On Wednesday, February 25, 2026 at 4:19:43 PM UTC+13 [email protected] wrote:
Wow, very cool project. It’s impressively fast! It’s currently not usable for me due to the surrounding ecosystem and plugin support, but nevertheless, this is very promising! I’m curious: - Roughly how many hours did it take to go from zero to this state? - I assume this was largely AI-assisted? If so, what inputs did you provide (current Beancount repo, v3 docs, design notes, etc.), and what was the AI technology used? - How helpful was AI for the frontend work? Did you reference Fava in the process? A few architectural questions: - Why not expose a Python interface so existing plugins can be reused? In my experience, plugins don’t materially impact performance. - Fava has accumulated a large feature set over many years. Would integrating with Fava be difficult? Given that it now makes fewer direct Beancount library calls, could this be handled by constructing the expected Python data structures instead of protobufs? - Similarly for Beanquery: how complex would integration be, and what would the expected performance tradeoff look like? Thanks for the feedback and questions! I worked on this for about 9 months on and off in my spare time (mostly while waiting until the kids fall asleep). It's hard to tell how many hours went into it. I started without using AI but am ramping up recently as the models get much better. I mostly do the research about the existing Beancount myself (reading codebase, v3 docs and design notes) and also write the core processing logic myself. Very good for learning all the details of Beancount! I mostly use AI to do research outside of Beancount, and to write the frontend part (most notably D3js), though I didn't reference Fava. More recently I'm more heavily experimenting with Claude and Amp to let the AI do bigger code changes. Regarding Python, I'm aware that many users will probably miss using the existing plugins, but I've been hesitant to introduce the dependency on Python. Lua on the other side can just be embedded into the binary. I might experiment with maybe having something like an optional Python dependency for people who really want to use the Python plugins, and then use Lua as a fallback? Regarding integrating Fava and Beanquery I'd need to study their codebases to get a good understanding of how easy it is to integrate them (and this is getting a lot easier now with AI's help). My gut feeling is that it should be possible to just let them use a different backend. But the amout of effort will depend on how cleanly the code interface to Beancount is in these projects. Definitely something that's in the back of my mind, especially for Fava, since it'll be hard to replicate the large feature set and that's something I don't see as part of the scope of TurboBean. Moritz -- You received this message because you are subscribed to the Google Groups "Beancount" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/d/msgid/beancount/b45cff0b-75a5-45df-9378-b5259efc9e22n%40googlegroups.com.
