BeeCount: local-first bookkeeping where the AI assistant is a cloud model call
Local-first bookkeeping for iOS/Android/Web. Self-hosted cloud + iCloud/WebDAV/S3 sync. AI capture. MCP.
At a glance
- What is it?
- A Flutter expense tracker with five sync backends and a self-hosted server in its own repository, whose comparison table promises zero data collection while its conversational and voice features are described as running on a third party model.
- Who is it for?
- BeeCount fits a person who wants an expense tracker whose ledger they can read, export and host themselves, and who is comfortable running a Docker container if they want multi-device realtime sync. Read four things first.
- Can I use it commercially?
- Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
- Is it still maintained?
- Yes. The repository last received commits 1 day ago.
- What is it written in?
- Mainly Dart, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- TNT-Likely/BeeCount2,814 ★ · +485
| Repository | 2026-09-11 | 2026-11-15 |
|---|---|---|
| TNT-Likely/BeeCount | 2.3k | 2.8k |
485 Stars in the last 65 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
Four licensing files, one of them commercial
The repository root holds more licensing material than most projects: a licence file, an English version of it, a separate commercial licence document, a third party notices file, and a privacy document. The package metadata carries no recognised licence value.
The commercial file is not an accident of tooling. The README carries a business cooperation section that explicitly offers commercial licensing, deployment assistance, custom development and consulting, with a contact address for it. So this is a project with an open licence and a paid path, and the boundary between them lives in a document rather than in a sentence.
The third party notices file matters for a different reason: an app that vendors a local OCR runtime and calls a hosted model is distributing someone else's code and depending on someone else's service, and this is where those obligations are recorded.
If you are considering embedding this in a product rather than running the app, read the commercial document first and treat the metadata's empty value as an absence of information rather than as permission.
The comparison table and the AI features are different claims
The pitch table sets the app against conventional finance apps on five rows: data held by a third party and unauditable, privacy that can be analysed, data lost if the vendor folds, paywalls on advanced features, and advertising. The responses are fully open source and auditable, offline first with a self-hosted cloud the developer cannot reach, data sovereignty across five sync options, completely free including the AI, OCR and voice features, and zero ads, zero tracking, zero data collection.
The AI section then describes what those features do. The assistant handles bookkeeping by natural language dialogue and is based on the Zhipu GLM model. Voice capture is press and hold, with the same model interpreting a colloquial sentence such as spending fifty on vegetables today. OCR runs two engines, a local one and a cloud one, and recognises payment screenshots from Alipay, WeChat and UnionPay.
So the two statements are compatible but not equivalent. The local OCR engine is the path that keeps recognition on your device, and the ledger itself is local by default. The conversational assistant and the voice path are described as model calls to a hosted service. If that distinction matters to you, it is per feature rather than per app.
Five sync backends with a routing table for picking one
Sync is presented as a menu of five, each with a stated scenario rather than a technical difference.
The self-hosted cloud is the one for multi-device realtime collaboration and multi-user shared books, with a Docker deployment, a bundled web interface and shared ledgers. iCloud is for people on a single iOS device and is described as zero configuration and native. Supabase is for cross-platform users without a NAS, chosen for its free allowance and simple configuration. WebDAV is for NAS users, with Synology, UGREEN cloud and Nextcloud named. The S3 protocol option is for flexible storage, with Cloudflare R2, AWS S3 and MinIO named as targets.
The FAQ turns that into a decision, and the decision is mostly about what hardware you already own: iOS single device means iCloud, cross-platform with realtime means the self-hosted cloud, cross-platform without a NAS means Supabase or S3, and a NAS means WebDAV.
The default, it is worth noting, is none of them. The FAQ says the app stores locally out of the box, that every feature works that way, and that CSV export is available at any time.
Backups are encrypted zips you are meant to open without the app
The backup design is the most interesting engineering decision in the project.
Backups are encrypted with AES-256 and fanned out to several remote targets at once, naming R2, S3, WebDAV and B2. Then it says something most encrypted backups do not: if you lose the service, you can recover with standard unzip tools. The backup is an encrypted zip rather than a proprietary archive.
That is a deliberate choice about lock-in. A backup format only the application can read is a backup that disappears with the application, and the promise here is that a family member with an unzip tool can recover the data years later even if the project, the server and the author are all gone.
The data safety answer in the FAQ is consistent with it: use your own server, storage or bucket, and the recommendation is HTTPS transport for the WebDAV and S3 paths, with the self-hosted cloud defaulting to encrypted backups.
