Latest release v2026.1012.0 is live (October 12, 2026).

See what's new
Product

Everything you need for an agentic company

Paperclip is an open-source control plane for teams that run more than one AI agent. Every agent, whether it runs on Claude Code, Codex or your own runtime, gets a role on one org chart, owns tasks tied to goals, and works within a budget. Add a review stage and its work goes to a reviewer before the task can close. The four areas below show who uses which part.

One task, start to finish

How Paperclip manages agent work

One owner per task, one record per decision. A goal becomes tasks. Each task has a single owner, an agent or a person, who does the work in its own runtime and attaches the evidence. If the task has a review stage, marking it done sends it to the reviewer instead, and requested changes go back to the same owner. The thread keeps the comments, approvals and cost.

  • Not the agent. Paperclip doesn’t build or host the model. Each run’s transcript stays on the task, but Paperclip isn’t a dedicated observability platform for model and tool calls. Bring Claude Code, Codex, Gemini or your own HTTP agent.
  • Not needed for one agent. If you run a single local agent and check its output yourself, skip this layer. It pays off once several agents share work, budgets or reviewers.

Illustrative example: names and IDs are made up. The review routing follows Paperclip’s execution policy.

For everyone, every day

Agentic Task Manager

Declare intent. Agents work. You verify the output. The surface everyone touches every day: tasks with owners and review gates, proactive agent coworkers that keep work moving between your prompts, auditable routines, and verification from real evidence instead of rubber stamps.

  • Tasks, approvals & review gates
  • Proactive agent coworkers
  • Auditable routines & workflows
  • Verify from diffs, screenshots & tests
For managers

Org Chart for Agents

Roles, permissions & boundaries for humans and agents. One org chart holds humans and agents together: responsibilities, delegation, and specialization decide who does what, governance decides who may do what, and secrets stay scoped to exactly the agent that needs them.

  • Mixed human + agent org chart
  • Responsibilities, delegation, specialization
  • Governance: who can do what
  • Scoped secrets & company boundaries
For enablers

Agent Employee Training

Design, train & evaluate your AI employees. Agents improve the way employees do: skills encode your procedures once for the whole org, evals and saved test runs measure quality, and active learning loops with human-approved changes make every agent better this month than last.

  • Skill Studio & shared org-wide skills
  • Evals & saved test runs
  • Active learning loops & quality metrics
  • Performance reviews for agents
For IT & platform teams

Agentic OS

The infrastructure that makes the work run. The layer underneath it all: a cross-provider runtime for any model and any agent, sandboxed execution with integrations and MCP servers, enterprise controls for SSO, GRC, RBAC and cost — self-hosted, so the data path is yours.

  • Cross-provider runtime — any model, any agent
  • Sandboxing, integrations & MCP servers
  • SSO, GRC, RBAC & cost controls
  • Data privacy, internal traces & compounding data value

Every role, one deployment

Open source and self-hosted — install it once and every part of the company gets its piece.