Godek87/atlas-core ? reverse-engineered prompt

Build me a Python library that turns a pile of content into a searchable map of meaning.

I want to be able to feed it documents or rows from a database, have it chunk long text, create embeddings, group related items into topics, generate short summaries and a plain language overview, then search by meaning instead of exact words. It should work offline by default, but also let me plug in different embedding and LLM providers later, plus store data in memory or Postgres with pgvector.

Please make it handle multilingual content well, especially Arabic and Turkish, and support separate workspaces or groups in the same index. It should also do incremental sync so reruns only process changed content. If helpful, use the README examples as the main guide and add a couple of simple demos and tests so I can see it working end to end. Look up current docs online if you need to.