Fathom vs SuperMemory

A dedicated learner model extracts, merges, infers and forgets facts, storing each learning in a vector-graph database that understands entities, sources and time.

Learnings are injected into the model's context in real time via hooks, so agents recall relevant information without stuffing full histories into the prompt.

Memory lives outside any single model provider, so it carries across models and harnesses and is delivered through the API, plugins and MCP.

Mandatory memory is only billed once: repeated or unchanged content is not charged again, and the same rates apply on every plan.