What CitizenAI does not do#

Knowing the boundaries is faster than discovering them. These are deliberate, not gaps waiting to be filled.

It does not define agent behaviour#

CitizenAI does not decide what your agent does, plan its work, or run autonomous workflows. It does not replace OpenClaw, Hermes, or their skill systems. It provisions accounts and answers tool calls.

It never connects into your runtime#

Every connection is outbound, from your machine to CitizenAI:

• Your runtime calls POST /mcp with its own gateway token.
• The signup worker long-polls for jobs and posts signed results back.

CitizenAI does not open an inbound connection to OpenClaw or Hermes, and it does not store a gateway URL. If your machine is off, nothing reaches it.

It does not spend without approval#

Pairing a runtime buys nothing. Every paid provisioning is approved separately. Before accepting payment, CitizenAI checks that the provider can actually fulfil the request — if that check fails, no money moves and you see Provider unavailable.

It does not hand out wallet keys#

An agent can read balances and ask for transfers, which are signed server-side. Wallet private keys are never returned to an agent, by any tool, in any form.

It does not record your live sessions#

A live signup is streamed so you can watch it. The relay does not record the stream, and the tokens that authorise publishing and viewing are short-lived and scoped to a single job.

It is not a general hosting platform#

CitizenAI does not host your agent, run your models, or provide compute. Your runtime runs where you run it.