claude-warehouse

claude-warehouse

Your Claude Code sessions are gold. Stop throwing them away.

Every session with Claude Code produces knowledge — solutions, debugging paths, architecture decisions, tool patterns. When the session ends, it all evaporates. Git stores what changed. Claude Warehouse stores how you got there.

It syncs every session into a local DuckDB database. Search it, query it, visualize it. The longer you use it, the more valuable it gets.

Install

/plugin marketplace add sderosiaux/claude-plugins
/plugin install claude-warehouse@sderosiaux-claude-plugins

That's it. The dashboard launches automatically at every session start.

Prerequisites: uv (DuckDB and sentence-transformers install automatically via uv)

Background sync (recommended)

Set up a launchd daemon to sync sessions every 10 minutes in the background:

cat > ~/Library/LaunchAgents/com.claude.warehouse.plist << 'EOF'
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
    <key>Label</key>
    <string>com.claude.warehouse</string>
    <key>ProgramArguments</key>
    <array>
        <string>/bin/bash</string>
        <string>-c</string>
        <string>timeout 240 uv run --script "$(claude info plugins-dir)/cache/sderosiaux-claude-plugins/claude-warehouse/*/scripts/sync.py" --verbose</string>
    </array>
    <key>StartInterval</key>
    <integer>600</integer>
    <key>RunAtLoad</key>
    <true/>
    <key>LowPriorityIO</key>
    <true/>
    <key>ProcessType</key>
    <string>Background</string>
    <key>ExitTimeOut</key>
    <integer>300</integer>
    <key>KeepAlive</key>
    <dict>
        <key>SuccessfulExit</key>
        <false/>
    </dict>
    <key>StandardOutPath</key>
    <string>/tmp/claude-warehouse-sync.log</string>
    <key>StandardErrorPath</key>
    <string>/tmp/claude-warehouse-sync.log</string>
</dict>
</plist>
EOF
launchctl bootstrap gui/$(id -u) ~/Library/LaunchAgents/com.claude.warehouse.plist

You can also trigger a manual sync anytime:

uv run --script "$(claude info plugins-dir)/cache/sderosiaux-claude-plugins/claude-warehouse/*/scripts/sync.py" --verbose

What You Get

Dashboard — live analytics at a glance

A visual dashboard auto-launches with every Claude Code session at http://localhost:3141.

• Overview cards — sessions, messages, tokens, projects, cost (30d vs prev 30d)
• Daily activity bar chart (14d)
• Cost breakdown by project (top 10)
• Tool distribution donut chart
• Write/Read ratio trend (weekly)
• Session efficiency buckets (abandoned → long)
• First-prompt quality vs. cost
• Recent sessions table
• Wrapped card — streak, peak hours, dev type, marathon session

No setup. Refreshes every 60s. Idempotent — starting multiple Claude Code sessions won't crash it.

Recall — find what you solved before

You ask Claude naturally. It searches your history.

• "I hit a CORS error last week, how did I fix it?" → searches for CORS error
• "Which project had the Kafka consumer work?" → searches for kafka consumer
• "I built a data table component recently" → searches for react table

Semantic search (v0.4.0): When keyword search isn't enough, recall falls back to vector similarity — finding related concepts even when the exact words don't match. Powered by local embeddings (sentence-transformers, no API calls), stored in DuckDB with HNSW indexing.

# Keyword search (exact match)
/claude-warehouse:recall CORS error

# Semantic search (meaning-based, finds related concepts)
vsearch.py "how to debug a performance issue" --project myapp --days 30

Report — understand how you use AI

Zero-effort analytics on your AI-assisted development habits.

/claude-warehouse:report

Shows you: token costs per project, session efficiency trends, most-used tools, busiest projects, abandoned sessions, and actionable suggestions to improve your workflow.

Costs — know where your tokens go

/claude-warehouse:costs

Token usage mapped to actual dollar amounts. Per project, per week, per model. See which projects are expensive and why.

Wrapped — your AI coding stats, shareable

/claude-warehouse:wrapped

Your personal "Claude Code Wrapped": total sessions, tokens consumed, favorite tools, longest session, most active project, top prompts. Fun to share, useful to reflect on.

Query — raw SQL for anything

/claude-warehouse:query SELECT project_name, COUNT(*) sessions FROM sessions WHERE created_at >= current_date - INTERVAL '7 days' GROUP BY 1 ORDER BY 2 DESC

Full SQL access to the entire warehouse.

Architecture

SessionStart hook
  └── dashboard.py & → HTTP server on :3141

launchd (every 10min)
  ├── sync.py        → incremental ETL into DuckDB
  └── embed.py       → incremental vector embeddings (sentence-transformers)