We leave fragments of ourselves everywhere.
A conversation with ChatGPT.
A half-finished project.
An email we never answered.
A note written at 3 AM.
A message on Telegram.
A saved idea.
A decision we made six months ago.
A project we started with enormous enthusiasm — and quietly abandoned.
Our digital lives are full of signals.
The problem is that those signals are fragmented.
We have data everywhere, but almost no system that can answer the question that actually matters:
«What does all of this mean for the decision I should make next?»
That is the problem I am exploring with PsycheForge.
The Engine That Turns Your Digital Life Into Decisions.
GitHub:
https://github.com/modarresi1913/PsycheForge
The Problem Isn't Lack of Information
We already have too much information.
We have:
- AI conversations
- emails
- messages
- notes
- projects
- documents
- bookmarks
- social posts
- ideas
- tasks
- decisions
- failures
- successes
Yet most AI systems operate with an extremely narrow window.
You ask a question.
The model answers.
Then the conversation ends.
Even when an AI remembers previous conversations, it usually remembers facts and context, not the deeper behavioral structure emerging across months or years.
That's the gap PsycheForge is trying to address.
From AI Assistant to Personal Cognitive Engine
The central idea behind PsycheForge is simple:
«Your digital life is a dataset.»
But it is not a normal dataset.
It contains traces of:
- what you care about
- what you repeatedly think about
- what you start
- what you finish
- what you avoid
- what you change
- what you decide
- what repeatedly goes wrong
- what consistently works
If we can connect these traces, we can move beyond simple retrieval.
We can begin modeling patterns.
And patterns can inform decisions.
The architecture therefore evolves from:
Input
↓
LLM
↓
Answer
toward:
Digital Life
↓
Personal Memory
↓
Cognitive Model
↓
Pattern Detection
↓
Decision Intelligence
↓
Intervention
↓
Action
↓
Outcome
↓
Learning
↺
That loop is the heart of PsycheForge.
What Does PsycheForge Actually See?
PsycheForge is designed to ingest fragmented information from multiple digital sources.
The current project includes parsers/connectors for sources such as:
- ChatGPT
- Claude
- Gmail
- Outlook
- Telegram
- Slack
- X/Twitter
- Apple Notes
- manual text
The important part isn't the number of integrations.
The important part is what happens after ingestion.
Different sources are normalized into a common representation so that the system can reason across them.
For example:
ChatGPT
│
├── "I want to build this product."
│
Gmail
│
├── "Still haven't launched it."
│
Notes
│
├── "New idea!"
│
Projects
│
├── 7 active projects
│
└── 3 abandoned
Individually, these fragments don't tell us much.
Together, they may reveal a pattern.
From Memories to Patterns
Imagine PsycheForge discovers:
11 projects started
8 became inactive
3 reached MVP
7 new projects proposed
4 projects expanded significantly after starting
The system shouldn't immediately say:
«"You are afraid of failure."»
That's an unjustified psychological conclusion.
Instead:
«Observed pattern: Projects frequently expand in scope before reaching MVP.»
Then:
«Possible interpretation: Scope expansion may be contributing to reduced completion rates.»
And finally:
«Suggested intervention: Define a fixed MVP boundary before adding new features.»
This distinction is extremely important.
PsycheForge should separate:
OBSERVED
↓
INFERRED
↓
HYPOTHESIZED
↓
RECOMMENDED
An AI should not confuse an inference with a fact.
The Cognitive Model
Over time, PsycheForge can construct an evolving model of the user.
Not a personality test.
Not a clinical diagnosis.
A Personal Cognitive Model.
Conceptually:
USER
│
┌───────────┼───────────┐
↓ ↓ ↓
Goals Interests Skills
│ │ │
└──────┬────┴────┬──────┘
↓ ↓
Projects Decisions
│ │
└────┬────┘
↓
Patterns
↓
Outcomes
Every important inference should ideally have:
- evidence
- confidence
- timestamp
- source
So instead of:
«"You are highly interested in AI infrastructure."»
PsycheForge could show:
«High recurring interest in AI infrastructure
Evidence: 37 conversations, 8 projects, 12 saved notes.
Confidence: 91%»
The user can then inspect the evidence.
The Most Important New Primitive: Decisions
Most personal AI products focus on information.
PsycheForge focuses on decisions.
A decision isn't just a chat message.
It has a lifecycle:
Question
↓
Context
↓
Options
↓
Evidence
↓
Choice
↓
Action
↓
Expected Outcome
↓
Actual Outcome
For example:
Decision:
Should I start another project?
