Hi Chris, I recently had this idea and am building a prototype: a Q&A system backed by an LLM (large language model) and grounded by LilyPond documentation. The issue I see is that building such a chatbot requires using the LLM API provided by OpenAI, Google or Anthropic, which is very costly. There are open source models, but hosting and running inference on them still require expensive compute. Hence, I haven't found a way to make this approach affordable and accessible to more users.
As for personal use, Kieren's approach with agent skills sounds very promising to me. If you have an AI subscription, you should try it. Overall, in my experience, ChatGPT, Claude, Gemini are okay for triaging LilyPond errors. One needs to be very specific, and ask them to ground their solutions with real references to Lilypond extensive documentation, in order for them from making things up. Hope it helps. On Fri, May 8, 2026 at 11:31 PM Chris Crossen <[email protected]> wrote: > > > > On May 8, 2026, at 2:07 PM, Kevin Cole <[email protected]> wrote: > > > > As AI's go, I've found Perplexity.ai is very good at citing sources > > and providing links to them without being badgered into doing so. > > > Thanks, Kevin, I’ll check it out. > > Chris >
