跪拜 Guibai
← All articles
Juejin Community · AI Programming

Juejin Usage Reads Your Local AI Logs and Tells You Exactly How Dependent You Are

By JavaDog程序狗 ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

AI coding agents produce local logs that most developers never inspect. A tool that surfaces those logs as concrete Token and call-count trends turns vague dependency into measurable cost and workflow data, without the privacy risk of uploading conversation content.

Summary

Juejin Usage, released by the Xitu Juejin community, acts as a local-first AI accounting tool. It scans a developer's machine for installed coding agents like Claude Code and Codex, reads their local usage logs, and surfaces a dashboard of call counts, Token consumption, and model preferences over time. A desktop app auto-registers agent hooks on first launch; a CLI version runs via a single npx command.

The dashboard covers daily and weekly usage trends, model distribution across Claude and Codex, and multi-device merging after logging into a Juejin account. An opt-in leaderboard ranks users against the wider developer community by total usage. Privacy defaults keep all data in a local `~/.ai-usage/` directory — prompt text, hostnames, project names, and API keys are never collected, and cloud sync only uploads anonymized Token counts with a device identifier.

Dog Brother, a Qingdao-based developer, installed the tool and found his own numbers exceeded expectations, prompting a reassessment of which coding tasks genuinely benefit from AI versus manual work. The post ends with a call for readers to share their own stats in the comments.

Takeaways
Juejin Usage reads local Claude Code and Codex logs and displays daily/weekly call counts, Token consumption, and model distribution.
Desktop installation auto-registers agent hooks; a CLI version is available via `npx @juejin-opensource/jusage start` (Node.js ≥ 20).
Multi-device usage merges after logging into a Juejin account, and an opt-in leaderboard ranks users by total AI usage.
Privacy is local-first: data stays in `~/.ai-usage/` by default, and prompt text, hostnames, project names, and API keys are never collected.
Cloud sync, when enabled, uploads only anonymized Token counts and an anonymous device identifier; users can delete the local directory to wipe all records at any time.
Dog Brother's own dashboard revealed higher-than-expected AI dependency, prompting a review of which tasks should stay manual.
Conclusions

AI coding agents already produce detailed local usage logs; Juejin Usage is essentially a log-reader and dashboard, not a new instrumentation layer — the data was always there, just invisible.

The tool's privacy model (local-only by default, no prompt or project metadata collection) removes the main objection developers have to AI usage trackers, which is conversation surveillance.

A public leaderboard turns individual AI dependency into a social metric, which could normalize heavy AI use or, conversely, surface outliers who are over-reliant on specific models.

Concepts & terms
Juejin Usage
An open-source AI usage statistics tool by Xitu Juejin that reads local logs from Claude Code and Codex to display Token consumption, call counts, model distribution, and an opt-in community leaderboard.
Agent Hook
A mechanism Juejin Usage's desktop app uses to automatically register with installed coding agents like Claude Code or Codex on first launch, enabling it to read their local usage logs.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