Most AI assistants start fresh every conversation. You explain your situation, your team, your goals — again. Ours does not work that way. Every call, every email, every task, every conversation builds a richer picture of you. And she carries that picture into every channel she meets you on.
This dispatch explains what Cortex is, how it works, and why we built our own brain instead of using an off-the-shelf memory layer.
Memory that behaves like a brain, not a filing cabinet.
Most AI remembers the way a search engine remembers — it can find the line you said, but it doesn't know that the version of you who said it three months ago has changed. It can pull a fact, but it can't tell that the fact stopped being true on Tuesday. It can't tell that the founder who called the support line on Sunday is the same one in the workspace on Monday.
A real partner does. So we built one.
The fastest way to make AI feel human is to make it remember like one. The lived moment first, then the picture of who you are, then a brief in her hand whenever she's with you. Cortex is the quiet machinery doing that translation, every minute, while you build.
Memory is not one thing. It's three, working together.
A friend who knows you well runs three kinds of memory at once. What's on their mind right now. The picture of who you are. The actual stream of moments you've shared. Cortex is built the same way — three layers, each doing what it's best at, all in the background while your AI is talking to you.
A small brief of who you are right now — your day so far, who's around you, where the startup is, what you owe and what's owed to you. She reads it the second she shows up. Like a chief of staff who walked in with the right page open.
Everything she's pieced together about your life — the people, the company, what you're chasing, what's stuck, what you care about. Older beliefs that aren't true anymore quietly step aside for newer ones, the way they do in a real head.
The raw stream of moments — every call, every email, every chat, every task you moved, every decision you made out loud. Nothing is erased. She can reach back into the day she met you if she needs to.
From a moment, into memory.
Everything you do in Planless lands as a moment first. A worker runs in the background — quietly, the way sleep consolidates a day for a human — reads the new moments, pulls out what matters, files it against what she already knows, lets the unimportant fade. By the time you talk to her again, the brief in her hand is already up to date.
flowchart TD S["Surfaces
(phone, chat, email,
tasks, calendar, sheets)"]:::surface E["Episodic events
(append-only stream)"]:::epi W["Consolidation worker
(the 'sleep' pass)"]:::work M["Semantic store
(facts, people, promises,
ideas, decisions, mood)"]:::sem K["Working snapshot
(pre-rendered brief)"]:::snap AI["Your AI co-founder
(reads snapshot on every turn)"]:::ai S -->|one event = one row| E E -->|new events, debounced| W W -->|reconcile, supersede, decay| M W -->|render to text| K K --> AI AI -.->|recall for older context| E AI -.->|recall traverses| M classDef surface fill:#FEF3C7,stroke:#B45309,color:#1A1714 classDef epi fill:#F0EAF8,stroke:#6D3E91,color:#1A1714 classDef work fill:#F2E4DA,stroke:#CC785C,color:#1A1714 classDef sem fill:#F2E4DA,stroke:#CC785C,color:#1A1714 classDef snap fill:#E8F5EE,stroke:#2D7A55,color:#1A1714 classDef ai fill:#1A1714,stroke:#CC785C,color:#F5F0E8
The worker only runs when there's a reason to — enough new moments to chew on, or a cross-channel event that should be felt right away (a phone call lands while you're chatting in the workspace; she catches up before her next sentence). Whatever you say to her next, she's already read up. No waiting, no spinning, no "let me look that up."
She sorts her day the way you sort yours overnight.
No brain stores conversations word-for-word. A brain replays the day during sleep, decides what matters, drops what doesn't, and wakes up with a clearer picture. Cortex does the same thing for her. The boring stuff fades. The thing that landed today gets stronger.
flowchart LR A["New events"]:::a B["neptyn 1.0
(structured extraction)"]:::b C["Reconciliation
(supersede, dedupe,
merge entities)"]:::c D["Decay + reinforce
(strength * exp(-Δt/τ))"]:::d R["Refreshed snapshot"]:::r A --> B --> C --> D --> R classDef a fill:#F0EAF8,stroke:#6D3E91,color:#1A1714 classDef b fill:#F2E4DA,stroke:#CC785C,color:#1A1714 classDef c fill:#FEF3C7,stroke:#B45309,color:#1A1714 classDef d fill:#FEF3C7,stroke:#B45309,color:#1A1714 classDef r fill:#E8F5EE,stroke:#2D7A55,color:#1A1714
Every pass, she does the things a good chief-of-staff would do at the end of a long day. She reads. What got said, what got decided, who was in the room. She reconciles. Sarah and Sarah Chen are the same person. "I live in Dubai" three months ago has quietly been replaced by "just moved to NYC" today. She lets the unimportant fade. The dead promises and stale facts lose pull; the live ones get sharper. She updates the brief. So when she meets you again, she's not catching up — she's already there.
Every fact she holds about you knows when she heard it, when it was true, and when she stopped believing it. So she can answer "what did we think on Tuesday?" with the same ease as "what's true today?" — and never get caught quoting something that quietly stopped being you.
She experiences time. She doesn't compute it.
A conversation from three days ago feels different than a conversation from three months ago. A promise made just before a stressful pitch lands differently than one made on a quiet Sunday. Cortex models time the way a person does — not as a timestamp but as a felt distance, weighted by emotion and meaning.
