CAMARA-CHILENA-INTELIGENCIA-ARTIFICIAL/GRAIL ? reverse-engineered prompt

Build me a Python library that turns a folder of documents into a searchable knowledge graph, and also works as a long term memory layer for coding agents like Claude Code or Codex.

I want one engine with two modes, one for indexing documents into a graph and one where an agent can write memories, entities, and relationships directly. It should support querying the graph in a few different ways, including normal document search, broader graph search, and a recall style search that can filter memories by things like date, category, tag, entity, and confidence without needing an LLM call.

Please include a simple CLI for creating a project, indexing content, updating or deleting items, and exporting results. Add a small Python SDK too, plus a ready to use agent skill or tool format so an assistant can save and search memories easily. Use local storage by default, but make it easy to swap in vector backends and model providers from config. Look up current docs online if you need to.