🚧 Active Development
AURIX is currently under active development.
⚠️ The current executable build is based on the AURIX Alpha prototype and is intended for research and experimentation purposes only.
Privacy-first AI Productivity Coach powered by Local LLMs
AURIX is a privacy-first AI productivity system that evolved from a simple Todo application through user research, technical experimentation, and continuous product iteration.
Rather than integrating AI as a feature without a clear purpose, AURIX focuses on solving a fundamental problem discovered through user validation:
How can technology help people better understand their own behaviors and improve over time?
AURIX did not begin as an AI project.
It started as a simple Todo Dashboard built around one question:
"Would anyone actually find this useful?"
Through early user validation, I discovered that people do not simply need another task management tool.
They need a system that helps them understand:
- Why tasks remain unfinished
- Which behaviors lead to better outcomes
- How they can continuously improve their workflow
This insight became the foundation of AURIX:
Execution → Feedback → Improvement
While traditional productivity tools often stop at task completion, AURIX explores the missing feedback layer powered by local AI.
While reviewing previous projects, I identified a recurring pattern:
- Development was heavily focused on implementation
- User feedback loops were often missing
- Projects ended once a functional prototype was completed
I wanted to shift the focus from:
"Can I build this?"
to:
"Does this actually solve a user's problem?"
To validate this assumption, I built a lightweight Todo Dashboard and conducted early-stage usability testing.
The study focused on four key questions:
- Is the interface intuitive?
- Is the workflow efficient?
- What challenges do users encounter?
- What improvements do users expect?
- 3 users participated in usability testing
- Both positive and negative feedback were collected
- The goal was to identify fundamental usability issues rather than visual improvements
- Fast and responsive user experience
- Simple workflow without onboarding requirements
- Clear separation between different task categories
- No priority management
- No scheduling or reminder system
- No feedback after task completion
Users were not looking for another Todo application.
They wanted a system that could help them improve.
The main limitation was not task management itself, but the absence of a feedback loop after execution.
Based on this insight, AURIX was designed as an AI-powered behavioral feedback system.
Instead of only recording tasks, AURIX analyzes personal productivity patterns and generates meaningful insights.
Examples:
- Weekly productivity summaries
- Habit consistency analysis
- Task completion trends
- Goal progress tracking
- Personalized behavioral recommendations
AURIX is designed as a personal AI system, not a general-purpose chatbot.
The system primarily focuses on the user's own data:
- Personal habits
- Task history
- Productivity patterns
- Long-term goals
Example:
Your long-term goals showed lower completion rates this week. Tasks started before 10 AM have a higher completion rate. Consider scheduling important work earlier in the day.
Privacy is a fundamental design principle of AURIX.
All user data remains on-device by avoiding external AI APIs.
- Offline-first operation
- Local inference
- No cloud dependency
- User-owned data
- Lower operational cost
Large language models provide impressive capabilities, but productivity data has unique characteristics:
- Small and structured datasets
- Limited context requirements
- High privacy expectations
- Need for fast response times
AURIX explores the idea that:
Smaller models can become highly capable when the surrounding system is designed effectively.
- C++
- Gemma 2B
- Ollama
- Local RAG
AURIX has evolved through multiple iterations to explore both product value and the challenges of building reliable local AI systems.
Goal: Validate the user problem
Implemented:
- Basic task management
- Local-first architecture
- Early-stage usability testing
Outcome:
Discovered that task management alone does not provide enough long-term value.
↓
Goal: Explore personalized AI feedback using local LLMs
Implemented:
- Local LLM inference
- Gemma 2B integration
- Basic memory retrieval
- Context-aware responses
During development, several architectural challenges were identified:
- Prompt injection risks
- Memory poisoning vulnerabilities
- Lack of semantic retrieval
- Absence of memory reliability scoring
These findings guided the next stage of architectural improvement.
↓
Goal: Improve system reliability and context safety
The beta version explored stronger protection mechanisms.
However, improving security introduced new engineering trade-offs:
- Reduced retrieval effectiveness
- Challenges maintaining RAG quality
- Increased architectural complexity
This experiment highlighted the difficulty of balancing retrieval quality, security, and usability in local AI memory systems.
↓
After evaluating both product direction and technical constraints, AURIX is being rebuilt around its original vision:
Building a privacy-first personal intelligence system.
Current focus:
- Reliable local RAG architecture
- Structured memory management
- Semantic retrieval
- Behavioral feedback generation
The goal is not simply to build another AI assistant.
The goal is to create a system that helps people understand themselves through their own data.