One sentence. One meeting. One live app.
What • Council • Brainstorm • How • Architecture • Stack • Start • Deploy
vibeDeploy takes a sentence describing your app idea — or a YouTube URL — and turns it into a live, deployed application on DigitalOcean. No coding required.
Between your idea and the deployed app, The Vibe Council — a panel of 6 specialized AI agents — holds a live debate meeting to evaluate feasibility, challenge assumptions, and score your idea before a single line of code is written.
"I want a restaurant queue management app with QR codes"
↓
🏛️ The Vibe Council convenes
🏗️ Architect → 🔭 Scout → 🛡️ Guardian → ⚡ Catalyst → 🎯 Advocate
🧭 Strategist synthesizes → Vibe Score™: 78.5 → 🟢 GO
↓
📄 PRD + Tech Spec + API Spec generated
💻 Frontend + Backend code generated
🚀 Deployed to DigitalOcean
↓
✅ https://your-app.ondigitalocean.app
| Traditional Vibe Coding | vibeDeploy |
|---|---|
| Generates code only | Idea → live deployed app |
| Single AI writes code | 6 AI experts debate before building |
| You judge quality | AI team self-validates feasibility |
| Deployment is separate | Auto-deploy + live URL |
| Text input only | YouTube URL also works |
| One mode fits all | Evaluate or Brainstorm — you choose |
vibeDeploy's unique decision-making framework. Six specialized AI agents engage in structured debate — challenging, questioning, and refining your idea through a live meeting.
| Agent | Role | Focus | Core Question |
|---|---|---|---|
| 🏗️ Architect | Technical Lead | Tech stack, complexity, feasibility | "How do we build this?" |
| 🔭 Scout | Market Analyst | Competition, trends, market fit | "Who wants this?" |
| 🛡️ Guardian | Risk Assessor | Security, legal, failure modes | "Why could this fail?" |
| ⚡ Catalyst | Innovation Officer | Uniqueness, disruption, impact | "What makes this special?" |
| 🎯 Advocate | UX Champion | User perspective, accessibility | "Will people use this?" |
| 🧭 Strategist | Session Lead | Synthesis, scoring, verdict | "What's the verdict?" |
Phase 1: INDIVIDUAL ANALYSIS → 5 agents analyze in parallel
Phase 2: CROSS-EXAMINATION → Structured debate (Architect↔Guardian, Scout↔Catalyst, Advocate challenges)
Phase 3: SCORING → Each agent scores their axis (0-100)
Phase 4: VERDICT → Strategist calculates Vibe Score™ → GO / CONDITIONAL / NO-GO
Vibe Score™ = (Tech × 0.25) + (Market × 0.20) + (Innovation × 0.20) + ((100 - Risk) × 0.20) + (UserImpact × 0.15)
≥ 70 → 🟢 GO Build and deploy immediately
50-69 → 🟡 CONDITIONAL Propose scope reduction, user decides
< 50 → 🔴 NO-GO Detailed failure report + alternatives
Not ready to evaluate? Brainstorm first. The same 5 council agents switch from judge mode to creative ideation mode — generating possibilities instead of scores.
Your idea → 💡 Brainstorm Mode
↓
🏗️ Architect: "3 innovative tech stack combos..."
🔭 Scout: "3 untapped market segments..."
⚡ Catalyst: "3 ways to make this 10x more disruptive..."
🛡️ Guardian: "3 risks turned into competitive advantages..."
🎯 Advocate: "3 delightful UX micro-interactions..."
↓
🧭 Strategist synthesizes all insights into:
• Top 5 ranked ideas across agents
• Recurring themes & synergies
• 3 quick wins to implement first
• Strategic direction recommendation
Each agent generates ideas, opportunities, wild cards, and action items from their unique perspective. The Strategist then synthesizes everything into a coherent creative brief — no scoring, no judgment, pure creative fuel.
