Cortex

Every AI assistant forgets. You explain your project, your constraints, the decision you made and why, and next session it is gone. Switch model and you start over. Cortex is a semantic memory that lives outside the models and is shared by all of them. It stores decisions, facts, constraints and the reasons behind them, and gives them back to any assistant that asks, through MCP, an open standard. It is in the official ChatGPT connector directory: enable it there, nothing to configure. Claude, Claude Code, Cursor and any MCP client read and write the same memory. What makes it more than a place to put text is what it refuses. Cortex rejects around 80% of what it is offered: duplicates, noise, things that will not matter in a month. What it keeps, it classifies as accepted, corroborated, contradicted or rejected, and marks whether something was observed from a real source or inferred. When two memories disagree it says so instead of silently picking one, and a nightly consolidation merges what belongs together. Key features: - Write-time quality gate (about 80% of proposals rejected in production) - Facts extracted as claims, conflicts tracked as objects with an open or resolved state - Superseded memories stay linked to the ones that replace them - Every answer can cite the ids of the memories it comes from - Same memory from ChatGPT, Claude and any MCP client - EU infrastructure, one isolated tenant per user, export and deletion in one click Built and run by one person, in production, patent pending in Italy. Free trial with no time limit while paid plans are being prepared; plans from 0.99 EUR per month.