Free shows you the waste. Pro watches it live.

Live Token Hygiene

Naqi watches every Claude Code session in real time, detecting 9 waste patterns — mega-sessions, dive-deep openers, log pastes, subagent swarms, auto-compact waste, skill auto-load overhead, and more. The live watcher polls every 60 seconds and fires a notification the moment any session crosses the threshold.

The 7-day snapshot is free and shows your top 3 waste patterns with a $ estimate. Pro unlocks the full live watcher, all 9 detectors, 30-day historical drill-downs, and the weekly TokenDiet digest.

• 9 waste-pattern detectors running continuously
• SessionThresholdAlert — 60-second polling, fires at 50 turns or 200M tokens
• Weekly TokenDiet digest delivered every Monday
• 30-day historical drill-down with per-session breakdown
• ProjectHotspot nudge — flags when one project consumes 60%+ of weekly tokens
• Prompt waste detection — flags expensive patterns before you send

AI-Powered Recommendations

Naqi analyzes your entire workspace — server configs, memory entries, skills — and returns ranked recommendations with specific actions. Not generic advice: it knows which server is broken, which memory contradicts another, and which config key is wrong.

Powered by your own API key. Your data never touches Naqi servers — it goes directly from your machine to your chosen AI provider.

• Ranked by severity: Critical, Warning, Info
• Each recommendation includes a specific action
• Anonymized before sending — no paths, no secrets
• Works offline with Ollama (fully local, no API cost)

Works with any of these providers

Anthropic

OpenAI

Google Gemini

xAI Grok

Ollama

OpenRouter

Anthropic

OpenAI

Google Gemini

xAI Grok

Ollama

OpenRouter

Contradiction Detection in Memories

AI clients accumulate memory entries that contradict each other over time. "Prefer tabs" conflicts with "always use spaces". "Use GPT-4" conflicts with "default to Claude". Naqi Pro detects these contradictions, near-duplicates, and redundant entries automatically.

Case-insensitive matching, stop-word filtering, negation detection. Each pair is labeled: Contradictory, Near-duplicate, or Redundant.

• Detects across all memory files for a client
• Labels each pair with a specific conflict type
• One-click resolution — keep, merge, or delete
• Re-runs on every scan so new contradictions surface immediately

One-Click Batch Cleanup with Diff Preview

Free tier shows you what is wrong. Pro lets you fix it. Select all recommendations or just the ones you want, preview every change as a diff before it applies, then confirm. Naqi applies changes surgically — it never rewrites your entire config file.

Diff-based editing: only the targeted key is modified. Formatting, comments, and ordering are preserved.

• Diff preview for every change before applying
• Batch apply across multiple clients at once
• Automatic backup created before every change
• Full undo stack — reverse any action instantly

Config Profiles

Capture your current MCP server setup as a named profile. Apply it across clients with one click. Export as JSON to share with your team or restore on a new machine. Stop rebuilding the same setup from scratch.

Profiles are client-agnostic — a profile captured from Cursor can be applied to Claude Desktop or VS Code.

• Capture current server setup as a named snapshot
• Apply across any supported AI client
• Export and import as JSON
• Version-controlled — see what changed between profiles

Visual Config Editor

Click any config file to open an inline JSON editor with line numbers, real-time validation, and syntax highlighting. Format with one click. Save with Cmd+S. No more opening config files in a text editor to fix a trailing comma.

A backup is created automatically before every save. Safe Mode blocks saves if you have it enabled — all writes are gated.

• Inline editor — no context switching to a text editor
• Real-time JSON validation as you type
• Format button applies standard JSON formatting
• Backup created before every save

6 AI Providers — Bring Your Own Key

Choose the AI provider that fits your workflow. Cheap and fast for quick scans. Powerful for deep analysis. Fully offline with Ollama — no API cost, no data leaving your machine.

Your API key is stored in the macOS Keychain, never in a config file or Naqi servers.

• Anthropic Claude — best reasoning for config analysis
• OpenAI GPT — fast, reliable, widely available
• Google Gemini — large context window
• xAI Grok — fast inference
• Ollama — fully local, zero cost, zero data egress
• OpenRouter — access any model through one key

FAQ