Total token usage
Uncached input + output + cache write + cache read. Context sent again is counted again. This measures cumulative usage, rather than conversation size. Thinking tokens are already included in output.
Compare your Claude Code sessions and see where the tokens went. Input, output and cache, counted separately.
Run one command to find your Claude Code sessions and compare their token usage.
curl -fsSL 'https://byetokens.com/tools/claude-code-export-v1.mjs' | node --input-type=module - --site 'https://byetokens.com'Requires Node.js 18+ and curl. Runs once, with no package installation.
The report includes session and project names, dates, models and token counts. Prompts and tool contents are left out. No API key is needed, and the report is not uploaded to our server.
Choose the JSON report saved by the command, Claude Code JSONL files, or a project folder.
Processed in your browser. No upload, account or API key.
Manual session files are in ~/.claude/projects/ (Windows: %USERPROFILE%\.claude\projects\). Choose a project folder to include nested subagents. A new import replaces the current report.
Uncached input + output + cache write + cache read. Context sent again is counted again. This measures cumulative usage, rather than conversation size. Thinking tokens are already included in output.
Repeated responses are merged by message ID, or request ID when unavailable, keeping the largest count for each field. Nested subagent logs are grouped with their parent when folder paths are available. Only imported records are counted.
The command saves a JSON report on your computer. Automatic mode passes a compressed summary to the browser through the part of the link after #. This is not sent to our server and is cleared when read. Prompts and tool contents are excluded.
Cache reads, cache writes and output have different prices. These counts do not represent your bill or the percentage of your Pro / Max allowance. They also do not measure the quality of a completed task.