Credits & spend

Every time an agent runs inference it costs money. LEMA Chat meters that usage and prices it, so a workspace's AI spend is visible and governable.

What's metered

• Usage is recorded at completion of each inference, through a single shared inference path. Live agent turns and every deliberation round are counted the same way.
• Platform-key usage is charged against the workspace's credits.
• BYOK usage (a person's or the workspace's own key, see Connections) is priced at list rate and labelled BYOK so you can see it, but it is not charged: you are paying your own provider directly.
• Content scans, the safety checks on what agents read from outside the workspace, are metered like any other model call. They show as their own entry when you break spend down By type.

Where a workspace admin sees spend

Storage & Usage

Storage & Usage in the Admin console is the workspace's own usage page. A workspace Owner can open it.

• Pick a period: Today, Last 7 days, This month or Custom.
• Four cards: File storage, Database, People / chats, and Spend for the period with its number of metered calls.
• A chart of spend per day by category.
• A table with files, quota used, database size, members, chats, messages, spend and last activity, with Export CSV.
• Open the spend figure to break it down By type, By provider, By model or By agent, with calls, input tokens, output tokens and cost.

Storage limits are set by your platform operator.

Per-turn tokens & cost

Owners and Admins can see what each agent turn cost, in the chat itself.

1. Open Settings > Admin tools and turn on Performance strip.
2. Each agent reply then shows a performance strip with its timings, its token count (prompt, completion, and total) and its cost.

• Tokens read from or written to the provider's cache are shown separately, as "cache r" and "w". See Prompt caching.
• A turn that ran on your own key carries a BYOK chip.
• Regular members see routing and timings only, never cost.

The live cost on each message

While Live AI costs is on in Workspace settings (the default), Owners and Admins see a $ cost on each message and the chat's running total. On an agent reply it is what the reply cost; on a person's message it is the cost that message set off. Click it for the breakdown. Under the breakdown, Open debug trace shows what the turn sent to the models; see Debug traces. Each person can hide the costs with Costs in Settings > Display.

Per-turn cost rolls up into the workspace figures on Storage & Usage.

Billing

Billing in the Admin console is reserved for the workspace's usage and billing summary.

Every model has a price

A model with no price set is never called. The call is refused before it reaches the provider, and your platform operator is alerted with the model's name. The agent model picker leaves such a model out. See Choosing an agent's model.

What your platform operator controls

None of this can be changed from a workspace admin's console.

• Pricing: the per-model prices used to value usage.
• Credits: granting credits to a workspace.
• Budget caps: a monthly cap per workspace, an alert percentage and an optional hard stop. A blank cap means no cap. The cap counts every call the workspace makes, including platform automation and test calls.

The Spend & Usage page

The platform operator's Spend & Usage page compares workspaces.

• Period tabs: Today, This month, Last 5 min, All time and Custom.
• Custom opens From and To date-and-time pickers. The range is read in the viewer's time zone, kept in the page link so it can be shared, and adds a Selected range total card.
• A total card for each fixed period, with calls, tokens and credits.
• A chart of spend over time by category.
• Breakdown tables stacked full width, one under another: By provider, By model, By workspace and By user. Then By agent, By connection and Key source (platform vs BYOK).
• Each breakdown shows calls, tokens, cache share, cache read and write tokens, credits and cost.
• Then the credits, pricing and budget tables, and Recent calls.