┌──────────────────────────────────────────────────────────────┐
│ INPUT │
│ Text: "Build a recipe sharing community with AI suggestions" │
│ — or — │
│ YouTube: https://youtube.com/watch?v=... │
│ (auto-extracts idea from transcript + key frame analysis) │
└──────────────────┬───────────────────────────────────────────┘
▼
┌──────────────────────────────────────────────────────────────┐
│ THE VIBE COUNCIL (Live AI Debate) │
│ │
│ Phase 1: 5 agents analyze independently (parallel) │
│ Phase 2: Cross-Examination debates │
│ Phase 3: Each agent scores their axis │
│ Phase 4: Strategist → Vibe Score™ → Decision │
└──────────────────┬───────────────────────────────────────────┘
▼
┌──────────────────────────────────────────────────────────────┐
│ BUILD (if GO or CONDITIONAL-accepted) │
│ │
│ 📄 PRD + Tech Spec + API Spec + DB Schema │
│ 💻 Next.js frontend + FastAPI/Express backend │
│ 🎨 App logo + placeholder images (AI-generated) │
└──────────────────┬───────────────────────────────────────────┘
▼
┌──────────────────────────────────────────────────────────────┐
│ DEPLOY │
│ │
│ 📦 GitHub repo created automatically │
│ 🚀 Deployed to DigitalOcean App Platform │
│ 🔗 Live URL: https://your-app.ondigitalocean.app │
└──────────────────────────────────────────────────────────────┘
┌─────────────────────────────────┐ ┌──────────────────────────────┐
│ Frontend (App Platform) │ │ Agent Backend (Gradient ADK) │
│ Next.js 15 + shadcn/ui │────▶│ Python + LangGraph │
│ Static Site │◀────│ @entrypoint → POST /run │
│ .do/app.yaml │ SSE │ .gradient/agent.yml │
└─────────────────────────────────┘ └──────────────────────────────┘
│
┌──────────────────────────┼──────────────────┐
▼ ▼ ▼
┌─────────────┐ ┌──────────────┐ ┌──────────────┐
│ Gradient AI │ │ Managed │ │ DO Spaces │
│ Inference │ │ PostgreSQL │ │ (S3) │
│ (8 models) │ │ │ │ │
└─────────────┘ └──────────────┘ └──────────────┘
Two independent deployments:
- Agent →
gradient agent deployto ADK infrastructure (FREE during preview) - Frontend → DO App Platform via
.do/app.yaml
| Component | Technology | DigitalOcean Service |
|---|---|---|
| Frontend | Next.js 15 + shadcn/ui + Tailwind CSS | App Platform |
| Agent Backend | Python 3.12 + Gradient ADK + LangGraph | Gradient ADK |
| Database | PostgreSQL 16 | Managed PostgreSQL |
| Storage | S3-compatible object storage | Spaces |
| AI Models | 4 open-source model families via Serverless Inference | Gradient AI Platform |
| Knowledge Base | RAG over DO docs + framework patterns | Gradient KB |
| # | Feature | Usage |
|---|---|---|
| 1 | ADK (Agent Development Kit) | Agent system hosting |
| 2 | Serverless Inference (Text) | 6 council agents + code generation |
| 3 | Serverless Inference (Image) | App logos, UI mockups, placeholders |
| 4 | Knowledge Bases (RAG) | DO docs, framework best practices |
| 5 | Multi-Agent Routing | LangGraph Send API parallel execution |
| 6 | Evaluations | Agent response quality measurement |
| 7 | Tracing | Full pipeline debugging |
| 8 | Function Calling | DO API, GitHub API, YouTube API |
| Role | Model | Type | Cost (per 1M tokens) |
|---|---|---|---|
| Council (5 agents) | openai-gpt-oss-120b |
Open-Source (120B) | $0.10 / $0.70 |
| Strategist | deepseek-r1-distill-llama-70b |
Open-Source (70B) | $0.99 / $0.99 |
| Cross-Examination | deepseek-r1-distill-llama-70b |
Open-Source (70B) | $0.99 / $0.99 |
| Code Generation | openai-gpt-oss-120b |
Open-Source (120B) | $0.10 / $0.70 |
| CI Repair | openai-gpt-oss-120b |
Open-Source (120B) | $0.10 / $0.70 |
| Doc Generation | alibaba-qwen3-32b |
Open-Source (32B) | $0.25 / $0.55 |
| Input Processing | openai-gpt-oss-120b |
Open-Source (120B) | $0.10 / $0.70 |
| Decision Gate | deepseek-r1-distill-llama-70b |
Open-Source (70B) | $0.99 / $0.99 |
| Brainstorm Agents | openai-gpt-oss-120b |
Open-Source (120B) | $0.10 / $0.70 |
| Brainstorm Synthesis | deepseek-r1-distill-llama-70b |
Open-Source (70B) | $0.99 / $0.99 |
| Web Search | mistral-nemo-instruct-2407 |
Open-Source (12B) | $0.30 / $0.30 |
| Image Generation | fal-ai/flux/schnell |
Open-Source (fal) | ~$0.003 / image |
Multi-model open-source AI: 4 model families via DO Serverless Inference Cost per full deployment: ~$0.12 | $200 credits = ~1,600+ deployments
- Python 3.12+
- Node.js 20+
- Gradient CLI
- DigitalOcean account (free $200 credits for hackathon)
# Clone
git clone https://github.com/Two-Weeks-Team/vibeDeploy.git
cd vibeDeploy
# Agent Backend
cd agent/
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env # Fill in your keys
gradient agent run --host 0.0.0.0 --port 8080
# Frontend (new terminal)
cd web/
npm install
NEXT_PUBLIC_AGENT_URL=http://localhost:8080 npm run devOpen http://localhost:3000 and enter your idea.
