AI CODING COST & FILE-ACCESS RECEIPT
See what the session cost.
See what it touched.
Drop a native Codex CLI rollout or generic JSON / JSONL session export. RepoMeter separates uncached input, cached input, output, and tool-call counts, then reports file scope when the log exposes structured paths.
01 / INPUT
Session log
Where is my Codex CLI rollout?
Choose one recent rollout-*.jsonl file:
Linux / macOS / WSL~/.codex/sessions/YYYY/MM/DD/
Windows%USERPROFILE%\.codex\sessions\YYYY\MM\DD\
Use a personal, non-sensitive session. Do not use company logs or logs containing confidential data.
02 / RECEIPT
What you get
- Token usage breakdownUncached input, cached input, and output
- File-access scopeFull local paths redacted
- Runaway-growth signalsSelf-log reads and sharp input jumps
- Shareable reportPrint or save as PDF
This receipt only describes events present in your log. It is not a security guarantee.
LOCAL RECEIPT
Session summary
Findings
Models
| Model | Uncached input | Cached input | Output |
|---|
Logged file scope
redacted locally| Displayed path | Accesses | Tools | Flag |
|---|
Receipt boundary
EARLY VALIDATION
Does this solve a real problem for you?
This prototype is testing one narrow promise: a trustworthy, vendor-neutral receipt for AI coding sessions.
Privacy note: the page counts a fixed set of anonymous funnel events (page view, sample, file selected, parse success/failure, receipt, feedback). It never sends your log, filename, path, prompt, model name, or report contents.