How do I find which model each call used in Cursor (beta)?

Rius records the model name on the span for each answer in a Cursor (beta) chat, so you find it by opening the chat in the trace list or by filtering the Spans tab to LLM spans. Cursor (beta) chats appear under an agent named cursor, one trace per chat. The model name is captured in both structure-only and content modes, so you need no extra setup to see it. Cursor (beta) reports no token counts, so those spans show the model but no tokens and no cost.

How it works

The Rius plugin turns each Cursor (beta) agent chat into one trace. Each answer in that trace is an LLM span, and the span carries its model. Model names are sent whether a folder is enabled for structure only or with content.

The trace tree looks like this:

• An AGENT span for the Cursor (beta) session.
• A CHAIN span for each turn, meaning one prompt and everything it caused.
• An LLM span for the answer, with its model.
• TOOL spans for tool calls, and AGENT spans for subagents under the Task call that started them.

Spans arrive while you work: when the chat starts, after each answer, and every 20 tool calls. A chat’s trace closes when Cursor ends the session. If Cursor never says so, your next chat closes it once it has been idle for 12 hours.

Span times are when Cursor reported each step, not when the model call started. Tab completions and Cursor cloud and background agents are not traced.

Set it up

1. Install the plugin in Cursor. In Cursor’s plugin settings, choose Import from Repo and enter glassflow/rius-coding-agents. Cursor support needs plugin 0.6.0 or later. The Cursor (beta) guide covers install, sign-in and folder rules in full.
2. Sign in and turn tracing on for a folder, then check it:

/rius-login
/rius-enable-here
/rius-status

3. Work as usual, then open the Traces page in the console. Click the cursor cell in a row to filter the list to Cursor (beta) chats. The Sent by column shows whose key sent each chat, and from plugin 0.6.2 the User column shows the email you signed in with.
4. Open a chat and select an LLM span in the waterfall. The Model section shows the provider, the requested and response model. Use the previous and next buttons, or the ]l and [l keys, to jump between LLM spans.
5. To list calls across all chats, open the Spans tab. In the All kinds menu choose LLM, then filter to the cursor agent. Sort by time and read the model on each span.
6. To see only calls from one model, open an LLM span, go to the Raw tab, and on the model attribute choose Filter traces by this. You can also click Add filter, pick the model field from the attribute list, choose is or is any of, and pick a value. Attribute filters match exact values only. Use Exclude this value to remove one model from the view.
7. Click Save current search… on the Traces filter bar to keep that filter and come back to it. The exploring traces page covers filters, saved searches and the keyboard keys.

What you get

• The model behind every answer in a Cursor (beta) chat, on the span for that answer.
• A filterable list of Cursor (beta) LLM calls by model, agent, status and time range, with URLs you can copy so a teammate opens the same filtered view.
• The full span tree around each call: tool names, error types and subagents. A shell command that exits non-zero is marked as an error.
• Prompts, answers, and tool input and output on the Input / Output tab, only for folders you enable with /rius-enable-content-here. Recognised secrets are replaced before anything is sent.
• A question you can ask in plain language. The plugin registers the Rius MCP server as rius, so after you click Connect next to it in Cursor’s MCP settings, you can ask about your recent Cursor (beta) chats.

Cursor gives no token counts, so cost and token columns stay empty for these chats. A blank cost means no usage was reported, not that the call was free. The Claude Code usage guide shows the per-developer token and cost views that Claude Code sessions get.

When something else fits

If you need tokens or cost per model, Cursor cannot supply them, because its hooks do not report usage. Claude Code sessions traced by the same plugin carry model calls with tokens and cost. For the other coding agents Rius traces, see Coding agents.

Related