Claude Code leaves logs. Sniffly reads them.

What it does

Sniffly spins up a local web dashboard that chews through your Claude Code logs and visualizes how you actually use the AI assistant. It charts usage patterns, categorizes error types, and lets you replay your instruction history. All processing happens on your machine unless you explicitly opt to share a dashboard link.

The interesting bit

The privacy posture is unusually strict for an AI tool: no telemetry, no cloud processing, and conversations never leave your computer. The error-breakdown view specifically surfaces where Claude Code makes mistakes so you can refine your prompts or avoid repeating them.

Key highlights

• Runs fully offline with no telemetry; data never leaves your machine
• Error-type breakdown to spot where Claude Code fails
• Message history walkthrough with optional sharing via private or public links
• Simple CLI for configuration and server management
• Python 3.10+, MIT licensed

Caveats

• The README insists all data stays local, yet describes a “public gallery” for shared dashboards; the mechanics of that sharing are unclear.
• Remote access requires manual port forwarding; it is clearly built for local, single-user access.

Verdict

Grab it if you live in Claude Code and want usage analytics without surrendering your conversation history. Ignore it if you do not use Claude Code or have no interest in quantifying your AI assistant usage.