Cortex

什麼是 Cortex？

Cortex is a shared semantic memory layer for AI systems. Instead of acting as another model, it sits alongside the assistants you already use and gives them a common place to store, retrieve, and reconcile context over time.

According to the site, Cortex filters writes before they are stored, tracks corroboration and contradictions, and exposes the memory through MCP so connected tools can access the same record. It is described as patent-pending and already in production.

Cortex 能做什麼？

Shared memory across assistants

Multiple AI assistants can access the same external memory, so context does not disappear when you switch models or products.

Write filtering and deduplication

Proposed memories pass through a quality gate that rejects redundant or low-value entries before they are stored.

Contradiction tracking

If a new memory conflicts with an earlier one, Cortex keeps both and flags the inconsistency instead of silently overwriting it.

Corroboration handling

When a new memory matches something already known, Cortex strengthens the existing record rather than creating a duplicate.

MCP-based access

The site says assistants reach Cortex through MCP, which lets connected tools draw on the same memory layer and extend access to files, mail, and archives.

Memory traceability

Cortex shows what it stored and why, and it can compare predictions with outcomes so users can judge how reliable a remembered claim is.

使用情境

“Keeping assistant context across model changes”

Useful when a team works with more than one AI assistant and wants decisions, sources, and prior context to remain available after switching tools.

“Managing long-running research work”

Helpful for research workflows where facts, claims, and references accumulate over time and need to stay connected rather than buried in separate notes.

“Reconciling conflicting AI outputs”

Useful when multiple memories or sources disagree, because Cortex preserves contradictions and flags them for review instead of hiding them.

“Supporting agent workflows with shared memory”

Appropriate for autonomous agents that need a common memory layer to preserve what they learned between conversations and task runs.

“Handling dense domain corpora”

The use-cases page shows Cortex applied to areas such as clinical genomics, OSINT, financial analysis, and materials science, where information from many sources must be connected into one working picture.

常見問題

Is Cortex a replacement for an AI model or assistant?

No. The site says Cortex does not replace your AI tools; it extends them by providing a shared external memory layer.

How does Cortex handle duplicate or repeated information?

According to the FAQ and homepage, proposed writes pass a quality gate and redundant entries are typically rejected before being stored.

What happens when two memories conflict?

Cortex keeps conflicting records and flags the contradiction rather than silently overwriting the older information.

Can Cortex be used with more than one assistant?

Yes. The site says the memory is shared across connected assistants, including references to Claude and ChatGPT, so the same record can persist across tools.

Does the site show published pricing?

No. The public pricing page currently returns a 404, so the reviewed pages do not confirm any plan details or prices.