The Architecture

How SCIXO remembers you

Every AI conversation builds a continuously evolving profile of your projects, your preferred working habits, and your completed tasks.

Three things happen
every night

While you sleep, a quiet process reads everything you did today, extracts what matters, and prepares a briefing so tomorrow's AI session starts from where you left off.

Layer 01

Conversations become structured records

Every chat is saved as an organized daily record. This includes which project you worked on, how long you spent, what you built, and what's still unfinished.

Runs once per day · per session
Layer 02

Key facts are extracted

An AI extracts essential details like technology choices and working patterns, storing them as permanent and searchable memories.

Persistent · conflict-resolved
Layer 03

A living profile is written

All those facts get synthesized into a rich document that captures a portrait of your project and your working style, ready to be handed to your AI the moment you start a new session.

Regenerated nightly · instant to read
Behind the scenes

What happens at
11:30 PM

Every night, a pipeline quietly runs through everything you did that day. It reads your conversations, identifies the moments that matter, and updates your memory so your next session starts seamlessly.

Runs automatically without needing any action from you
Takes under 5 minutes for a full day of activity
Everything stays local and nothing ever leaves your machine
1

Conversations collected

Everything you said today, across all your AI tools, is gathered into one structured record.

11:30
2

Key moments found

The system finds where you got stuck and where you had your breakthrough, identifying the highest-signal parts of any chat.

11:31
3

Facts categorized into five types

Technical details, decisions made, open problems, how quickly things were solved, and your working patterns.

11:32
4

Old knowledge updated

Anything outdated gets replaced rather than piled on. If you switched tools, the old belief is removed first.

11:33
5

Project profile rewritten

A fresh, up-to-date document is written for each active project, detailing its current state, open issues, and recent decisions.

11:34
6

Your habit profile updated

Patterns that appear 3+ times across separate days get promoted from "maybe" to confirmed. SCIXO learns your style over weeks.

11:35
Memory, categorized

Five things your AI
actually remembers

Not everything is worth keeping the same way. SCIXO sorts every piece of knowledge into five distinct categories, each serving a different purpose.

Technical facts about your project

The specific tools, libraries, and setup used. Your AI gives you accurate answers instead of generic ones.

Example memories
Using Express.js with SQLite for the backend API
Auth switched from sessions to JWT tokens last week
Connection pool capped at 10 concurrent connections

Architectural decisions you've made

When you switch approaches, the old belief is replaced. Memory stays current.

Example memories
Migrated from cookie-based to token-based auth (March 28)
Upgraded database driver to fix async transaction bug
Switched to stateless architecture to support mobile client

What's still broken or unfinished

Open issues that weren't resolved. Your AI walks into the next session already knowing your to-do list.

Still open
Frontend not yet updated to send Bearer auth header
Token refresh logic not implemented, resulting in sessions expiring in 1h
Mobile client hasn't been tested end-to-end yet

How you personally like to work

Your style, preferences, and common habits are confirmed across all projects rather than just one.

Confirmed about you
Always pastes full error messages, so responses include line numbers
Prefers functional programming style over class-based
Gets stuck on CSS layout, so explanations are more thorough there
Day 1: Initial observation 1 of 3

"You always paste the full error message into the chat"

Seen once. Stored but not yet trusted.
Day 4: Second observation 2 of 3

"You always paste the full error message into the chat"

Seen twice. Growing confidence. Still watching.
Day 9: Final confirmation Confirmed ✓

"You always paste the full error message into the chat"

Seen three separate times. Now a confirmed habit.
Trust through repetition

A pattern isn't real until
it's seen three times

SCIXO doesn't jump to conclusions about how you work. A habit has to appear across three separate days before it's considered confirmed.

This prevents wrong impressions from sticking. One session where you did something unusual doesn't permanently change how your AI treats you. The truth has to prove itself.

The Moment of Truth

When you hit Resume,
something different happens

Without memory, every new chat starts from zero. You spend the first few minutes re-explaining your project. With SCIXO, the AI begins the session fully briefed and ready to assist.

Without memory for every session
YOU

Hey, I'm working on a backend API in Node.js. I switched to JWT auth last week, away from sessions. I'm having trouble with the refresh logic…

AI

Sure! Could you share more about your setup? What framework are you using? What does your auth middleware look like?

YOU

It's Express. Using the jsonwebtoken package. Here's the middleware…

AI

"Tokens? What tokens? I'm not sure which ones you're referring to from our last session."

3+ messages just to re-explain context

With SCIXO memory
YOU

I'm having trouble with the JWT refresh logic.

AI already briefed

Got it. Based on where you left off, specifically that JWT is active and using jsonwebtoken, but refresh rotation isn't implemented yet, meaning sessions expire after 1h. The frontend AuthContext also still needs updating. Want to tackle that first, or go straight to refresh?

First message. Already in context.

What gets loaded when you hit Resume

Project profile

Current stack, recent decisions, architecture state

Live facts

Any new knowledge from today, before synthesis

Open issues

Unresolved problems from previous sessions

Your habits

Confirmed preferences, style, working patterns

Loaded once at the start of the session and never re-fetched mid-conversation for instant performance.

Living knowledge

When you change,
your memory changes too

SCIXO never just adds new information on top of old. When you switch from one technology to another, the outdated belief is explicitly removed first. Your AI always holds the current truth rather than a pile of history.

This is why the system gets more accurate over time instead of accumulating noise. Every conflicting fact triggers a deliberate decision: keep, update, or replace.

When What happened Action
Day 1
Started using session-based login Memory records: "Project uses sessions"
Learned
Day 8
Migrated to token-based login Old belief removed. New one replaces it.
Replaced
Day 1
Got stuck on a CSS layout problem Habit noted: "Struggles with CSS" (1/3)
Noted
Day 9
Third time the pattern appeared Promoted from pending to confirmed habit
Confirmed
Day 30
Solved a complex CSS layout solo "Struggles with CSS" → "Proficient with CSS"
Updated
Where memory lives

Three stores,
three kinds of knowledge

Not all knowledge belongs in the same place. SCIXO keeps them separate to ensure every query is precise, meaning that asking about a project won't accidentally pull in unrelated facts from another one.

Per project

Project memory

Everything about a specific thing you're building. Tech stack, open bugs, decisions made, what's working. Each project has its own private memory, ensuring that they don't bleed into each other.

Current tech stack
Architectural decisions
Open bugs and unfinished work
Across all projects

Personal habits

Facts about how you work that are true everywhere. Your preferences, strengths, patterns. These cross-apply, as every project profile also pulls in what SCIXO knows about you personally.

Confirmed coding preferences
Recurring struggle areas
Communication and style habits
Per AI model

AI performance

How well different AI tools have worked for you. Which resolved things quickly, which needed more back-and-forth, which worked best for what kind of task.

Resolution speed by task type
How many attempts problems took
Which AI handled which task best
Gets better with time

Memory that grows
smarter every day

On day one, SCIXO knows a little about you. By week four, it knows your habits, your strengths, the things that trip you up, and the exact state of every project. Every session builds on the last.