FutureOS

FutureOS is a multi-agent Personal AI operating system that transforms long-term goals into actionable daily plans. Powered by NVIDIA Nemotron through Nebius Token Factory

Comment

FutureOS — Your Personal AI Life Operating System

Elevator Pitch

FutureOS is a multi-agent Personal AI operating system that transforms long-term goals into actionable daily plans. Powered by NVIDIA Nemotron through Nebius Token Factory, it combines intelligent planning, persistent graph-based memory, progress analytics, personalized coaching, and opportunity research into one unified assistant. Unlike traditional productivity tools that simply track tasks, FutureOS brings specialized AI agents together to help users plan, execute, learn, and continuously improve.

Inspiration

We noticed that most productivity applications are passive. They allow users to create tasks, set reminders, and track progress, but they rarely understand the bigger picture behind someone's goals.

People often struggle to maintain consistency, identify their weaknesses, find relevant learning opportunities, and adapt their plans when circumstances change.

We imagined a personal AI system that works more like a team of specialized advisors than a conventional chatbot. An assistant that remembers previous activities, understands long-term objectives, breaks complex ambitions into achievable milestones, and provides actionable guidance based on actual progress.

This led us to build FutureOS: an adaptive AI life operating system where multiple agents collaborate around an individual's goals and long-term development.

What it does

FutureOS coordinates six specialized agents:

  • Planner Agent: Converts long-term goals into structured roadmaps, milestones, and actionable daily missions.
  • Execution Agent: Manages mission completion, tracks achievements, and awards experience points (XP).
  • Memory Agent: Maintains persistent graph-based memory of goals, completed missions, and learning activities across sessions.
  • Analytics Agent: Analyzes completion rates, accumulated XP, and progress to generate meaningful performance insights.
  • Coach Agent: Provides personalized recommendations, identifies potential consistency issues, and encourages users at important milestones.
  • Research Agent: Searches for relevant internships, courses, project ideas, and professional communities based on the user's goals.

Key features:

  • Goal-driven mission generation.
  • Persistent graph-based memory.
  • Multi-agent activity tracking.
  • Personalized progress insights and coaching.
  • Opportunity discovery through web search.
  • Interactive analytics and achievement tracking.
  • Structured AI outputs for roadmaps, strategies, and insights.

The system is designed to turn a user's long-term ambition into a continuously trackable execution process.

How we built it

FutureOS is built using Jac, a graph-native programming language and full-stack development framework.

Our architecture combines Jac's graph-based data model with specialized AI functions and a multi-agent workflow.

AI inference and model integration:

We integrated NVIDIA Nemotron 3.5 Lightning through Nebius Token Factory using its OpenAI-compatible chat-completions interface and Jac's byLLM integration.

Nemotron is configured to power the Planner, Analytics/Coach, and Strategy generation workflows. This enables the application to generate structured mission plans, personalized progress insights, and goal-oriented strategies.

Graph-based memory and persistence:

We use Jac's graph-native data model to maintain relationships between goals, missions, memory entries, and agent activity logs. This allows the application to preserve user progress and maintain context across sessions.

Agent architecture:

Each specialized agent has a defined responsibility. Their workflows are connected through shared application data, structured AI outputs, and graph-backed state.

Technology stack:

  • Jac / Jaseci
  • Python
  • Jac byLLM
  • NVIDIA Nemotron 3.5 Lightning
  • Nebius Token Factory
  • Jac graph database
  • React-based frontend
  • Tailwind CSS
  • DuckDuckGo Search

We also introduced environment-based API configuration to keep credentials separate from application source code and added an inference smoke test for validating Nebius connectivity.

Challenges we ran into

One of our biggest challenges was coordinating multiple specialized agents without making the application feel like a collection of disconnected AI features.

We needed a consistent data model so that planning, execution, memory, and analytics could work with the same user goals and progress history.

Another challenge was generating reliable structured outputs. Mission plans, progress reports, and strategies need predictable fields and formats so they can be consumed by the application's workflows rather than simply displayed as conversational text.

Integrating Nebius Token Factory with the existing Jac byLLM architecture also required adapting the model configuration, endpoint, and environment-based authentication.

Finally, we had to account for inference failures. We retained deterministic fallback mechanisms so core application workflows can remain usable when an external model request is unavailable, while explicitly distinguishing fallback events from successful Nemotron inference.

Accomplishments that we're proud of

  • Built a multi-agent personal AI system with six specialized agents.
  • Implemented graph-backed persistence for goals, missions, memories, and agent activity.
  • Connected long-term goal planning with daily execution and progress tracking.
  • Integrated NVIDIA Nemotron through Nebius Token Factory into the AI planning and coaching workflows.
  • Implemented structured AI outputs for mission plans, strategies, and progress insights.
  • Created an interactive dashboard for monitoring agent activity and personal progress.
  • Designed fallback mechanisms to improve application resilience when model inference is unavailable.
  • Extended an existing personal AI application with a dedicated Nebius and NVIDIA inference integration.

Our biggest accomplishment is bringing planning, execution, memory, research, and coaching together into one unified personal AI experience.

What we learned

Building FutureOS taught us that useful personal AI requires more than generating intelligent responses.

Persistent memory, structured outputs, specialized responsibilities, and meaningful interactions between agents are essential for creating an assistant that can support users over time.

We also learned the importance of choosing the right model for structured reasoning tasks, integrating open models through hosted inference infrastructure, managing credentials securely, and designing fallbacks for external service failures.

Working with NVIDIA Nemotron through Nebius Token Factory gave us practical experience integrating hosted open-model inference into an existing multi-agent application.

Most importantly, we learned that AI should not simply tell people what to do. It should help them turn intentions into measurable actions.

What's next for FutureOS

We envision FutureOS evolving into a genuinely adaptive personal AI companion.

Our next priorities include:

  1. Long-term contextual memory: Build richer memory relationships that identify recurring patterns, preferences, achievements, and obstacles over time.
  2. Autonomous execution: Enable agents to coordinate and carry out approved actions across users' selected applications and workflows.
  3. Adaptive planning: Automatically revise mission roadmaps based on changing priorities, completed tasks, and observed performance.
  4. Intelligent opportunity discovery: Improve research using live information retrieval and more relevant opportunity ranking.
  5. Personal AI skills: Introduce reusable skills that users can customize for career development, learning, productivity, and personal projects.
  6. Private and user-controlled AI: Strengthen data ownership, privacy, and control over the tools and information accessible to agents.
  7. Scalable inference: Explore Nebius serverless infrastructure and intelligent model routing to improve responsiveness and control inference costs.

Our long-term vision is to make FutureOS a personal AI operating system that does more than manage tasks: it understands individual goals, remembers the journey, and helps people make consistent progress toward the future they want to build.

Built With

Share this project:

Updates

Submission history