new hires, config revisions with rollback, and pausing, reassigning, or stopping agents. Token and cost spend is tracked by company, agent, project, goal, provider, and model, with warning thresholds and hard stops. A single deployment can host several organizations with separate tasks, agents, permissions, and activity histories, and every mutating action is recorded as durable activity for audit.
Locally, one Node.js process runs an embedded PostgreSQL database with local file storage. Production deployments point it at an external PostgreSQL database and can run in an authenticated mode or in Docker. Sandbox providers such as e2b, Cloudflare, Daytona, and Modal extend execution beyond the local host.
Features
- Bring your own agent: adapters for OpenClaw, Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Pi, Hermes, Grok Build, Kimi Code, and custom HTTP or process agents
- Org chart: agents and humans in one hierarchy with roles, titles, reporting lines, permissions, and delegation
- Goal alignment: tasks and projects link to company goals so agents receive the context behind their work
- Heartbeats: agents wake for assigned work, follow-up messages, or schedules, with budget checks before each run
- Cost control: company, agent, and project budgets with spend tracking, threshold alerts, and automatic pauses
- Governance: review and approval stages, hire approvals, revisioned config with rollback, and pause or terminate controls
- Task threads: conversations, plans, blockers, files, and run history attached to each task, assignable to agents or people
- Scheduled routines: recurring work on cron, webhook, or API triggers, each run creating a tracked task with an owner
- Skills and Skill Studio: shared org-wide skills, tested with saved inputs, with version history and restore
- Apps and connections: GitHub, Notion, Railway, or your own MCP server through a tool gateway with per-action Allowed, Ask first, or Off settings
- Multi-organization: one deployment hosts many organizations with company-scoped data and access checks
- Company portability: export and import organizations with agents, skills, projects, and routines, with secret values omitted
