# Agent seat playbook

What to do once you can see agents working through your customers' seats: the terms that set expectations, the notice your users see, an offer that gives agents a better route, and how to use the [agent-usage ledger](/docs/usage-ledger) when you talk to the customer. The agents work for your customers' people, so the goal is a better route and a fair price, never blocking the person whose seat it is.

Two things to keep straight in every message:

- Telltale detects agents. It does not detect several people sharing one login, and you should not claim it does.
- Charge or escalate only on **chargeable** agent units, the high-confidence ones: verified identity, an automation trace, injected input, sustained behaviour, a behaviour change, or two or more independent behaviour families agreeing with no physical-device input ([what counts as chargeable](/docs/usage-metering#what-counts-as-chargeable)). Medium-confidence (behaviour-only) agent units are a signal for a conversation, not a charge. Unknown is not human, and it is not an agent either.

## 1. Terms of service clause

> **Template, not legal advice.** Adapt it with your own counsel to your contract structure, jurisdiction and existing seat definitions.

> **Automated agents.** A "Seat" may be used by its named user directly or through software agents (including AI agents and browser automation) acting on that user's behalf ("Agents"). The Customer is responsible for the actions of Agents used with its Seats as if they were the actions of the user. Agent use is subject to these limits: (a) Agents must not be used to share one Seat among several people; (b) Agent activity in excess of [the fair-use allowance in the Order Form / X units per Seat per month] may be charged at the Agent rate in the Order Form, or moved to an Agent Seat or our API; (c) we may require Agents to use the API or an Agent Seat for [named high-volume actions]. We measure Agent activity using automated detection and will share the records behind any Agent charge on request. We will give at least [30] days' notice before charging for Agent activity that was previously included.

What the clause does: it makes agents acting for a user legitimate, sets a fair-use line, names the better route (API or agent seat), and commits you to showing your evidence, which the ledger lets you keep.

## 2. In-app notice

Shown in the page when an agent is detected (the SDK's `onAgent` handler, see [Agent monetization](/docs/agent-monetization)). Name the agent only when it is named, and always with how we know.

**When the agent is named (Verified or Unconfirmed):**

> **Claude in Chrome (unconfirmed) is working in this account.** That's fine. Agents get faster, more reliable results through our API or an Agent Seat. [See agent options]

**When it is not named:**

> **An automated tool seems to be working in this account.** If that's you, our API or an Agent Seat gives agents a faster, more reliable route. [See agent options]

Keep it short, dismissible and once per session. Never block the person, never call the tool an AI model unless it is named, and never show the notice on Unknown.

## 3. The agent seat offer

A simple offer that turns agent use into a route your customer chooses:

- **Agent Seat**: a seat for one agent working for one named person in the account. Priced per month like a user seat, with a monthly allowance of agent actions (for example 1,000) and a per-unit rate above it.
- **Included allowance on user seats**: each user seat includes a small allowance of agent units a month (for example 100), so occasional agent help never becomes a charge.
- **API access**: for high-volume, repeatable work. Point agents at your API documentation from the in-app notice.

Mark accounts that take the offer as **Agent seat** in Charge › Usage; their usage is then reported with that label. Use the [pricing estimate](/docs/usage-metering#pricing-estimate) (Charge › Pricing simulator) to see what past agent use would have been worth under each model before you announce prices.

## 4. Using the ledger in the customer conversation

1. **Get told when it starts.** Turn on account alerts (Charge › Ledger › Account alerts) with your sales or customer-success Slack channel and a monthly threshold. "Acme started using agents" is the moment to offer the better route, before any charge.
2. **Open with what you saw, in plain words.** "Since 3 October we've seen 1,240 completed exports in your account done by an agent, 1,100 of them by Claude in Chrome (unconfirmed)." Use the account page: chargeable and behaviour-only units per day, and who in the account.
3. **Offer the route, then the price.** Agent Seat or API first; mention the clause and the fair-use line second.
4. **If they dispute it, show the evidence.** Each chargeable unit has its session, the detector's basis, and how we know (for example a WebDriver flag, or a verified operator signature). Download the entries CSV for the period.
5. **When they are right, correct it.** Move the units to the right class (Change class on the entry). The ledger adds an adjustment on today's date and, if you send to Stripe, cancels or lists the credit. Nothing earlier is rewritten.
6. **Charge from closed days only.** Read the ledger for complete UTC days, or send meter events to Stripe; every entry carries an idempotency key, so nothing is charged twice.

What not to say: that Telltale caught someone sharing a login; that a tool is a specific AI model when it is not named; or that unknown sessions were people.