whatidid

"What did I actually build this week?"

Cross-agent AI work digest that reads sessions from every local AI tool you run — Hermes, OpenAI Codex CLI, Claude Code, GitHub Copilot CLI, Continue, and more — and translates them into a business-readable value report.

Inspired by the What-I-Did-Copilot project. Built to work across all agents, not just one.

What it does

• Reads session data from: Hermes Agent (SQLite), OpenAI Codex CLI, Claude Code, GitHub Copilot CLI, Continue
• Groups related work by project and workstream
• Estimates human-equivalent hours using effort heuristics
• Calculates value delivered (hours × blended hourly rate)
• Estimates AI credit cost and shows value multiple (e.g. 1300×)
• Outputs: Markdown report, JSON data, and a self-contained HTML report (Linear.app-inspired design)
• Redacts credentials, tokens, and sensitive paths automatically

Quick start

pip install whatidid2
whatidid --days 7
# → ~/whatidid-reports/AI_WORK_DIGEST_LATEST.html

CLI options

whatidid [OPTIONS]

  --days INT           Look-back window in days (default: 7)
  --max-sessions INT   Cap on sessions to process (default: 100)
  --hourly-rate FLOAT  Blended hourly rate for value calc (default: 125.0)
  --no-html            Skip HTML output
  --output-dir PATH    Where to write reports (default: ~/whatidid-reports)

Supported sources

Sample output

Sessions : 93
Projects : 9
Est. hours: 169.25
Value     : $21,156
AI cost   : $16.05
Multiple  : 1318×
Confidence: medium (0.72)

Reports saved to ~/whatidid-reports/.

License

MIT — Tyler Thompson