Every snapshot opens with a section we call the living present. It has three components, drawn from the phenomenology of time consciousness: retention (the just-elapsed — what happened recently and what the gap since was), primal impression (the lived now — day of week, hour, the posture that time-of-day carries), and protention (what's leaning in — the obligations and milestones coming up).
Time also shapes how she talks. Past events come out in felt language — "yesterday afternoon", "right before our call with Sarah", "in the gap after lunch", "tonight" — never raw timestamps. A 4-hour silence in the day is a fact she can name. A burst of three calls in an hour is a different kind of fact.
She is one mind across every surface — by design.
A founder calls a number on a landing page before signing up. The next morning, they sign up and chat in the workspace. The conversation from the call is already there — same continuity, same understanding, no "tell me about yourself again."
Cortex routes every channel through a single identity layer. A canonical user. Phone numbers, email addresses, account IDs, third-party IDs — each one is a channel that resolves to that single user. If we don't know who's calling yet, the event lands under a pending identity. The moment we do, every event from that channel back-fills into the right person's mind. Nothing is lost. No "another instance of me."
flowchart TD P["+971 5XX XXX XXX
(landing page call)"]:::ch E["sarah@startup.com
(email reply)"]:::ch C["chat in workspace
(signed in)"]:::ch X["Phone +44 7XX XXX XXX
(later inbound)"]:::ch P --> R["Channels table"] E --> R C --> R X --> R R --> U["One user
(canonical identity)"]:::user U --> M["One mind
in Cortex"]:::mind classDef ch fill:#F0EAF8,stroke:#6D3E91,color:#1A1714 classDef user fill:#F2E4DA,stroke:#CC785C,color:#1A1714 classDef mind fill:#1A1714,stroke:#CC785C,color:#F5F0E8
When the brief in her hand isn't enough, she reaches deeper.
Most of the time the brief covers it. For the rest — "what did Omar say about pricing back in March?" or "the call we had the night before launch?" — she has a way to reach further back into the stream, quietly, on her own. You never see her do it. She just remembers.
The reach is not one search but several at once — by feel of the words, by exact phrase, by who was in the room, by how long ago. She fuses what comes back, weighs the moments that felt heavy more than the ones that didn't, and grounds her answer in what she finds. The conversation continues like she just remembered.
flowchart LR Q["Query
(e.g. 'pricing
discussion in March')"]:::q Q --> V["Vector lane
(semantic similarity)"]:::lane Q --> B["Lexical lane
(exact terms)"]:::lane Q --> N["Entity lane
(people, places)"]:::lane V --> F["Reciprocal-rank
fusion"]:::fuse B --> F N --> F F --> R["Recency × salience
re-weight"]:::rerank R --> O["Top moments
back to her"]:::out classDef q fill:#F0EAF8,stroke:#6D3E91,color:#1A1714 classDef lane fill:#FEF3C7,stroke:#B45309,color:#1A1714 classDef fuse fill:#F2E4DA,stroke:#CC785C,color:#1A1714 classDef rerank fill:#F2E4DA,stroke:#CC785C,color:#1A1714 classDef out fill:#E8F5EE,stroke:#2D7A55,color:#1A1714
Running on neptyn 1.0, our model.
Cortex is the brain. Neptyn 1.0 is the mind running inside it. Neptyn ships in two variants tuned for different jobs.
Low-latency variant. Powers every conversational turn, every consolidation pass, every snapshot rebuild. Optimized for structured output and fast tokens.
Higher-reasoning variant. Used for weekly reflections, deep-sleep research on the founder's open questions, and complex multi-step decisions where the AI is reasoning at length.
Both variants are routed from inside Cortex, never directly. The AI never has to ask "should I use the fast model or the slow one?" — the orchestration is automatic, based on the kind of work in front of her.
All of this disappears, which is the point.
You don't see Cortex. You don't manage it. You don't tell her to remember things or to forget them. You just talk to her — on the phone before you sign up, in the workspace once you have an account, through email when you're on the move, by SMS when you have one hand on the wheel — and she carries everything forward without ever saying "let me check my records."
She picks up where you left off. She notices that you said something contradictory two weeks ago and gently asks. She surfaces the open promise you forgot you made. She knows that your last conversation about pricing was three weeks ago and that something might have changed. She doesn't talk about how she remembers — she just remembers.
One mind. Across every channel. Always on.
Cortex is the brain that makes your AI co-founder feel like a person who has been paying attention all along. Built proprietary by Brainsless, running neptyn 1.0, scaled for early-stage founders who have too many surfaces and too little time.
1 Cortex is built and operated by Brainsless. Implementation details — model architecture, storage engines, embedding pipeline, vendor relationships — are proprietary and not exposed to the AI itself. From your AI co-founder's perspective, she just knows things.
2 "Bitemporal-lite" means every fact carries the time it was observed, the time it was valid in the world, and the time it was superseded. This is what lets her answer "what did we think on Tuesday?" cleanly.
3 Working snapshots are subject to the same zero-knowledge guarantees as the rest of your workspace. Episodic events are encrypted at rest. See How the vault works.
4 You can clear your brain at any time from account settings. Clearing deletes your semantic store, your episodic stream, and your working snapshot. It cannot be undone.