llamaindex-python-memory

LlamaIndex Python memory attribution

This example signs memory commands from a real Python llama_index.core.memory.Memory instance while memory storage stays host-owned.

Run it from the repo root:

pnpm --filter @atrib/integration llamaindex-python-memory-smoke

What the smoke proves:

• A published llama-index==0.14.22 package import can run in a transient uv environment.
• Memory.from_defaults(...) can use a StaticMemoryBlock.
• App code still calls put, put_messages, get, get_all, set, and reset.
• Public atrib records carry hashes, operation names, signatures, chain roots, and informed_by links.
• Raw memory text and retrieved memory context stay in local sidecars, not public records.

What it does not prove yet:

• VectorMemoryBlock retrieval or an external vector database. In this package version, SimpleVectorStore is not enough for that proof because it does not store nodes directly.
• A full LlamaIndex agent or workflow.
• LLM calls, hosted persistence, or external adoption.
• A current LlamaIndex review target.

The integration point is intentionally narrow: sign the memory commands at the host boundary, keep LlamaIndex storage and retrieval semantics intact, and keep private memory material out of public records.