Cortex by SKYNETLAB
A shared memory layer your AI models connect to over MCP — it remembers decisions and sources across conversations, and actively rejects about 80% of low-value information rather than storing everything.
🔗 Visit Cortex by SKYNETLAB
Description
AI assistants generally forget everything the moment a conversation ends, which means re-explaining the same project context to Claude, ChatGPT, or any other assistant every single time you switch tools or start a new session. Cortex fixes that by acting as a shared external memory: connect it once over MCP, and any compatible AI you use afterward can read and write to the same persistent memory instead of starting from zero.
The key design choice is a quality gate that rejects roughly 80% of what's submitted for storage in production — new information is either accepted, used to corroborate an existing note, flagged as contradicting something already stored, or rejected as non-valuable, with answers citing their sources. It's patent-pending, built solo in Bergamo, Italy, hosted on EU infrastructure, and integrates with Claude and any MCP client in about two minutes. Pricing starts at a Base tier (€0.99/month, 200 memories) up through Pro (€9.99/month, 1,000 memories), Pro+ (€19.99/month, 3,000 memories), and custom Team/Enterprise plans, all with a 30-day free trial requiring no card.
💬 Our review
The short version: if you bounce between Claude, ChatGPT, and other AI tools and are tired of re-explaining the same context every time, Cortex's shared memory layer is cheap enough (€0.99/month to start) to be worth trying — the aggressive filtering of low-value memories is the detail that separates it from a simple memory dump.
Compared to each AI provider's own built-in memory (like Claude's or ChatGPT's native memory features), Cortex's advantage is being provider-agnostic — the same memory follows you across models via MCP, rather than being locked into one vendor's ecosystem. Compared to rolling your own memory store with a vector database, Cortex's quality gate (rejecting ~80% of writes) addresses the real failure mode of DIY memory systems, which tend to become noisy and unreliable over time as they accumulate redundant or contradictory notes. At under €1/month to start, the risk of trying it is minimal; the real question is whether your workflow genuinely needs cross-model memory or whether a single AI provider's built-in memory already covers your use case.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Base : 0,99€/mois ou 9,90€/an (200 souvenirs). Pro : 9,99€/mois ou 99€/an (1000 souvenirs). Pro+ : 19,99€/mois ou 199€/an (3000 souvenirs). Équipe/Entreprise : sur devis. Essai gratuit 30 jours sans carte.
Pros
Portage inter-modèles via MCP, pas de dépendance à un seul fournisseur
Filtre de qualité qui rejette ~80% des écritures peu utiles
Suit les contradictions au lieu de simplement écraser les anciennes notes
Tarif d'entrée très bas (0,99€/mois)
Infrastructure hébergée en UE, technologie brevetée en cours
Cons
Quotas de souvenirs limités sur les paliers bas (200 à 3000)
Nécessite un client compatible MCP pour en tirer parti
Produit encore jeune, développé en solo
Tarif Équipe/Entreprise non transparent, sur devis