vibeDeploy

One click starts discovery. A live Kanban ranks ideas. You choose a GO card to build. vibeDeploy ships the app to DigitalOcean.
Zero-Prompt discovery, AI evaluation, type-safe code generation, build validation, and deployment in one flow.

Demo video link: add the final public YouTube or Vimeo URL before submission.

---

Overview

vibeDeploy is a dual-runtime application that turns an idea into a live deployed app on DigitalOcean.

• web/ is the Next.js 16.1.7 frontend for Zero-Prompt, evaluation, brainstorm, and operational views.
• agent/ is the Python 3.12.4 backend that runs the Gradient ADK entrypoint, FastAPI gateway, LangGraph pipelines, and deployment orchestration.
• The flagship flow is Zero-Prompt Start: one click starts discovery, a live Kanban fills with GO and NO-GO ideas, and the user picks a GO card to build.

Main Flow

One click starts Zero-Prompt discovery
  -> YouTube discovery selects candidate videos
  -> transcript extraction pulls source material
  -> Gemini extracts the app idea
  -> paper search adds academic context
  -> competitive analysis checks saturation
  -> deterministic scoring marks GO / NO-GO
  -> live Kanban shows the ranked cards
  -> user clicks Build on a GO card
  -> contract-first code generation runs
  -> validation + deploy produce a live app URL

Product Modes

Mode	Input	Output
Zero-Prompt Start	One click	Live Kanban of GO / NO-GO ideas, then on-demand build
Evaluate	One sentence or YouTube URL	Vibe Council score, verdict, and optional build
Brainstorm	One sentence	Structured idea expansion and product brief

Why The Flow Works

• Contract-first generation keeps frontend and backend aligned through OpenAPI-derived types and models.
• Layered generation separates deterministic scaffolds from LLM-written business logic.
• Validation before deploy includes syntax, imports, build checks, and contract verification.
• Live streaming UI turns the pipeline into a visible product experience instead of a black box.

Public Live Apps

These public links returned HTTP 200 during the latest submission audit.

App	What it shows	URL
NutriPlan	AI meal planning MVP	https://nutriplan-meal-planning-818722-dei83.ondigitalocean.app
FlavorSwap Lite	Recipe and ingredient swap workflow	https://flavorswap-lite-637431-dl43t.ondigitalocean.app
GardenBreak	Lightweight habit and garden routine tracking	https://gardenbreak-377462-h9bmg.ondigitalocean.app
Creator Batch Studio	Creator planning and batching workspace	https://creator-batch-studio-640303-segyb.ondigitalocean.app

DigitalOcean Capabilities Used

The current submission material is aligned around 13 DigitalOcean capabilities across Gradient AI and the broader deployment stack.

#	Capability	Current use in vibeDeploy	Code
1	Gradient ADK	Agent entrypoint and deployment	agent/main.py
2	Knowledge Bases	Deployment and framework retrieval	agent/tools/knowledge_base.py
3	Evaluations	Generated app quality checks	agent/evaluations/
4	Guardrails	Input moderation and prompt safety	agent/guardrails.py
5	Tracing	Tool and LLM trace instrumentation	agent/llm.py, agent/tools/*.py
6	Multi-Agent Routing	Council and Zero-Prompt orchestration	agent/gradient/router.py
7	A2A Handoff	GO card handoff into build pipeline	agent/gradient/a2a.py
8	Serverless Inference	Model execution via the provider registry	agent/providers/registry.py
9	App Platform	Hosts vibeDeploy and generated apps	.do/app.yaml
10	Spaces	Archives and build artifacts	agent/tools/spaces.py
11	Image Generation	Logos and OG image generation	agent/tools/image_gen_do.py
12	Agent Versioning	Version-aware pipeline iteration	agent/gradient/versioning.py
13	MCP Integration	Platform API access	agent/gradient/mcp_client.py

Current Model Plan

These are the current default role mappings from agent/llm.py and the canonical registry in agent/providers/registry.py.

Role	Model	Provider
Zero-Prompt discovery	gemini-3.1-flash-lite-preview	Google
Zero-Prompt brainstorm	gemini-3.1-flash-lite-preview	Google
Council analysis	claude-sonnet-4-6	Anthropic
Cross-exam	claude-sonnet-4-6	Anthropic
Strategist / docs	gpt-5.4	OpenAI
Code generation	gpt-5.3-codex	OpenAI
CI repair	gpt-5.2	OpenAI
UI design	gemini-3.1-pro-preview	Google
Image generation	fal-ai/flux/schnell	DO Inference

Current Stack Versions

Layer	Current version / service
Frontend	Next.js 16.1.7, React 19.2.4, Tailwind CSS 4, Framer Motion 12.37.0
Backend	Python 3.12.4, FastAPI 0.115+, Pydantic 2.9+, uvicorn 0.34+
Agent runtime	Gradient ADK 0.2.11+, LangGraph 1.0+, LangChain Core 1.2.11+
Database	Managed PostgreSQL
Storage	DigitalOcean Spaces
Build validation	Docker SDK

Local Development

# agent runtime
cd agent
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python run_server.py

# web runtime
cd web
npm ci
NEXT_PUBLIC_AGENT_URL=http://localhost:8080 npm run dev

Verify Locally

# agent
cd agent
ruff check .
ruff format --check .
pytest tests/ -v --tb=short

# web
cd web
npx eslint .
npm test
NEXT_PUBLIC_AGENT_URL=http://localhost:8080 npm run build

For the latest verification evidence, see docs/E2E_VERIFICATION.md.

Deployment

• Live app: https://vibedeploy-7tgzk.ondigitalocean.app
• Deployment inventory: docs/DEPLOYMENT_STATUS.md
• App Platform spec: .do/app.yaml
• Agent deployment script: agent/scripts/deploy.sh

# deploy the agent
cd agent
gradient agent deploy

# deploy the app platform stack
cd ..
doctl apps create --spec .do/app.yaml

Project Structure

vibeDeploy/
|- agent/               # Gradient ADK entrypoint, FastAPI gateway, LangGraph pipelines
|- web/                 # Next.js app router frontend and Zero-Prompt UI
|- .do/app.yaml         # App Platform spec for api + web
|- docs/                # submission text, verification notes, reference docs
`- screenshots/         # demo and README assets

License

MIT — see LICENSE.

_{Maintained by Two-Weeks-Team}