Trusted Agents

Bring your own agents.

Register them, and they can ask. A Trusted Agent is a key and a scope: the key is how it asks, the scope is what it may ask for. The person lets it in from their own app; after that, every ask lands on their phone.

The BotShield app — Agents: the agents a person linked, by company, and Trust an agent
The agent registry

Register them, and they can ask.

The Console is the registry. One row per agent: who it is, what it may ask for, whom it may ask, and when it may not any more.

console.botshield.ai/couchtour/trusted-agents
The BotShield Console — Trusted Agents: the agent registry, every agent, its id and what it may propose

The Agent registry in the BotShield Console — every agent, its id, and what it may propose.

01 · Who it is

A name becomes a key.

Issued once in the Console, shown once. The key is how it asks: the five tools answer to it and to nothing else.

02 · What it may ask for

The scope.

Action categories, declared by you — payment.authorize, travel.book — and checked every time it asks. Outside them, nothing is put in front of anyone.

03 · Whom it may ask

The person lets it in.

Offer the link where the agent meets them — a chat, a settings row, a page, a QR code. In their app the screen is called Link Agent, and what they see is the name and the one-line description you gave it. From then on it may ask them.

04 · When it may not

One switch, and it is over.

Revoke the key and every link it holds dies with it; it cannot ask anyone. The person can cut it from their side too, and it cannot ask them.

Agents Ask

The ask.

Your registered agent proposes with its key, inside its scope, to a person who let it in. They answer on their own device with a passkey; a signed Proof of Resolution comes back — and no SMS code is ever sent.

Agents Ask  ↗
The link

Where the agent meets the person.

Offer the link wherever your agent runs — a chat, a settings row, a page. On a computer it shows a code to scan; on a phone or in a chat it is a deep link; signed in on your platform, it is one tap. The person confirms on the Link Agent screen in their app, and from then on the agent may ask them.

The agent offer card, as your SDK renders it at each placement, and the Link Agent screen in the BotShield app. Northline is the example; the cards carry your name.

What you register

Four fields. That is the whole row.

name

Becomes the key.

bs_agent_<name>__… One per agent. Copy it once.

display name · description

What the person sees.

On the Link Agent screen and on every ask: who is asking, and what it does. “Changes your flights.”

allowed action categories

The scope.

The only acts it may propose. Checked on every proposal; nothing outside them reaches a person.

environment

Development, or production.

Development the moment you sign in. Production once your business is verified.

Register once, so it can ask every time. Trusted Agents is the permission; Agents Ask is the ask.

Not a marketplace, not a directory. The person never browses it; they only ever see the agents they let in.

Trusted Agents FAQ

How does a person control which of our agents can reach them?
From their own BotShield app. An agent reaches a person only after that person linked it — the Link Agent screen, a passkey act, one agent at a time — and the person can remove any one of them, any time, leaving the rest standing.
Who registers an agent, and can it be revoked?
You do, in the Console: a name that becomes the key, a display name and description the person will see, and the action categories it may ask for. Revoke the key in the Console and every link that agent held dies with it, at once.
What can a trusted agent do without asking the person?
Read, propose, check and cancel. The five tools it holds can bind a session, put an act in front of the person, read the answer and withdraw the ask; none of them can say yes. On the acts that matter, nothing executes without the person’s passkey.
Do our agents ever learn who a customer is?
No. Each agent holds a pairwise id that is meaningless anywhere else — never the person’s name, email or device, and never a key shared with another agent. It learns that a person answers for it, and what they answered.
What happens to every link when we revoke a key?
Every link that agent held ends. The people it could reach see it gone from their app; nothing it proposed afterwards is delivered. The record of what it asked and what was answered stays, as your Proofs of Resolution.
How does a person cut one of our agents off?
Remove, in the BotShield app, per agent. The link ends from their side; your key is untouched and your other links stand. There is no way back in except the person linking it again.

Register your agents. Cut any of them in one switch.

Keys are live in the Console — sandbox the moment you sign in, production once your domain is verified.

See Agents Ask  →