The server half is a second repository
Everything about multi-user, realtime and web lives somewhere else. The deployment tutorial, the backup system, the progressive web app and the operational detail are all in the separate cloud repository, and the app README defers to it rather than duplicating it.
What that repository is described as doing is specific. One Docker image is both server and web, so a browser pointed at the server address is the interface. Multiple users register on one server with their data isolated from each other. Changes made on one phone appear on another phone and in the browser within seconds, over a WebSocket push rather than polling.
The shared ledger feature has the most detail, and it is the one that matters for a household. An owner generates an invite code, family or team members join the same book, there are owner and editor roles, and every transaction records who entered it and who edited it, with per-member totals. The file states that this is equivalent on all three platforms, which is the claim to test first if you plan to share a book across phone and web.
One platform was dropped and moved out, and MCP arrived
The platform line has a deletion in it. HarmonyOS is struck through and marked as no longer updated, with a link to a separate repository holding it. So the third platform target is not dead so much as separated: the code left this repository and its maintenance status is now a property of that other project.
What remains is Android from version 5.0, iOS from 15.5, and the web interface that arrives with the self-hosted cloud rather than as a separate target.
The new item is MCP support, announced at the top of the file as new, with documentation on the project site and a pairing requirement: it works together with the self-hosted cloud, letting a large language model drive the ledger directly. That is the feature to look at first if you want the app to be scriptable, and it is also the feature that ties the mobile app, the server and an external model together, so it inherits the same privacy questions as the AI features.
A Flutter client with generated code and hand-drawn header skins
The client stack is named precisely: Flutter 3.27 or newer, Riverpod for state, Drift over SQLite for the local database, and the same four cloud backends for sync. Four commands get you from a clone to a running app, including a code generation step and separate dev and release flavours:
# install dependencies
flutter pub get
# code generation
dart run build_runner build --delete-conflicting-outputs
# run the application
flutter run --flavor dev
# build a release
flutter build apk --flavor prod --releaseThe tree explains the rest. There is a localisation configuration for generated translations, a demo directory holding a hundred record and ten thousand record fixtures produced by a Python generator script, a preview directory, a tool directory and a packages directory.
The most documented contribution is cosmetic. Header skins are either code skins drawn with a custom painter that follow the theme colour, or image skins as SVG that fill the header and can be tinted. The stated simplest route is to copy an example SVG into the assets directory, register it in one Dart file, add a localised name and regenerate translations, with the full specification in a readme inside that assets directory.
Editorial conclusion
BeeCount fits a person who wants an expense tracker whose ledger they can read, export and host themselves, and who is comfortable running a Docker container if they want multi-device realtime sync. Read four things first. The privacy claims and the AI features are two different stories: the ledger and the local OCR path stay on your device, while the conversational assistant and voice capture are described as running on a third party model, so decide per feature rather than per app. Licensing is not a single file: the repository carries an open licence, an English translation of it, a separate commercial licence and third party notices, and the project solicits business licensing, so read all four before you embed it. Sync is a real choice with five backends and no default beyond local storage. And the server half lives in another repository, so a deployment is two projects, not one.
Frequently asked questions
Can BeeCount be used without configuring any cloud service?
Yes. The app stores locally by default, every feature works that way, and you can export CSV backups at any time. The five sync options are choices rather than requirements.
Which BeeCount sync option should I choose?
The FAQ routes by what you own: iCloud for a single iOS device, the self-hosted BeeCount Cloud for cross-platform realtime collaboration, Supabase or S3 for cross-platform use without a NAS, and WebDAV if you have one. S3 covers Cloudflare R2, AWS S3 and MinIO.
Does BeeCount send my data to an AI provider?
Partly, and the file separates the cases. OCR has two engines, a local one and a cloud one, while the natural language assistant and the voice capture feature are described as based on the Zhipu GLM model. The ledger itself is local by default.
How are BeeCount backups encrypted and recovered?
Backups are AES-256 encrypted zip archives fanned out to several remote targets, including R2, S3, WebDAV and B2. The project states that if you lose the service you can recover with standard unzip tools, so the format does not require the app.
What licence is BeeCount released under?
The repository carries an open licence file, an English version of it, a separate commercial licence document and third party notices, while the package metadata records no recognised licence. The README also solicits commercial licensing enquiries, so read the commercial document before embedding the app.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/tnt-likely-beecount)Star history chart for your README
[](https://hysenlabs.com/projects/tnt-likely-beecount)Was this useful?