promptconduit

blog

· PromptConduit · 7 min read

Cursor AI Spend by Repo and Model

Quick answer

Cursor’s docs explain models and pricing, but the product UI focuses on plan limits — not a ledger of which workspace spent which model today.

PromptConduit attaches git repo name to every captured event and rolls spend up by model in the app and extension:

curl -fsSL https://promptconduit.dev/install | bash
promptconduit install cursor

Open PromptConduit: Show AI Cost Breakdown in the editor for per-session detail, or sign in at app.promptconduit.dev for multi-day charts.

What Cursor shows today

Cursor Settings → Usage gives account-level consumption against your plan. That’s enough to know you’re “running hot” this week.

It doesn’t answer:

  • Did frontend/ or infra/ drive the spike?
  • Was it Composer, Auto, or a named API model?
  • How much came from cache reads vs fresh context?

Repo attribution from git context

When the PromptConduit hook runs inside a git repo, the CLI records repo_name (and branch when available) on each event envelope. In the web app:

  • Overview shows period spend with deltas
  • Cost analytics groups by model and day
  • Team plan adds member-level rollups

All of this is built from the same local capture path — the server aggregates dollars your CLI already computed.

Model breakdown in the editor

The extension prices each turn using a bundled rate table aligned with Cursor’s published API rates. The panel shows:

  • Input, output, cache read, cache write tokens per request
  • Running session total in the status bar
  • “What if” comparison against other models (rate-only, not capability)

100% local until you sync.

Claude Code on the same repo

Many developers run Cursor and Claude Code in the same repository. PromptConduit captures both tools into one event log, so repo rollups reflect total AI spend on that codebase — not just one vendor.

Bedrock and Vertex

If your Claude Code sessions run on AWS Bedrock or Google Vertex, Anthropic’s console won’t show per-session analytics. PromptConduit still prices from local transcripts because capture happens at the tool — independent of which cloud hosts the model.

See also: Track Claude Code token usage and cost.

Next steps

  1. Install the CLI — no account required
  2. Install the extension for live totals
  3. Optional: Pro trial for 90-day history and digests