An identity layerfor every model you'll ever talk to.
Cortext exists because the intelligence keeps getting better and the relationship keeps resetting. Every new conversation starts from zero. Every new model means explaining yourself all over again. We're building the piece that persists — so the context stays with you, not the platform.
Mission
Give everyone a single, portable, user-owned memory layer that any AI can read and reason over — without surrendering that memory to a platform in the process.
The current model is backwards. Your preferences, your history, your quirks, and the shape of how you think all live scattered across a dozen chat products, each of which owns a slice and none of which share. That's not durable, and it's not yours. Cortext flips the ownership: your identity is the API, models are the clients.
How it works, briefly
Three moving parts, in order:
Bring in conversation history from the AI tools you already use. Cortext extracts durable memories — preferences, facts, patterns — and throws away the filler.
Memories get grouped into segments (work, personal, research, whatever) with a confidence score that decays over time and reinforces on repeated contact.
When you start a new conversation, Cortext compiles a model-shaped context and hands it to whichever provider you're talking to. You don't notice it — the model just already knows you.
The story, short version
Cortext started in early 2026 as a very personal frustration. I'd spend the first ten minutes of every Claude or ChatGPT conversation re-explaining who I was, what I cared about, how I wanted the answer framed, and what I'd already tried. Then I'd open a new tab and do it again.
I built the first version as a single HTML file that pasted a pre-composed system prompt into whatever model I was using. That file turned into an extension, then a small web app, then this. The thesis hasn't changed since day one: your identity is worth building once, not a thousand times.
Cortext is a solo project today. That's on purpose. The scope is narrow, the opinions are strong, and the surface area that touches your data is small enough for one person to hold in their head.
What we hold to
Every memory Cortext stores on your behalf is exportable, editable, and deletable by you. No read-locks, no silent retention, no 'we'll get to it'.
Cortext doesn't infer latent psychographic profiles. It stores the things you've actually said and done. That's it. The model-you is always inspectable.
You can take your memory layer and leave at any time. The export is complete, standard, and will import cleanly into the next thing — even if that next thing isn't us.
If something in Cortext is too complicated for one engineer to reason about, it's too complicated to be trusted with your identity. We pick simplicity every time.
Your identity is the permanent fixture. OpenAI, Anthropic, Google — whoever — they're just the current visiting speaker. Cortext's job is to brief the speaker.
Flashy where it earns the attention. Encryption, backups, uptime, and the permissions model are boring. That's the point.
Want to see it running?
The fastest way to understand Cortext is to import one conversation and watch it build the first few memories.
Get started →