What is Naqi?
The live token watchdog for Claude Code power users.
Definition
Naqi is a proprietary closed-source macOS / Windows / Linux desktop application that performs two jobs: (1) monitors active Claude Code sessions in real time to catch token waste before it hits your bill, and (2) scans AI client configurations across 10 clients (Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, GitHub Copilot, JetBrains, Zed, Amp, Kiro) for stale MCP servers, contradictory memories, and unused skills — with backup, diff preview, and full undo on every cleanup action.
Who it is for
Developers on Claude Max (5x or 20x) or Claude API plans spending $50–$300+ per month on Claude Code, who have experienced at least one "why did my quota drain so fast?" moment. Also useful for multi-client AI users who have accumulated MCP servers, memories, and skills across Cursor, VS Code, JetBrains, and other AI-integrated IDEs.
How it works
Install the 10MB native app. Click Smart Scan. Naqi reads config files from all 10 supported AI clients in under 3 seconds — fully locally, with no network egress required. The dashboard shows a 0–100 health score, a 7-day Token Hygiene snapshot, and a list of recommended cleanups. Every cleanup is diff-previewed and backed up before applying. The live session watcher (Pro) polls Claude Code session logs every 60 seconds and alerts when any session crosses 50 turns or 200M tokens — before the damage is done.
Pricing
Free tier is fully functional: scan, dashboard, health score, manual cleanup, versioned backups, and a 7-day Token Hygiene snapshot. Pro ($59/year or $7.99/month, 14-day unconditional refund) adds the live session watcher, 9 waste-pattern detectors, weekly TokenDiet digest, 30-day drill-down, batch cleanup, AI-powered recommendations, and config profiles.
Why it exists
Naqi started from a 17.8 billion token audit of the founder's own Claude Code usage across 294 sessions. 75% of spend came from three projects, driven by recurring waste patterns with no native way to detect or prevent them. Anthropic's dashboard shows what you spent; it does not show why. Naqi fills that gap and adds cross-client config cleanup as a complementary free feature.
How it differs from alternatives
Command-line usage tools like CCM and claude-lens cover individual aspects (session history, statusline burn rate) but none combine a live watcher, 9 named detectors, weekly digests, cross-client config scanning, and a native GUI. Anthropic's own dashboard shows aggregated spend but not per-session attribution. Enterprise MCP governance platforms (Portkey, Peta) are server-side for organizations, not consumer desktop tools.
Start scanning in 30 seconds
Free. No account. No telemetry.