Cortex launches quality-first AI memory infrastructure

Cortex targets the hardest part of agent memory: keeping persistent knowledge useful instead of letting it become an opaque vector-database dump.

• –Its reported 80% redundant-write rejection rate suggests the product prioritizes memory quality at ingestion, though the figure is vendor-reported.
• –Contradiction tracking and source citations could make long-running research assistants and enterprise copilots substantially more auditable.
• –MCP support makes the memory layer model-agnostic, while the quick Claude setup lowers experimentation friction.
• –The hosted architecture still leaves important questions around exportability, privacy, latency, and performance against Mem0, Zep, or custom stacks.