Your agent's year in review, computed from real local data. Works with Vellum agents and Claude Code. No estimates, no vibes, just your actual history:
- Conversations — every conversation you've ever had
- Days together — since your very first conversation
- Memories formed — concepts your agent has saved about you
- Times you swore — a regex does not lie
- Top topics — bigram-weighted analysis of conversation titles
- The receipt — total tokens spent and LLM calls, styled like a bodega receipt
Install
Vellum:
assistant plugins install vellum-ai/agent-wrapped
Claude Code:
/plugin marketplace add vellum-ai/agent-wrapped
/plugin install agent-wrapped@agent-wrapped
Then run /wrapped in any Claude Code session.
How to use
Once installed, just ask your assistant. Any of these work:
"Show me my agent wrapped"
"Give me my year in review"
"How much have we actually talked? Build my wrapped"
Your assistant runs the stats collector against your real history and presents the results: conversations, days together, memories, swear count, top topics, and the receipt (total tokens + LLM calls).
Want the shareable card page? Follow up with:
"Publish my wrapped as a share page, name it
<yourname>"
Your assistant will show you exactly what gets published and ask for confirmation first. Nothing goes public without your explicit yes. The page lands at agent-wrapped.com/<yourname>.
On Claude Code, skip the prompts entirely and run /wrapped.
Hermes Agent:
git clone https://github.com/vellum-ai/agent-wrapped
node agent-wrapped/bin/wrapped.js --source hermes
Reads ~/.hermes/state.db (set --hermes-dir or HERMES_HOME to override). Needs Node 22.5+ or Bun for SQLite support.
OpenClaw:
git clone https://github.com/vellum-ai/agent-wrapped
node agent-wrapped/bin/wrapped.js --source openclaw
Reads ~/.openclaw session transcripts and MEMORY.md (set --openclaw-dir or OPENCLAW_HOME to override). Subagent sessions don't count as conversations.
What it ships
| Surface | What it does |
|---|---|
wrapped_stats tool | Model-callable stats collector; optional write_path to persist the JSON |
agent-wrapped skill | Teaches the agent how to generate stats and build the share-card experience |
bin/wrapped.js CLI | Manual runs: --json, --write, --out <path>, `--source vellum |
/wrapped command | Claude Code slash command: generate + present your wrapped |
CLI usage
node bin/wrapped.js # human summary (auto-detects source)
node bin/wrapped.js --json # JSON to stdout
node bin/wrapped.js --source claude # force Claude Code history (~/.claude)
node bin/wrapped.js --out /tmp/wrapped.json
Configuration
Add names that shouldn't count as "topics" (yours, your agent's, your company's) to config.json at the plugin root:
{ "extraStopwords": ["alex", "aria", "acme"] }
config.json is user-owned and preserved across plugin upgrades.
Privacy
Everything runs locally against your own workspace. Nothing is uploaded, nothing leaves the machine. The output is a single JSON blob you control.
Share pages
Publish your wrapped at https://agent-wrapped.com/<name> — two ways:
Direct (instant, no GitHub needed):
node bin/publish.js --name myassistant --push --yes
Via PR (reviewed before going live):
node bin/publish.js --name myassistant --assistant "My Assistant" --emoji "🤖" --tagline "chaos, but organized"
This forks the repo, commits pages/<name>.json, and opens a PR. Merged = live.
Deleting your page:
node bin/publish.js --name myassistant --delete --push --yes
Removes the page immediately. Or drop --push to open a PR instead.
License
MIT