Skip to main content
Every call is paid one of three ways, called license models, and the gateway picks one per call in a fixed order. The session-hour rate is 8 credits for each active hour of an agent turn whose model calls run on your own key or plan (What a run costs). Credits, and so this rate, apply on xpander cloud only (Cloud, Hybrid and Air-Gapped). Admins choose which ones people may use in Policies (License model).

​
1. A person’s own Claude plan

To turn it on for your organization, contact xpander.
On xpander cloud, when xpander has turned it on for your organization, anyone with a Claude plan can bring it. The plan, including its extra usage, then pays for two kinds of calls:
  • Their agents in xpander. Claude Code agents that run for them, Omni included. See Agent runs that use a plan.
  • Claude Code on their own computer, signed in with their Claude plan and sending their xpander key in the x-xpander-api-key header.
In both cases the model must be a Claude model from Anthropic, and Bring your own subscription must be on for the person’s organization, group and agent. Otherwise the call is refused with a message naming what is missing. It is never silently moved to company models.
Anthropic’s Claude Code legal and compliance page says: “Advertised usage limits for Pro, Max, and Ultra plans assume ordinary, individual usage of Claude Code and the Agent SDK.” Don’t run a team bot or a shared, always-on agent on one person’s plan. Use a company key for it.

​
Agent runs that use a plan

The plan also pays for Claude Code agents that xpander runs, when the person the run acts for connected their Claude plan and picked it in the model picker for that agent. Another person’s plan pays only when it is pinned to the agent (Pin a plan to an agent). The plan covers the model usage of the run. Your organization pays the session-hour rate for it, on xpander cloud. A person connects their plan in the model picker of a conversation in xpander Chat, under Subscriptions. The plan cannot pay for:
  • An agent whose Claude Code runs on a Bedrock model, or on a provider other than Anthropic. An admin can move the agent to an Anthropic model.
  • An agent that is not a Claude Code agent.
When that plan is limited or its login expired:
  • In xpander Chat, the turn stops, and the conversation shows a notice with Continue on organization models. The turn waits until the person clicks it. When the admin’s When someone’s plan hits its limit or the person’s own When my Claude plan hits its limit is Pause, the notice has no such button, and the turn is refused until the plan resets or the person reconnects it.
  • Any other run (a channel, a schedule, a webhook or an API call) has no one to ask, so it runs on company models and the person gets a notification. When either setting is Pause, the run is refused with plan_unavailable instead, nothing is charged, and the person gets a notification.

​
Pin a plan to an agent

Someone who can edit an agent can pin their own connected Claude plan to it, in the agent’s settings under Advanced > Billing > Subscription plan. A pin makes that one plan pay for every person’s turns on the agent, not only the pinner’s, and the pinner answers for the plan’s terms. A turn that names no person never runs on a pin. Picking your plan in the model picker already covers your own turns. A pin adds the turns of everyone else on the agent. The Warning above applies to a pin. Pin a plan only to an agent that you alone use. Run a team agent, a bot or an always-on agent on company models instead: a key your organization stored, or xpander’s key (Company models). When the plan cannot serve sets what the agent’s turns do while the pinned plan is limited or signed out: Use the org’s models (the default), Wait until it resets or Stop the turn. When When someone’s plan hits its limit is Pause, the turns wait instead of using company models.

​
When the plan hits its limit

When someone’s plan hits its limit in Policies decides what happens until the plan resets:
  • Fall back to company models (default): the call continues as in step 2, and the person sees their plan is limited and when it resets. Each person can still choose Pause under When my Claude plan hits its limit in AI models & keys. These calls are paid like any company call: per token on your stored key, or in credits on xpander’s key, and they count toward budgets.
  • Pause: the call is refused until the reset, and nothing is charged. The choice is locked for everyone, with Your admin set calls to pause when a plan hits its limit.
AI models & keys shows a limited plan, for example Your Claude plan’s usage limit is reached until Thu, Oct 8, 8:00 AM. Calls use company models until then.

​
2. Company models

Every other call, and any call a limited plan falls back from, runs on a key your organization stored or on xpander’s key. Default license model in Policies decides which: With Bring your own key, the first stored key assigned to the agent, the person, their primary group or the organization wins (Whose key pays).

​
Cloud, Hybrid and Air-Gapped

On Hybrid and Air-Gapped, every provider your agents use needs a key in Platform > AI governance > Models or on the install (secrets.static). A call with neither is refused, naming the missing key. Agents of the same organization running in xpander cloud may still use Xpander AI Credits.

​
See who paid

A request in Logs shows its License model, and its Gateway decisions show which key served it and why. The Paid by column names the payer of each call or trace, and the Paid by filter narrows Logs to one license model or one of your stored provider keys. Usage splits spend into Xpander AI Credits, Your provider keys and Subscription: personal Claude plans. A ChatGPT plan cannot pay for calls. Only a Claude plan can.

​
Next steps

Choose license models

The License model setting.

Assign keys

Whose key pays.