declare { Life }

Logo

Here's your Devpost submission copy:

Inspiration

Every developer knows Infrastructure-as-Code. You declare the desired state of your system — your servers, your network, your containers — and tools like Terraform automatically reconcile reality to match. We asked a simple question: what if you could do that for your life? You set intentions every day: sleep 7 hours, no meetings before 10am, gym three times a week, two hours of deep work. But reality drifts. Meetings get scheduled at 8:30am. Deep work never gets blocked. Sleep debt accumulates. And no system has ever done anything about it. We built one.

What it does

declare { life } is an infrastructure-as-code system for your personal operating system. You write a YAML manifest declaring your desired state. 3 specialized Gemini agents: Monitoring, Drift Detection, and Orchestration all run continuously, detect when reality diverges from your declared state, and take autonomous action to fix it. The remediation agent is ran on FeatherlessAPI, using Gemma 4 E4E it, to handle the complex task of resolving the schedule. The agent reschedules violating meetings, books gym sessions, carves out deep work blocks, and automatically declines incoming invites that violate your policies, generating a polite Gemini-written decline message explaining why. Every action follows a plan/apply workflow: the agent shows you exactly what it intends to do before executing. High-impact actions require your approval. Everything is logged to an immutable audit trail with Gemini's full reasoning chain.

How we built it

The backend is FastAPI with four specialized agents orchestrated in Python. Gemini 2.5 Flash powers all agent reasoning via Vertex AI, using function calling to propose and execute tool actions against calendar, health, and productivity APIs. Firebase Firestore handles real-time state synchronization — when an agent takes action, the frontend updates instantly via onSnapshot listeners without polling. The frontend is Next.js 15 with a terminal-aesthetic dashboard that visualizes desired vs. actual state as SLO cards. ElevenLabs synthesizes a spoken morning briefing summarizing your infrastructure status. Featherless API provides a specialist scheduling optimization model for complex constraint resolution. The whole system is deployed on Vercel and Railway, installable as a PWA.

Challenges we ran into

The hardest problem was designing the plan/apply workflow correctly. Agents that act too autonomously are dangerous, rescheduling your boss's meeting without asking is worse than doing nothing. We built a severity classification system (P1–P4) where high-impact actions gate on human approval while low-impact actions execute silently with an undo window. Getting Gemini's function calling to reliably return structured tool invocations, rather than text descriptions of actions, required careful prompt engineering and fallback logic for when the model returns prose instead of tool calls. Real-time synchronization between the FastAPI backend, Firestore, and the Next.js frontend through SSE streams and onSnapshot listeners also required careful coordination to avoid race conditions.

Accomplishments that we're proud of

The auto-decline feature genuinely surprised us when we first saw it work end-to-end. Someone sends a 9am meeting invite, and within seconds the agent has evaluated it against the manifest, called Gemini to write a polite decline, updated Firestore, and surfaced the response in the UI without anyone touching a keyboard. Watching the audit trail populate with Gemini's actual reasoning chains for each decision made the system feel real in a way we didn't expect. We're also proud of the IaC metaphor itself. It reframes a personal productivity problem in terms that any engineer immediately understands, and the YAML manifest editor makes the concept tangible in a way that a settings page never could.

What's next for declare { life }

The natural next step is a native mobile app with push notifications when an agent declines an invite, haptic feedback when a plan applies, a home screen widget showing your SLO status at a glance. We also want to connect to real systems beyond mock APIs: Google Calendar is already scaffolded in the codebase, Apple Health via HealthKit, and Todoist for task management. Longer term, the manifest system could support team-level SLOs, a shared agreement between collaborators about each other's time boundaries, enforced by agents that negotiate on their behalf rather than declining unilaterally.