Sift
Cross-platform usage analytics for AI coding tools. Sifts through local session data, computes 27 metrics, and generates reports, an interactive dashboard, and a machine-readable JSON export.
Install as Claude Code Plugin
# Add the marketplace
/plugin marketplace add AndurilCode/sift
# Install the plugin
/plugin install sift@sift
Then use /sift in any conversation to analyze your AI usage.
Standalone Usage
# All time, all sources
python3 -m sift
# Last 7 days
python3 -m sift --days 7
# Since a specific date
python3 -m sift --since 2026-03-01
# Filter by source
python3 -m sift --source claude-code
python3 -m sift --source claude-code --source copilot-cli
# Filter by project (substring, case-insensitive)
python3 -m sift --project my-repo
# Combine filters
python3 -m sift --days 30 --source claude-code --project my-repo
# List available sources and projects
python3 -m sift --list
python3 -m sift --list --days 30
Supported Sources
| Source | Data Location |
|---|
| Claude Code | ~/.claude/projects/ |
| Copilot CLI | ~/.copilot/session-state/ |
| VS Code Copilot Chat | ~/Library/Application Support/Code/User/workspaceStorage/ |
| Cursor | ~/.cursor/chats/ + ~/.cursor/ai-tracking/ |
| Gemini CLI | ~/.gemini/tmp/ |
| Codex CLI | ~/.codex/state_*.sqlite |
Output
All artifacts are written to ~/.sift/:
| File | Description |
|---|
report.md | Full markdown report with all metric sections |
dashboard.html | Interactive HTML dashboard with filters and charts |
export.json | Machine-readable JSON with all metrics and per-session data |
prompts/ | User prompts grouped by project |
Metrics
| Category | Metrics |
|---|
| Cost | Total cost, cost/session, cost/action, cost/minute, daily burn, platform comparison |
| Cache & Context | Cache hit rate, amortization, input freshness, context accumulation, tool definition overhead |
| Output | Output ratio (net/gross), stop reason distribution, model mix |
| Productivity | Edit/read ratio, turns before first write, lines ratio, prompt length distribution |
| Health | Session health (median/P90/bloat), session outcome (success/failure), retry ratio, duration trend |
| Adoption | Project adoption, top sessions, model routing efficiency |
Development
Build the plugin (first run also configures git hooks for auto-rebuild on commit):
bash build_plugin.sh
After this, any commit that touches .py files will automatically rebuild analyzer.pyz and bump the plugin version.
Requirements
- Python 3.10+
pyyaml (for Copilot CLI parsing)
No other dependencies. The dashboard uses Chart.js via CDN.
AndurilCode
