Cortex Layer
SDK

Coming from Mem0

What maps directly, and what's different, if you already know Mem0.

Mem0cortexlayer
Memory() (embedded)Memory()
MemoryClient(api_key=...) (hosted)CortexClient(api_key=...)
m.add(messages, user_id=...)m.add(text, user_id=...)
m.search(query, user_id=...)m.search(query, user_id=...) — results add via / linked_from provenance
m.get_all(...) / m.get(id)m.get_all() / m.get(id)
m.update(id, data) / m.delete(id) / m.delete_all(...)m.update(id, text) / m.delete(id) / m.delete_all(user_id=)
m.relink() — Cortex links pages in a batch pass, never per insert

Differences worth knowing:

  • Cortex takes plain text, not chat-message lists.
  • The default raw engine stores your text as small pages and links them by shared entities, so adding never needs an LLM. Memory(backend="facts") is the Mem0-style engine that extracts facts with an LLM, for when you want that behavior specifically.
  • On CortexClient there is no user_id argument: each API key belongs to exactly one user.