What you settled, and should not have to settle again next time.
settled
Governed cognition system for AI agents
Your agents can change.
Their experience stays.
Cortex captures the work carried out with your agents, builds an experience from it, learns, remembers, and returns the relevant context when they need it.
Trace → Experience → Learning → Recall → Context returned
On your machine · local-first · no remote telemetry
Claude Code
Codex
Antigravity
Gemini
ChatGPT
Claude.ai
Remembering is not rereading.
When a conversation ends, not everything in it matters the same for what comes next. Finding what was said is not the same as remembering what matters today.
Cortex does not reread your old conversations. It keeps what matters for what comes next.
What you took back, refined or put straight along the way.
taken backWhat the work revealed, and will still hold in six months.
discoveredWhat cost you once, and is better not done twice.
to avoidWhat you tried then set aside — and does not need trying again.
set asideThe rest can be forgotten. What Cortex hands back to your agents is what still matters.
The memory cannot be hidden.
If a piece of information can influence your agents, you must be able to see it and correct it.
Where the work stands, without rereading it.
You come back to a project several days later. What was decided is there, what is blocking is stated, what needs your view is flagged. Nothing to reread.
What you keep apart stays apart.
What it keeps, what it learns, and where it comes from.
What you said does not get mixed with what Cortex inferred about the way you work. Every piece of information says where it comes from, and you can correct it, flag that it is no longer true, or remove it.
Information you remove stops influencing your agents.
A system that claims to remember without showing it is worthless.
Cortex tells you how many sessions it kept, what still contradicts itself, what is waiting for your view. Never an "all good" with nothing behind it.
If something is missing, Cortex says so and shows you what to do.
You keep the subscriptions you already have.
Your ChatGPT, your Claude Pro or Max, a coding plan. Cortex adds no subscription: it uses what you already pay for, and only says it is ready once it has checked.
Your credentials stay in your machine's vault.


You decide what it consumes.
Routine work, background work and hard calls do not need the same model. You choose which serves what, and what takes over if it fails.
If something stops answering, Cortex tells you instead of pretending.
You will not spend your days approving things.
Cortex works while you work. It comes to you when a decision genuinely deserves your attention, not for every piece of information.
What cannot be undone always requires a gesture from you.
The same memory, everywhere you work.
You switch agents. The experience you built up stays available.
Claude Code
context comes back inside the toolYou start in the morning, a decision is made. Nothing to do on your side: it is kept.
Codex
context comes back inside the toolIn the afternoon, you carry on elsewhere. The morning's decision is there, with nothing to explain again.
Your web conversations
sessions are keptChatGPT, Claude.ai, Gemini, Antigravity: what you do there joins your memory. Context does not come back inside those tools, though — we do not pretend otherwise.
A session only enters Cortex if you allow it.
You choose the conversations Cortex uses, and you can go back on that decision. Nothing enters memory without your consent.
On a given conversation, you allow Cortex to follow it. Until you do, nothing is kept.
The conversation comes across as it goes. Nothing to copy, nothing to file.
The session is offered to you. You are the one who attaches it to a project, or sets it aside.
The gesture is identical for Claude Code and Codex. Your memory stays under your control, and you also choose which AI services Cortex uses.
Picking work back up without starting over.
Without rereading
The decisions made, what is still blocking, what remains open. Without reopening the conversation.
Without explaining again
The project's context comes back to your agent when it needs it. You do not have to lay it out again.
Without repeating
Your ways of working, once confirmed, follow you from one project to the next. You can see them and correct them.
Cortex is a system, not an interlocutor.
- A replacement for Claude, Codex, ChatGPT or Gemini. You keep talking to them.
- One more agent or chat. The agent that works and acts belongs to Cortex.deck.
- A notebook you have to file yourself.
- A memory that learns quietly, without you being able to see or correct.
- A layer that keeps the experience of the work done with your agents, and hands it back.
- A place to see what is kept, and correct it when needed.
- Something that lives beneath your tools, on your machine.
- The base of Cortex.deck: the same memory, if you want to go further.
Need an agent that acts?
Cortex.deck adds an agent able to work and act with the same memory. You find there exactly what Cortex kept.
Moving from one to the other is not a migration.
Give a memory to the agents you already use.
Cortex works with the agents you already use. Install it, open a first project, and carry on working as usual.