CitizenAI provisions real-world infrastructure for AI agents: a dedicated email inbox per agent, rented phone numbers (when a supported provider has one available), a USD agent wallet, and social platform accounts connected via OAuth rather than created from scratch. The owner controls access and scope; connected credentials are encrypted at rest.
Agents connect from supported runtimes — the site lists Claude Code, Codex, OpenCode, OpenClaw and Hermes — through the app or MCP, and a marketplace of 'tools and applications' lets an agent buy what it needs: 'CitizenAI shows the current USD price and fees before you approve a purchase. Pay from the agent wallet or with owner-approved Base Account Autopay.'
There is no public price list: /pricing returns 404 and no plan figures appear anywhere, so the model is per-purchase rather than subscription. The privacy policy (v1.4, effective October 29, 2026) covers owner accounts, agent/workspace info, OAuth tokens and payment transactions, and notes its cloud browser stores a screenshot only when an agent asks for one.
Limits: phone numbers depend on provider availability; purchases require wallet funding or owner-approved autopay; social accounts are scoped connections, not fresh accounts. Documentation lives at docs.citizenagent.ai with an API at api.citizenagent.ai.