April - Agentic AI Companion

April listens, understands, and acts. A voice agent built to complete tasks from start to finish without supervision

Comment

Inspiration

I'm a founder, and my day is a mess of tabs. Notion for tasks, Gmail for follow-ups, LinkedIn for outreach, Docs for notes, a code editor for prototypes and me in the middle, typing the same context into every single one of them.

One night around 2 AM, I was manually copying a contact's details from an email into my CRM, then into a task list, then drafting an intro message — the third time that week I'd done that exact sequence. And I thought: I have AI tools open in four tabs, and I'm still the one doing the typing. Every assistant I tried could talk about my work. None of them could do it.

I didn't need another chatbot. I needed a second pair of hands. Someone I could just tell "handle this" and walk away. That's the night April started.

What it does

April is a voice-first autonomous agent that works the way I always wished an assistant would: I speak, she executes.

Voice interaction I talk to her like a person, hands-free, while doing other things Autonomous task execution a background work queue runs multi-step jobs asynchronously; I don't wait, I delegate April Studio I describe an app or landing page out loud, and she builds working HTML with live preview and revisions Self-learning memory She remembers my preferences, my contacts, my stack, my rules — automatically, from our conversations, stored locally. I never re-explain myself Task board My startup, coding, and daily tasks in one place that she manages with me Superconnector she knows what I offer and what I need, tracks my network, spots strategic matches, and drafts the intros I used to write at 2 AM Google integration Connected to Gmail, Drive, and Sheets, so she acts on my real accounts, not a sandbox Hybrid inference Cloud when available, 100% local when not; she works even when my WiFi doesn't

How we built it

April runs on [Gemini via the Gemini API] for reasoning and planning, orchestrated through [Google ADK] in a loop of intent parsing → planning → tool execution → memory write-back. The backend runs on [Cloud Run] with [Firestore] for persistence. Voice uses the Web Speech API with neural TTS output. Memory lives on-device by default with export/import backup — my data stays mine.

Challenges we ran into

Trust Letting an agent touch real accounts means getting permissions, scopes, and guardrails right before anything else Voice turn-taking Teaching April not to talk over me, and to handle noise and interruptions gracefully Async execution Keeping the conversation alive while jobs run in the background Memory quality Deciding what's worth remembering automatically, without filling her head with noise

Accomplishments that we're proud of

  • I genuinely use April every day, she's my workflow
  • One spoken sentence becomes finished work: an app built, a task filed, an intro drafted
  • Her memory compounds; by week two she knew things about my business I'd only mentioned once
  • Local-first design: she works offline and my data never has to leave my machine

What we learned

The gap between "AI that answers" and "AI that acts" is enormous, and everything valuable lives on the far side of it. I also learned that memory is the real product: an agent that remembers you becomes irreplaceable in a way no smarter model can match. And building for myself first kept me honest if April couldn't save me time, nothing else mattered.

What's next for April - Agentic AI Companion

Deeper account control, full Gmail and Calendar automation with granular permissions, multi-agent delegation where April spawns sub-agents for parallel jobs, a proactive mode where she flags what needs doing before I ask, and mobile apps so the founder's second pair of hands is always in reach.

Built With

  • antigravity
Share this project:

Updates

Submission history