Semantic code search

[barredterra]

We currently have naive codebase search, as implemented in 0c090dd

https://github.com/alyf-de/ask_alyf/tree/codebase-indexing implements a local vector database (stored as JSON) through LlamaIndex with any-llm embeddings. This turned out to be not very performant: (re-)indexing takes a lot of time and search is pretty slow. It's also not very efficient, since every site reindexes roughly the same apps.

Idea for a more performant solution:

• Set up elasticsearch instance
• Regularly index popular apps using https://github.com/elastic/semantic-code-search-indexer
• Sites get an API key and can query the database

Alternative:

• Use https://cocoindex.io/examples/code_index
• Transform code chunks to text using a cheap model (e.g. gpt-5.4-nano) and use that for the embeddings
• https://pypi.org/project/Whoosh/
• https://deepwiki.com/
• https://github.com/safishamsi/graphify