Skip to content

mumbl

mumbl is a multimodal AI agent orchestration platform that transforms messy, spoken stream-of-consciousness into structured, real-world actions.

Code Video Cerebral Valley

Core trick

Gemini function calling splits spoken, unstructured input into intents and dispatches seven specialist agents in parallel.

Tags

voiceproductivityagents

Description

mumbl is a multimodal AI agent orchestration platform that transforms messy, spoken stream-of-consciousness into structured, real-world actions. Users speak naturally — "I was up too late, need to move my gym session, find somewhere healthy for dinner, and draft that business proposal" — and Mumbl. interprets the intent, dispatches 7 specialized AI agents in parallel (Email, Calendar, Maps, Notes, Nutrition, Context, Website), and delivers actionable outcomes: emails drafted, restaurants found with map embeds, documents generated in Notion, calendar events rescheduled, and more. Built with Next.js 16, Gemini 2.5 Flash with function calling for agent orchestration, ElevenLabs for speech-to-text and text-to-speech, Google Places API for real venue search, and Notion API for live document creation. The tagline: "Say it broken. Get it done."
GrigoryDabizha/mumbl★ 00 forksTypeScriptMITCreated Mar 21, 2026 · last push Mar 21, 2026

README excerpt

# Mumbl.

**Say it broken. Get it done.**

A multimodal AI agent orchestration platform where raw, human, messy input becomes real-world actions — executed in parallel, delivered as finished outcomes.

Built for the **Zero to Agent — Vercel × Google DeepMind Hackathon London** (March 21, 2026).

---

## 🚀 Live Demo

**We are currently hosted on a live URL as well!**
*(https://mumbl-five.vercel.app/)*

---

## 🧠 The Concept

Every productivity tool built to date has required users to think in the machine's language — to fill forms, structure prompts, navigate menus, and context-switch between a dozen isolated apps. **Mumbl inverts this entirely.**

You speak the way humans actually think: broken, half-formed, overlapping, real. A swarm of AI agents powered by **Google DeepMind's Gemini 2.5 Flash** interprets your intent, resolves conflicts, and executes across every domain of your life simultaneously. Your calendar, your dietary goals, your commute, your playlists, your inbox—all managed from a single stream of consciousness.

## 🛠️ Stack

*   **Brain**: Gemini 2.5 Flash (via Vercel AI SDK)
*   **Voice (TTS/STT)**: ElevenLabs
*   **Memory & State**: Supabase
*   **Framework**: Next.js 15 (App Router)
*   **Styling & UI**: Tailwind CSS, Framer Motion, Vanilla CSS (Glassmorphism design)
*   **State Management**: Zustand
*   **Integrations**: Notion, Google Calendar/Maps (simulated), Spotify (simulated)

## 💻 Getting Started (Local Development)

First, install dependencies:

```b

Team