AGENT WRAPPED

Your year with AI coding agents, as one card.

One command reads the logs Claude Code, Codex, Cursor and Hermes Agent already keep on your computer: tokens, prompts, streaks, your most-used model, a persona and one line of roast. Nothing is uploaded.

uvx --from git+https://github.com/DoAideAi/DoAide-Wrapped agent-wrapped

Needs Python 3.9+ and uv. Or clone the repository and run python3 -m agent_wrapped.

AGENT WRAPPED · 2026 My year in AI agents Claude Code · Codex · Cursor 4.2B tokens through my agents 6,180 prompts sent 1,240 sessions 214 active days 41 day streak 612h with my agents $3,480 API-equivalent cost MOST-USED MODEL Claude Opus 5 · 58% of replies THE NIGHT OWL 38% of prompts after 10pm You interrupted your agent 412 times. It never once interrupted you. agent-wrapped by DoAide · wrapped.doaide.com Made from local logs. Nothing left this machine.
Example card with made-up numbers. Yours is made on your machine.

Do this today: keep your Claude Code history

Claude Code deletes session transcripts after 30 days unless you change cleanupPeriodDays. Raise it now and December's card covers the year. Already-deleted sessions can't come back.

{ "cleanupPeriodDays": 3650 }

Add it to ~/.claude/settings.json. Claude Code's settings reference.

Private by design

Reads the logs where they are and never writes to them. Message text is only checked to tell your prompts from tool output. No network requests. The card shows totals, never project names, prompts or paths.

Four agents, one year

Claude Code, Codex (Desktop and CLI), Cursor and Hermes Agent, merged into one set of numbers. Missing an agent? Ask for it.

Honest numbers

Replies that agents log over several lines count once, background-task notifications and scripted runs don't count as prompts, and costs are what the agents themselves recorded.