Persistent semantic memory for AI assistants. Same memory from Claude, ChatGPT or any MCP client: your AI remembers your work, decisions and conversations across sessions, devices and models.

Unlike plain RAG or note archives, every write passes a server-side quality gate that rejects redundant memories (~80% of proposals in production), contradictions are tracked as first-class objects instead of being silently swallowed, and every answer can cite its sources.

Zero setup — hosted remote MCP server, OAuth sign-in, trial space auto-created on first connection (30 days, no credit card)
Governed writes — novelty filter, typed claims, deduplication at write time
Conflict tracking — contradictions surfaced with warnings, never hidden
Multi-model — one memory across Claude, ChatGPT and any MCP-compatible client
EU infrastructure — export and delete your memories at any time

Listed on the official MCP registry (io.github.FilippoPilo/cortex). Patent application filed (May 2026). Benchmarks published — including the inconvenient ones — at skynetlab-cortex.com, docs and examples on GitHub.

search	Cerca informazioni nella memoria dell'utente. Restituisce risultati con id, titolo e url da citare.
fetch	Recupera il contenuto completo di un elemento dato il suo id (ottenuto da search).
cortex_recall	Riassume cio che la memoria contiene su un argomento.
cortex_write	Salva un'informazione nella memoria dell'utente.
cortex_forget	Dimentica (elimina) una memoria dato il suo id.
cortex_conflicts	Mostra le contraddizioni rilevate tra le memorie.

Cortex — Persistent Semantic Memory

What it can do

What data it sees

Do you need an account