Usage and logs

Every model call through the LLM Gateway is one request with one record: who made it, what started it, which client, model and provider, tokens in and out, cost, and how long it took. Two pages under Platform > AI governance read those records: Usage for spend, Logs for the requests themselves.

​

Usage

Usage shows what the organization spent in the range you pick (Last 24 hours, Last 7 days, Last 30 days, This month or Custom range).

• The top row shows Total cost, with the change against the previous period, and one tile per payer: Subscription: xpander credits for calls on keys included with xpander, Subscription: your provider keys for calls on your own keys, and Subscription: personal Claude and ChatGPT plans for calls on people’s own plans. When your organization uses only one kind of provider credential, the provider keys tile reads Subscription: API key or Subscription: IAM assume role. Requests and Tokens in / out follow.
• Group by splits the chart and the table by Person, Group, API key, Model, Client, Provider, Account, Agent or License model. Hourly and Daily set the chart’s step for ranges up to 7 days.
• + Filter narrows everything to the same dimensions. In the table, a row’s menu offers a filter for that row, such as Filter by this person, and View requests, which opens Logs with the same filters.
• Export > Download cost by … (CSV) downloads the table.

Costs are estimated from token counts and xpander’s model prices, including cache reads and writes. For calls on your own key, your provider’s invoice is exact.

​

Logs

Logs lists requests newest first. Requests that belong together, such as the calls of one agent run or one coding session, are grouped into a trace with totals for the whole trace. Live shows new requests as they arrive, and Pause stops the updates. The range menu and the refresh button sit beside it.

​

Filters

+ Filter narrows the list. Every filter you set is kept in the page address, so a filtered view can be shared as a link, and Clear filters removes them all. Type a request or trace ID in the search box (Request or trace ID) and press Enter to open it. A request opens at https://platform.xpander.ai/governance/logs/<request or trace ID>, so you can share it as a link.

​

Export

Export downloads what the filters show: Download CSV, Download JSON, or Copy as table to paste into a spreadsheet. Each row carries the time, request and trace IDs, status, person, group, agent, operation, conversation, key, model, client, license model, tokens, cost, whether it was billed, latency and the error. Exports never include prompt or response text.

​

A trace or a request

Opening a trace shows its steps, with each turn’s User message and Result. A trace that is still running shows Live and refreshes on its own; the pause button stops the refresh and the play button starts it again. Opening a request shows: Hovering a cost shows who paid it, such as Subscription: xpander credits, Subscription: API key with the key’s name, or the person’s Claude plan.

​

Prompt and response text

Usage, cost and timing are always kept. Prompt and response text is kept for everyone unless an admin turns off Saved prompt text for it. A request without saved text says so on its row. Anyone whose role may view usage can open saved text, and every opened request is recorded in the Audit log. Exports and log destinations never include the text.

​

Requests xpander makes for you

Some features call a model on your organization’s behalf: conversation titles and summaries, building an API connector from its spec, generating an agent’s details, checking custom function code, image generation and transcription in built-in skills, and bot replies. These requests appear in Logs like any other, so you can see every model call made for your organization.

• The Person column shows the person who started the work, such as an admin creating an API connector, or - when no one did.
• The Source column names the operation’s step, such as Tags step. Hover it for the operation, such as Connector generator, what it does and who started it.
• The cost reads Not billed when the request ran on xpander’s own key; hover it for the cost xpander measured. These requests use no credits and count toward no budget.
• When the request ran on your own provider key, which is always the case on a self-hosted install, your provider bills it, so the row shows its cost.
• A few operations are charged in credits, such as scoring an agent’s answers in evaluations. Their rows show the cost like any other request.

Choose System calls above the table to list only these requests, and the xpander operation filter to narrow them to one operation. Calls returns to everyone’s own requests.

​

How long requests are kept

Keep requests for in Policies sets the retention (Retention). A request older than that opens as No request or trace with this ID. Requests are stored in your organization’s own database: on xpander cloud a database dedicated to your organization, on a self-hosted install the PostgreSQL database of your install.

​

Hybrid installs

On a Hybrid install, requests from agents that run on your install are stored inside your environment. When your browser cannot reach your environment directly, Logs shows Needs a direct connection to your environment. Usage totals for billing always show, and the calls people’s coding tools make through xpander cloud show in Logs as usual.

​

Your own spend

Each person sees their own spend for the day, week or month, with their spending limit when one applies, in Usage in Xpander Chat (Settings > Usage). At the limit it shows Limit reached and when calls resume.