Token-Tap is designed as a local meter, not a transcript archive.
By default it stores:
- timestamp
- agent name
- source
- model
- input/output/cached token counts
- estimated cost
- confidence level
- prompt and response hashes when available
- source file hash
- short redacted excerpt
By default it does not store:
- full prompts
- full responses
- raw log files
- source code contents
- private chat history
- secrets
- SMTP passwords
Short excerpts pass through a redactor that targets common secret shapes:
- Authorization bearer headers
password,secret,token, and API key assignments- OpenAI-style
sk-...keys - GitHub-style
ghp_...,gho_...,ghu_...,ghs_...,ghr_...tokens
Redaction is a defense-in-depth feature. Do not intentionally import logs containing secrets unless you understand the risk.
Hashes use SHA-256. They help detect repeated prompts and correlate events without storing the original content.
privacy.storeRawLogLines is present for future advanced scenarios, but the current implementation does not archive raw logs. Keep this off unless a later feature explicitly requires it.
Token-Tap does not:
- inject into VS Code
- intercept network traffic
- proxy agent calls
- upload usage to a remote service
- modify watched log files
SMTP sends alert emails only when explicitly configured.