token-analyser

You can analyze Claude Code session token costs by parsing local JSONL and SQLite telemetry logs. This process provides per-call and per-model breakdowns, estimates model pricing, and detects usage issues like context explosions across specific sessions or timeframes.

Tracking subagent and multi-model token costs involves deterministic parsing of local telemetry logs. It extracts per-call usage data and applies model-pricing lookups to generate a readable cost and efficiency summary for your agents.

Yes, you can review telemetry logs for the current session, today, the last 24 hours, the past week, or a specific session ID. The analysis covers input, output, cache reads, and context growth across the selected timeframe.

Detecting high token ratios and missing cache reuse requires adversarial efficiency analysis of your session logs. It identifies context explosions, subagent-heavy runs, and calls where a cheaper model would have been sufficient.

Yes, token analysis supports both Claude and Codex local logs. It deterministically parses JSONL and SQLite telemetry to provide model-by-model usage breakdowns and cost estimates across both environments.

High session costs often stem from context explosions, missing cache reuse, or subagent-heavy runs. Analyzing your telemetry logs isolates which specific agent, model, or call pattern drove the spending spike.