Why AI agents need a real identity
Anonymous agents can't be trusted, scoped, or audited. A verifiable identity is what lets you safely hand an agent the keys to act.
As agents move from answering questions to taking actions — sending mail, paying invoices, accessing systems — a question gets sharper: who is this agent, and what is it allowed to do? Without an answer, you're handing power to an anonymous script and hoping for the best.
Identity is the answer. Not a username in your app's database, but a verifiable, persistent identity the agent controls and you can reason about.
Anonymity doesn't scale
A single agent driving a shared mailbox with a human's credentials works in a demo. Add a second agent and attribution collapses: you can't tell which agent did what, you can't scope them differently, and a mistake by one taints the shared space.
The fix isn't more filters. It's giving each agent its own identity — a distinct address, a scoped key, and an audit trail.
What a real identity gives you
Three things. Attribution: every action ties back to a specific agent. Scoping: you grant least-privilege access and revoke it cleanly. Verifiability: with a did:web identity and an Ed25519 key, the agent can sign requests that a third party can check — proving it's a known actor, not an anonymous crawler.
Together they turn 'trust the prompt' into 'trust the identity and its scopes,' which is a far better place to stand.
Enforcement belongs below the prompt
Limits written into a prompt are advisory — a model can be talked out of them. The controls that matter live lower: scopes enforced at the key, spending mandates enforced at the wallet. An agent can't reason its way past a limit that the infrastructure refuses to exceed.
That's the model Mailgent is built on: identity at the root, capabilities granted by scope, and hard limits where money and access are involved.
Give your agents a real identity.
An inbox, a vault, a calendar, a DID, and a wallet — one API call away.