Live at: https://vibedeploy-7tgzk.ondigitalocean.app
The FastAPI backend and Next.js frontend are both deployed via DigitalOcean App Platform:
doctl apps create --spec .do/app.yaml
# Or push to main → auto-deploy (deploy_on_push: true)
# → https://vibedeploy-7tgzk.ondigitalocean.app# Create agent via doctl (attach Knowledge Base for RAG)
doctl gradient agent create \
--name "vibedeploy-agent" \
--project-id "<project-id>" \
--model-id "<model-uuid>" \
--region "tor1" \
--instruction "Your agent instruction" \
--knowledge-base-id "<kb-uuid>"doctl gradient knowledge-base create \
--name vibedeploy-docs \
--region tor1 \
--project-id "<project-id>" \
--embedding-model-uuid "<embedding-model-uuid>" \
--data-sources '[{"web_crawler_data_source":{"base_url":"https://docs.digitalocean.com/products/app-platform/","crawling_option":"UNKNOWN","embed_media":false}}]'vibeDeploy/
├── web/ # Next.js Frontend → App Platform
│ ├── src/app/ # Pages (landing, meeting, brainstorm, result)
│ ├── src/components/ # UI components (council, brainstorm, scores, deploy)
│ └── src/lib/ # SSE client, API helpers
├── agent/ # Python Agent → Gradient ADK
│ ├── main.py # @entrypoint (Gradient ADK)
│ ├── server.py # Local FastAPI server (dev)
│ ├── graph.py # Evaluation pipeline (LangGraph StateGraph)
│ ├── graph_brainstorm.py # Brainstorm pipeline (LangGraph StateGraph)
│ ├── nodes/ # Pipeline nodes (input, council, brainstorm, build)
│ ├── council/ # 6 Vibe Council agent definitions
│ ├── tools/ # YouTube (yt-dlp), GitHub, DO, search tools
│ └── .gradient/agent.yml # ADK config (3 lines)
├── .do/app.yaml # App Platform spec (frontend only)
├── docs/reference/ # 10 planning documents
└── LICENSE # MIT
Comprehensive planning docs in docs/reference/:
| # | Document | Description |
|---|---|---|
| 01 | Hackathon Rules | Rules, timeline, $200 credits guide |
| 02 | Judging Criteria | 4 criteria × 25% each |
| 03 | Submission Checklist | Devpost requirements |
| 04 | Prizes | $20K total, 6 categories |
| 05 | Gradient AI Platform | SDK patterns, models, pricing |
| 06 | DO Products | App Platform, DB, Spaces |
| 07 | Winning Strategy | Demo video, judge targeting |
| 08 | Product Concept | Vibe Council, scoring system |
| 09 | Agent Architecture | LangGraph, state, nodes, prompts |
| 10 | Technical Plan | Deployment, costs, DB schema |
| Event | DigitalOcean Gradient AI Hackathon 2026 |
| Prizes | $20,000 total (1st: $5,000 + $2,000 DO credits) |
| Deadline | March 18, 2026, 5:00 PM EDT |
| Team | Two-Weeks-Team |
| Prize | Why vibeDeploy Fits |
|---|---|
| 1st Place ($5,000) | Deep Gradient AI integration (8 features), full pipeline |
| Best AI Agent Persona | 6 distinct agents with personalities debating live |
| Best Program for the People | Democratizes app development for non-coders |
| Great Whale Prize | Largest scope: idea → live deployed app |
MIT - See LICENSE for details.
Built with DigitalOcean Gradient AI for the DO Gradient AI Hackathon 2026