// the find
TNT-Likely/BeeCount
Local-first bookkeeping for iOS/Android/Web · Self-hosted cloud + iCloud/WebDAV/S3 sync · AI capture · MCP | 本地优先的跨端记账 · 自建云 + iCloud/WebDAV/S3 同步 · AI 记账 · MCP
BeeCount is a Flutter-based personal finance tracker targeting privacy-conscious users who want local-first data storage with optional self-hosted sync via WebDAV, S3, iCloud, Supabase, or their own BeeCount Cloud server. It ships on iOS, Android, and Web, with AI-assisted transaction capture (OCR, voice, NLP) powered by Zhipu GLM-4. Primary audience is Chinese-speaking users, though English is supported.
- Five sync backends (BeeCount Cloud, iCloud, Supabase, WebDAV, S3) is genuinely useful differentiation—most self-hosted finance apps offer one or two, and the architecture separating sync engines into distinct files (sync_engine_pull.dart, sync_engine_realtime.dart, etc.) suggests it was actually thought through rather than bolted on.
- The AI capture pipeline has real implementation depth: dual OCR engines (on-device TFLite + cloud GLM), voice input, screenshot monitoring via Android accessibility service and iOS back-tap shortcuts—these are non-trivial platform integrations, not just API wrappers.
- Repository hygiene is solid: issue templates in both Chinese and English, flavored builds (dev/prod), a contributing guide, design tokens doc, and a release workflow. Someone can actually fork and build this without reading the author's mind.
- Transparency about operational costs (¥847/year itemized) and a donors list signals genuine indie-project accountability rather than vaporware.
- Business Source License (BSL) means this is source-available, not open source. Any org that wants to run it internally for their team hits a commercial license requirement. The README calls it 'open source' repeatedly, which is misleading and will cause friction when someone's legal team gets involved.
- AI features are hard-coupled to Zhipu GLM-4 (a Chinese-market LLM). Non-Chinese users will need API accounts they probably don't have, and there's no obvious provider abstraction in the README to swap in OpenAI or a local model—the ai_provider_factory.dart exists but the default is clearly GLM-centric.
- No automated tests visible in the directory tree—no test/ folder, no widget tests, no integration tests. For a finance app handling user ledger data and multi-backend sync conflict resolution, this is a real risk anyone adopting or contributing to it will feel immediately.
- The HarmonyOS port is already abandoned after what appears to be a short lifespan, and the web version requires running the full BeeCount-Cloud Docker stack (separate repo). This means the 'cross-platform' story has meaningful caveats that aren't obvious until you dig in.