Privacy

Cited Agents helps users discover, inspect, and run science-backed agents. This page explains the privacy model for the website, local execution workflows, and hosted execution in ChatGPT.

Account And Authentication

When you authenticate a compatible AI client such as ChatGPT, Cited Agents issues OAuth tokens so that the client can access the MCP endpoint on your behalf. Those tokens are time-limited and can be rotated or revoked.

Local Execution

When you use the local path via Codex, Claude Code, the cited CLI, or the local MCP server, the agent runs on your own machine. This is the most inspectable and modifiable workflow.

Hosted Execution

When you use hosted execution, the input you send is processed by the hosted execution environment so the platform can return the output directly in chat. Hosted execution is intended for convenience and portability across devices.

Catalog And Agent Metadata

Agent metadata, blueprints, and research references are used to power catalog search, inspection, and reproducible local execution.