Big Red Button

One button. Whatever you ask for.

The universal autonomous agent. State an outcome — build a website, compose a video, research a market, train a clone, automate a workflow — and the agent does the rest.

What Cowork does for files. What Manus does for tasks. What Perplexity does for research. What Lovable does for code. All combined. No gates. Fully autonomous.

What's in this repo

This repository hosts both the public landing page AND the live agent app at bigredbutton.app, submitted to the Runway 2026 API Hackathon.

File	Purpose
index.html	Landing page — hero with embedded demo, superpower grid, outcome examples, workflow diagram, Runway endpoint stack, receipts, memo to Cris
app.html	The live agent app — single-page UI that streams Mini-Me's reasoning and renders deliverables (images, videos, voice, HTML) as they arrive
api/agent.js	Vercel Edge Function — Claude tool-use loop wired to Runway image+video, ElevenLabs voice, web search, HTML emit, text emit. Streams Server-Sent Events
package.json	Node engines declaration for Vercel
vercel.json	Caching, security headers, function maxDuration
SETUP.md	Env vars Roger must set in Vercel dashboard
demo.mp4	60-second hackathon submission film (MINIME_HACKATHON_DEMO_v18)
listing.mp4	The real-estate listing video the agent generated end-to-end as the demo example
avatar.mp4	AI-Roger talking-head clip — Runway avatars.create + gwm1_avatars
hero1.png, hero2.png	gen4_image_turbo outputs from a brief

How it works

OUTCOME → MINI-ME → PLAN → APIS → EXECUTE → DELIVER

The user states what they want. Mini-Me plans. The plan calls APIs (Runway for video/voice, ElevenLabs for audio, web search for research, code execution for builds, file/desktop for orchestration). Output is delivered. The user only sees the button and the result.

Stack — Runway endpoints used in the demo

Endpoint	Role
gen4_image_turbo	Hero stills (Nano Banana Pro tier) — face-locked references
gen4.5	Image-to-video motion clips
gen4_aleph	Cinematic color grade pass
avatars.create	Persistent custom avatar (the user's clone)
gwm1_avatars	Audio-driven avatar talking-head video
eleven_multilingual_v2	Voice-over narration via cloned voice
eleven_text_to_sound_v2	Music bed and sound effects
tasks API	Submit / poll / retrieve orchestration spine

Plus ElevenLabs Instant Voice Clone for the founder's voice.

Demo example

The 60-second submission film shows the agent generating a finished real-estate listing video from a one-line brief. That's one example — picked for the hackathon because it's a non-trivial, multi-endpoint, end-to-end workflow with verifiable receipts.

The same agent also builds landing pages, researches markets, automates content pipelines, trains personal clones, and orchestrates MCP-style toolchains. Real-estate is the hackathon proof. Universal autonomy is the product.

Deploy (Vercel — no terminal required)

1. Open vercel.com/dashboard
2. Add New ▸ Project ▸ Import Git Repository (this repo)
3. Framework preset: Other · Output directory: ./
4. Click Deploy
5. Settings ▸ Domains ▸ assign bigredbutton.app

Done.

Receipts

The agent's actual outputs from this hackathon week — embedded on the landing page and committed in this repo:

• 60-second submission film (demo.mp4)
• Real-estate listing video (listing.mp4)
• AI-Roger founder-clone talking head (avatar.mp4)
• 14 hero stills, 6 motion clips, 1 cinematic button render
• Voice clone, persistent avatar, full orchestration source code

Total Runway compute: **