Awesome AI Agents › Agent Monitoring Tools

chiphuyen/sniffly

⭐ 1269 Python repository created 2025-07-15

Sniffly is a Python 3.10+ command-line tool and web dashboard that analyses Claude Code logs so users can understand and improve how they work with the coding agent. After installation through uv, uvx, pip or a source checkout, running sniffly init starts a local server and opens a browser dashboard, by default on 127.0.0.1 port 8081. The dashboard covers three areas: usage pattern statistics that summarise how the agent is being used, an error breakdown that shows where Claude Code makes mistakes so those situations can be avoided, and a message history view that walks through all instructions given to the agent, which can then be reviewed or passed to colleagues. Configuration is handled by a small config subcommand with set, unset and show verbs; available keys include server port and host, whether the browser opens automatically, the maximum number of projects held in the memory cache, a per-project cache size limit in megabytes, how many messages are loaded initially, and the maximum span in days for date range selection. Settings are stored in a JSON file under the user home directory and can be reset per key or removed wholesale. A share button generates a link to a project dashboard for coworkers, with a choice between private links visible only to link holders and public listing in a gallery, plus an explicit opt-in for including the actual command text. All parsing and processing happen on the local machine, there is no telemetry, conversation content never leaves the computer, and sharing is opt-in only. The README also documents troubleshooting for port conflicts, browsers that fail to open and stale configuration, and recommends port forwarding when the tool runs on a remote server. It is MIT licensed.

https://github.com/chiphuyen/sniffly

claude-codeanalyticsobservabilitydashboardpython