Inspiration
Most productivity tools start with an empty calendar or task list and ask the user to organize everything themselves.
Cronnus starts somewhere else: with the person.
Work, study, routines, goals, energy, constraints, habits, available time, and the way each person prefers to organize their life all affect whether a plan is actually realistic.
We built Cronnus to explore a simple idea:
What if your planner could understand your context before asking you to plan?
What it does
Cronnus is an AI-powered personal planning system built around Gemini.
During onboarding, Cronnus learns about the user's real-life context, including responsibilities, schedules, goals, habits, preferences, and constraints.
Gemini 2.5 Flash then transforms that context into a personalized seven-day plan.
But Cronnus is not designed to let an AI silently modify someone's life.
Its Coach can understand the user's current context, tasks, goals, recent activity, and conversation history, then propose structured actions such as creating, modifying, deleting, or reorganizing tasks.
Those proposals pass through deterministic validation for scheduling constraints, limits, ownership, and safety. Important changes remain pending until the user explicitly confirms them.
Gemini reasons and proposes. Cronnus validates. The user decides.
Cronnus also provides proactive daily support through scheduled notifications and weekly check-ins, turning planning into an ongoing experience rather than a one-time generated schedule.
How we built it
Cronnus uses Gemini 2.5 Flash through Vertex AI for three core production flows:
- understanding onboarding and progressively building the user's profile;
- generating the personalized initial plan;
- powering the contextual Coach and generating structured action proposals.
The application runs on Google Cloud using Cloud Run, Cloud SQL for PostgreSQL, Cloud Scheduler, Firebase Cloud Messaging, Secret Manager, Cloud Logging, Cloud Build, and Artifact Registry.
The planning architecture combines LLM reasoning with deterministic software:
User context → Gemini reasoning → structured proposal → deterministic validation → human approval → persistence
This hybrid architecture gives Gemini meaningful decision-making responsibility while preserving predictable constraints and human control.
What we accomplished
Cronnus is not a concept or prototype video. It is a deployed product with real beta users.
During beta testing, users have completed onboarding, generated personalized plans, interacted with the Coach, completed tasks, received proactive notifications, and provided feedback.
Gemini calls through Vertex AI have been verified in the deployed production environment without fallback.
We built and operated the beta with approximately US$5.24 in Google Cloud usage, demonstrating that personalized AI planning can be delivered with a very low infrastructure cost during early-stage validation.
Challenges we ran into
The hardest problem was not simply connecting an LLM to a planner. It was deciding where AI should have autonomy and where deterministic software or human confirmation should remain in control.
Scheduling must respect real constraints. Personal actions should not be silently changed by a model. AI failures should not make the application unusable.
That led us to build Cronnus as a hybrid system instead of a fully autonomous agent: Gemini handles contextual reasoning, deterministic code validates its proposals, and users retain final control over meaningful changes.
What's next
The current beta validates the core loop: understand the person, generate a realistic plan, accompany execution, learn from interaction, and propose useful adjustments.
Next, we want to deepen that loop by improving long-term adaptation, expanding integrations, strengthening personalization from behavioral signals, and validating Cronnus with a larger user base.
Our long-term goal is simple:
Turn personal planning from something people have to constantly manage into a system that actively helps them move forward.
Built With
- cloud-run
- cloud-scheduler
- cloud-sql
- firebase-cloud-messaging
- gemini
- google-cloud
- next.js
- postgresql
- prisma
- pwa
- typescript
- vertex
- zod
Log in or sign up for Devpost to join the conversation.