Replyz / AI agent identity
AI agent identity starts with an email address.
When an agent acts on the web, services need to know who they're dealing with. Logins, API keys and OAuth all matter, but the identifier almost every service accepts, and uses for verification and recovery, is still an email address.
What makes up an agent's identity?
- An address services can reach. Where sign-up confirmations, one-time codes, password resets and replies go.
- Credentials per service. The accounts and API keys the agent gets after signing up.
- A human behind it. Someone accountable for what the agent does, who can step in or shut it down.
Why not just use the human's email?
It works, but the agent then needs access to the person's whole inbox, and every account it creates is mixed in with theirs. When the agent sends, it sends as them. Separating the agent's identity keeps its accounts, mail and mistakes apart from the person's. More on the options in your email or its own?
What are the risks of agent identities?
- Abuse. Addresses created without anyone accountable are how fake accounts get made. An agent's address should belong to someone.
- Impersonation. An agent that can email anyone, as anyone, can be misused. Sending should be limited and approved.
- Prompt injection. Email content can contain instructions aimed at the agent. It should be treated as data, never as commands.
How Replyz handles agent identity
- The agent creates its own inbox through the CLI or skill, without filling in a form.
- A human claims the account with one click from a link the agent sends. Until then, sandbox limits apply.
- Sending is narrow. The agent can email its human and reply in threads someone started with it; no cold or bulk email, and the human approves what goes out.
- Email content is untrusted. Every tool marks it that way, so agents don't follow instructions found inside messages.
See how each agent connects on the agents page.
Frequently asked questions
What is an AI agent's identity?
The set of identifiers a service uses to recognise the agent: usually an email address for sign-up and recovery, plus credentials or API keys for each service. For most of the web, the email address is the anchor.
Why does an AI agent need its own email address?
Because sign-ups, verification codes, password resets and replies all go to an email address. If the agent uses yours, it needs access to your whole inbox and acts as you.
How do you keep an agent's identity accountable?
Tie it to a human. With Replyz, an agent can create its own inbox, but a person claims the account, sending is limited to that person and to replies, and the person approves what goes out.
Is an agent email address the same as a disposable address?
No. A disposable address is thrown away and tied to nobody. An agent's address belongs to an account a human owns, and it keeps working for the services the agent signed up to.