miguelito.abdibedel.com
One chat, five agents
Single surface. Orchestrator routes intent. The routing disappears.
← All products
03 · Approval-gated beta
Miguelito
The system that runs your day, not your to-do list.
Approval-gated betaFounder & Product Lead
RIGHT NOW · Apr 2026
STATUS
Beta · gated
VERSION
v3.4.0
PLATFORMS
iOS · Android · Web
NEXT
First 50 beta users
The bet
“Five apps. Zero of them talked to each other. The planning wasn't the problem — execution was. So I built the system that learns you.”
What it is
Miguelito is the execution layer. You talk to it. It manages your calendar, tasks, goals, fitness, and client commitments through specialised agents behind a single conversation.
Built for people who know how to plan but don't have the bandwidth to hold all of it in their head every day.
My role
I built the product end-to-end as founder and product lead, from orchestrator architecture through live multi-agent execution loops and wearable integrations.
See it
Inside Miguelito.
One screen. Two decisions. The shape of the product.
~/specialist-agents
Specialist agents
5
Calendar · Fitness · Todo · Goal · PM. Each agent does one thing well.
~/privilege-lock-system
Privilege Lock System
3 tiers
Gentle, Standard, Hard. Users enforce their own constraints. Guardian Mode can't be skipped.
How it's built
Under the hood.
One conversation on the surface. An orchestrator underneath that classifies intent and routes to the right specialist agent. Five agents - Calendar, Fitness, Todo, Goal, and Product Manager - each doing one thing well.
Stack
- Backend
- Node.js, Express, Socket.io, Firebase Firestore
- AI
- Deepseek LLM, aggressive in-memory caching, about $0.14 per 1M tokens
- Frontend
- Flutter across iOS, Android, and web
- Real-time
- WebSocket chat with push notifications
- Auth
- Firebase Auth with Google OAuth for Calendar access
- Integrations
- HealthKit and Health Connect for wearable data
Key product decisions
- 01
Orchestrator architecture, not a mega-prompt
Central intent classifier dispatches to the right specialist agent. Prevents feature sprawl. Keeps every agent sharp.
- 02
Privilege Lock System
Three tiers - Gentle, Standard, Hard - let users enforce their own constraints on themselves. Guardian Mode adds a timed focus pause you can't skip.
- 03
Chronotype-aware scheduling
The calendar agent reads whether you're a morning lark or night owl and builds around your actual energy, not convention.
- 04
Wearable data is ground truth, not decoration
Stress, sleep, and recovery signals feed real decisions - dropping cognitive load, protecting evenings, reshuffling deep work blocks.
- 05
Guided intake, not forms
Conversational flows for goals, fitness, and tasks. No cold starts. No blank screens.
The build log
What shipped. What changed. What the numbers say.
Updated monthly. What shipped, what changed, what the numbers say.
April 2026 - Beta opens
- What shipped
- v3.4.0 - all five agents live, daily briefing, real-time chat, wearable sync, privilege lock system
- What changed
- Moved from personal tool to approval-gated beta
- The numbers
- TBD - fill in end of month
Where it stands today
Version 3.4.0 - production-ready. Currently opening an approval-gated beta. Recent focus has been calendar intelligence, privilege lock hardening, and guided intake reliability.
Live on iOS, Android, and web.
Objectives
North star metric
Full-plan follow through
Days where the user follows Miguelito's plan end-to-end. Target: 40%+ of active users weekly.
Key success metrics
- 70%+ return daily in the first week of use
- 60%+ engage with real-time reschedule when triggered
- First paying user within 7 days of beta opening
Target milestones
- Full agent suite live - shipped
- First 50 approval-gated beta users
- First feedback loop that changes a product decision
- First 10 users through a full planned day
Lessons
Wins and failures. No highlight reel.
What’s working
The orchestrator pattern. Users don't think about which agent they're talking to - they just talk. The routing disappears and that's what makes it feel like one assistant instead of five tools stacked behind a chat window.
What isn’t
Cold-start onboarding. Miguelito gets sharper the longer you use it, but the first three days - before it's learned anything about you - don't feel much better than any other task app. The learning curve is the value and also the biggest churn risk.
What I’d do differently
Build the Daily Briefing before the agents, not after. It's the feature that shows users the system is working - the agents do the work, but the briefing is where the user feels it. I built the engine first and the dashboard second. Should've been the other way round.
Try it live
Miguelito — live at miguelito.abdibedel.com
Miguelito beta is approval-gated. If you run your own thing and want a system that learns you, request access.