TimeSplit

https://github.com/Zenieverse/TimeSplit

Inspiration Most tools that help people plan their future reduce life to checklists, metrics, or generic advice. We wanted to explore something more human: the emotional and ethical consequences of decisions over time. The idea for TimeSplit came from a simple question: If you could talk to your future self, would you listen—or argue back? We were inspired by role-playing games, behavioral psychology, and the realization that long-term thinking is difficult because consequences feel abstract. Gemini 3 gave us the opportunity to turn abstract futures into interactive characters, making long-term decision-making tangible, emotional, and engaging. What it does TimeSplit is an interactive simulation game where users make high-impact life decisions and then confront multiple AI-generated future versions of themselves. After answering a short set of questions, Gemini 3 generates distinct future selves (e.g., 1 year, 5 years, 20 years ahead). Each future self has its own personality, values, memories, and perspective shaped by the user’s decisions. Users can chat, debate, or negotiate with these future selves, uncovering tradeoffs, regrets, and opportunities. The experience turns self-reflection into gameplay, where every conversation reshapes the timeline. How we built it TimeSplit is built around the Gemini 3 API, which serves as a multi-agent simulation engine. We use Gemini 3 to: Generate autonomous future-self agents from a single user profile Maintain long-term memory and consistent personalities per agent Perform causal reasoning across years of simulated decisions Dynamically generate personalized life events and dialogue Each future self is isolated into its own reasoning context, allowing Gemini 3 to produce believable disagreements, emotional responses, and evolving perspectives. A lightweight web frontend presents these agents through chat, timeline visualizations, and decision summaries. Challenges we ran into One major challenge was preventing future selves from becoming too similar. Early versions converged into generic advice-giving personalities. We solved this by: Assigning distinct value systems and priorities to each agent Separating memory and reasoning context per timeline Introducing tension by forcing agents to defend their own outcomes Another challenge was balancing open-ended AI creativity with a clear game structure. We addressed this by framing interactions around decisions, consequences, and measurable timeline shifts. Accomplishments that we're proud of Turning long-term thinking into a playable, replayable experience Using Gemini 3 as a world and identity simulator, not just a chatbot Creating emotionally distinct AI agents that can disagree with the user Building a project that blends gaming, productivity, and self-reflection Most importantly, we built something that feels personal—no two playthroughs are the same. What we learned We learned that Gemini 3 excels at multi-perspective reasoning when given clear identity constraints and memory separation. Treating the model as multiple agents rather than a single assistant unlocked far richer behavior. We also learned that users engage more deeply when AI responses carry emotional weight and consequences, not just information. Narrative consistency matters as much as technical correctness. What's next for TimeSplit Next, we plan to: Add multiplayer timeline comparison (compare futures with friends) Introduce habit tracking that feeds back into simulations Expand into education and career-planning modes Visualize timelines with richer, interactive graphs Long-term, we see TimeSplit as a new kind of interface for decision-making—where the future isn’t predicted, but participates.

Built With

• all