Evidence:
- 4 active projects
- 2 unfinished projects
- 7 new ideas this month
- repeated scope expansion
Possible action:
Do not start another project yet.
Next step:
Choose one existing project and define
a 7-day MVP milestone.
Now the AI isn't simply generating content.
It is participating in a decision loop.
And Then Something More Interesting Happens
What happens after the decision?
Most AI systems forget.
PsycheForge shouldn't.
Suppose the user chooses the recommendation.
Seven days later:
«What happened?»
The user reports:
«"The MVP was completed."»
Now the system has something new:
Decision
↓
Action
↓
Outcome
↓
Learning
The next time a similar situation occurs, PsycheForge has more evidence.
This creates an adaptive loop:
Observe
↓
Understand
↓
Detect
↓
Decide
↓
Act
↓
Measure
↓
Learn
↺
That's fundamentally different from a static personal assistant.
The Intervention Engine
Another important component is the idea of contextual intervention.
Imagine the system notices:
«You have researched the same problem across 14 conversations.»
Instead of producing another 2,000-word explanation, it could say:
«You've already gathered enough information to make this decision.
Your next bottleneck is not knowledge.
It's commitment.
Write a one-page decision memo and choose.»
Or:
«You created five new project ideas this week while your current project has no recorded milestone.
Capture the ideas.
Don't start them yet.
Finish the current milestone.»
This is not about making the AI aggressive.
It's about making it useful at the right moment.
Why "Push Back" Matters
A personal AI that always agrees with you is not necessarily intelligent.
If the user's history contradicts their current narrative, the system should be able to surface that contradiction.
For example:
«"You said this project is your highest priority."»
Then:
«"However, your activity shows that you've spent 80% of the last two weeks working on three other projects."»
That's useful.
Not because the AI is judging the user.
Because it is showing the user a discrepancy they may not have noticed.
The goal isn't:
«Tell me what I want to hear.»
The goal is:
«Show me what I cannot easily see myself.»
Privacy Is Not Optional
This architecture inevitably touches extremely personal information.
That means privacy cannot be an afterthought.
A serious version of PsycheForge needs to treat the Personal Cognitive Model as user-owned infrastructure.
Users should be able to answer:
«What does PsycheForge know about me?»
«Why does it believe this?»
«Which source produced this inference?»
«Can I delete this memory?»
«Can I export my cognitive model?»
«Which data can be sent to external AI providers?»
The future direction is therefore strongly privacy-first and potentially local-first.
What PsycheForge Is Becoming
The original concept was closer to:
AI Thinking Partner
The new direction is much broader:
Personal Cognitive Engine
And the product category I am exploring is:
Personal Decision Intelligence
Not another chatbot.
Not another note-taking application.
Not another productivity dashboard.
Not another personality test.
A system that sits above the fragmented traces of your digital life and tries to answer one question:
«Given everything that has happened, what should I do next?»
The Architecture in One Picture
YOUR DIGITAL LIFE
│
┌────────────────┼────────────────┐
↓ ↓ ↓
Chats Email Notes
↓ ↓ ↓
└────────────────┼────────────────┘
↓
DATA INGESTION
↓
PERSONAL MEMORY
↓
COGNITIVE MODEL
↓
PATTERN ENGINE
↓
DECISION ENGINE
↓
INTERVENTION ENGINE
↓
ACTION
↓
OUTCOME
↓
LEARNING
│
└──────────→ ↺
This is the system I want PsycheForge to become.
The Bigger Question
We built search engines to retrieve information.
We built recommendation systems to predict what we might like.
We built productivity tools to manage tasks.
We built generative AI to create content.
Perhaps the next layer is something different:
«Systems that understand the longitudinal patterns of an individual and help them make better decisions.»
That is the experiment behind PsycheForge.
The long-term vision is not to build an AI that talks like you.
It is to build an intelligence layer that understands the trajectory of your decisions.
Because your digital life isn't just a collection of data.
It is a record of:
what you thought,
what you wanted,
what you did,
what you avoided,
what you decided,
and what happened afterward.
PsycheForge is an attempt to turn that record into something actionable.
Your Digital Life. Decoded.
Your Next Decision. Clear.
Open Source
PsycheForge is being developed openly on GitHub.
Repository:
https://github.com/modarresi1913/PsycheForge
If you're interested in:
- Personal AI
- AI memory
- cognitive architectures
- decision intelligence
- behavioral pattern detection
- personal knowledge systems
- privacy-first AI
- human-AI collaboration
I'd love to hear what you think.
The most interesting question isn't whether an AI can remember everything about us.
It's this:
«If an AI could see the patterns across our digital lives, could it help us make better decisions?»
Top comments (0)