Paper Compute is an open intelligence platform for teams building and using AI agents. It helps organizations capture AI sessions, turn successful work into reusable skills, and manage how AI workloads run in their own cloud. Its shared, authenticated endpoint connects providers including Anthropic, OpenAI, and Fireworks, while model access controls let teams set approved model lists. Paper records structured session histories, including prompts, tool calls, model details, token use, cost, duration, and completion status, with verified user attribution and OpenTelemetry compatibility.

Teams can generate skills from real sessions, evaluate their impact against recorded prompts, and monitor usage, spending, agent time, and completion rates. Weekly digests summarize team activity and highlight repeated work that may benefit from a reusable skill. Paper Compute’s open-source tools include tapes for capturing, searching, inspecting, and replaying agent telemetry; cassette extensions for skills, exports, search, and memory; and stereOS, a hardened Linux operating system designed for AI agents. The company also offers a CLI, console, documentation, demos, and pricing information.