BurnBar：本地优先的AI编码助手使用量追踪工具

BurnBar is an open-source macOS menu bar application developed by the Imagine That AI team. It addresses the 'black box' cost problem of AI coding assistants by using a local-first architecture to track usage, cost, and workflow of tools like Claude Code, Codex, and Kimi. Key features include real-time monitoring, privacy protection (data stored locally in SQLite, no API keys required), and optional cloud sync. Project repository: https://github.com/Imagine-That-Ai/BurnBar.

AI coding tools like Claude Code, GitHub Copilot, and Kimi boost productivity but hide unclear costs. Developers often use multiple tools without real-time spending visibility, leading to unexpected monthly bills. Provider statistics are delayed or lack detail, making it hard to track usage patterns and cost distribution.

BurnBar’s local-first architecture ensures: 1. All data stored in local SQLite databases (no account/API keys needed, no cloud dependency for core functions). 2. Reads local session logs of AI tools instead of API queries (API keys stay with providers, zero extra network requests). Supports Claude Code, Codex, Kimi, etc.

1. Menu bar integration (non-intrusive, no Dock icon). 2. Real-time tracking: time dimensions (today/week/month), units (dollars/tokens), provider-wise distribution. 3. InsightEngine: detects usage patterns (e.g., 40% cost increase vs yesterday), cache hit analysis, new model reminders. 4. Daily summary notifications. 5. Built-in chat panel for querying usage data.

1. Daemon-first design: core functions via OpenBurnBarDaemon (runs in background even if UI closed). 2. Local retrieval system: GRDB/SQLite as primary storage, Firestore for sharing only. 3. CLI tools: support health check, project control (command example: swift run --package-path OpenBurnBarDaemon OpenBurnBarCLI -- help). 4. Editor extensions (Cursor/VS Code) for seamless integration.

Optional features: Google/Apple login (Firebase auth), selective sync (usage records, chat metadata), explicit control over chat content sync, and easy opt-out without affecting local functions. Also supports routing models like Z.ai via local OpenAI-compatible gateway for Cursor/Factory/OpenCode.

BurnBar’s macOS Beta version (0.1.3-beta.1) can be installed via GitHub Releases (download DMG, drag to Applications) or source (make install). System requirements: macOS, already using supported AI tools (Claude Code, Codex, etc.).

Scenarios: Personal devs (cost awareness, budget planning), team management (usage pattern analysis, cost optimization), efficiency analysis (habit improvement). Limitations: macOS-only, Beta state (stability issues), VS Code extension not on market, only supports tools with local logs. Conclusion: BurnBar fills a gap in the AI tool ecosystem, offering privacy-focused usage tracking. It will become a standard tool as AI assistants grow in popularity.