Agent Lore

Why check costs early

AI tools charge per token. A single long session can consume more than you expect. Knowing where to look prevents surprises.

Claude Code: where costs live

Cost tracking depends on how you authenticate.

Personal account (Max plan)

Usage counts against your subscription. No per-token bill, but you have rate limits.

console.anthropic.com → Settings → Usage

API key

You pay per token. Track spending and set monthly caps:

console.anthropic.com → Billing → Usage

AWS Bedrock

Costs go through your AWS bill, not the Anthropic console. Check:

AWS Console → Cost Explorer → filter by "Amazon Bedrock"

Or use the AWS CLI:

aws ce get-cost-and-usage --time-period Start=2026-03-01,End=2026-03-09 \
  --granularity DAILY --filter '{"Dimensions":{"Key":"SERVICE","Values":["Amazon Bedrock"]}}' \
  --metrics BlendedCost --query 'ResultsByTime[].Total.BlendedCost'

Google Vertex AI

Costs go through GCP billing:

console.cloud.google.com → Billing → Reports → filter "Vertex AI"

In-session tracking

Claude Code logs token counts per response when you run in verbose mode:

claude --verbose

This shows input, output, and cache tokens after each turn. Useful for spotting expensive operations in real time.

Session cost report

Claude Code stores token usage in session files at ~/.claude/projects/. A Python script can read them and estimate costs across all billing methods.

Full scripts and the dashboard: releve.neorgon.com

curl -O https://releve.neorgon.com/scripts/releve-mini.py
python3 releve-mini.py --days 7

One number, no dependencies:

Releve  ·  7 days  ·  22 sessions  ·  13,621 turns
API-equivalent   $2,659.18
  excluded       113 turns, never billed
rates as of 2026-08-27. A counterfactual at list rates, not an invoice.

For the breakdown by model, project, skill and effort, releve-scan.py writes a JSON file you drop on the dashboard. Full walkthrough: /t/session-cost-report/.

Costs are estimates. Max plan users see theoretical API-rate costs. Check your actual billing dashboard for real charges.

Cursor costs

Subscription

Cursor Pro includes a monthly allowance of "premium" requests. Check remaining:

Settings → Subscription → Usage

Or visit: cursor.com/settings

Premium vs standard requests

Cursor splits requests into two tiers:

• Premium requests draw from your monthly allowance. They run on the strongest models (Claude Sonnet, GPT-4o) with priority.
• Standard requests are unlimited. They use smaller models or run during off-peak times with lower priority.

Once your premium allowance runs out, you can still work with standard requests at no extra charge. Cursor shows your remaining premium count in Settings.

Source: docs.cursor.com/account/plans-and-usage

Quick reference