Origami (Created with v0)

Origami is a hackathon-first "Doc-to-App Engine" created with v0 and built with Next.js and the Vercel AI SDK. It ingests pasted text, text files, text-based PDFs, or a GitHub repository URL, then turns the source into an overview-first interactive workspace:

• a calculator for formulas, fees, or penalty rules
• a multi-doc repo dashboard with an interactive architecture graph and node inspector
• a document insight view for individual markdown files
• a deterministic package dashboard for package.json and key manifests
• a scenario simulator for branching playbooks and if/then manuals
• a PDF breakdown view with extracted sections, notable metrics, and source-aware follow-up actions
• a v0 MVP Builder (powered by v0) that generates an in-app single-page MVP route with copyable React/Tailwind code
• an MCP-backed source Q&A box that can inspect the active PDF, repo, or text before answering
• a customization prompt bar on generated MVP pages for revising the design in place

Stack

• Front end developed by v0
• Tailwind CSS
• Vercel AI SDK with streamText
• OpenAI via @ai-sdk/openai
• Model Context Protocol via @ai-sdk/mcp and @modelcontextprotocol/sdk
• optional v0 Model API access via https://api.v0.dev/v1/chat/completions
• React Flow
• Recharts
• Zod
• unpdf

Getting started

1. Install dependencies:

npm install

2. Add your environment variable:

cp .env.example .env.local

3. Start the app:

npm run dev

4. Open http://localhost:3000

Environment

OPENAI_API_KEY=your_openai_api_key
OPENAI_BASE_URL=
OPENAI_MODEL=gpt-5.5
V0_API_KEY=
GITHUB_TOKEN=

OPENAI_BASE_URL is optional for the default OpenAI API. Set it when you want to point Origami at an OpenAI-compatible endpoint such as Microsoft Foundry.

Example:

OPENAI_API_KEY=your_foundry_key
OPENAI_BASE_URL=https://your-endpoint.example.com/openai/v1
OPENAI_MODEL=your-model-or-deployment-name

GITHUB_TOKEN is optional, but recommended if you plan to scan multiple public repositories and want to avoid stricter unauthenticated GitHub API rate limits.

V0_API_KEY is optional. When present, Origami will try the v0 model first for MVP site generation and fall back to the configured OpenAI-compatible model if needed.

Useful commands

npm run dev
npm run lint
npm run typecheck
npm run build

Notes

• GitHub repo mode now scans recursive .md / .mdx docs plus key manifests such as package.json, pnpm-workspace.yaml, turbo.json, and tsconfig.json.
• Repo dashboards show an Overview tab first, then one tab per included markdown or manifest file.
• Architecture nodes include evidence-backed explanations, related nodes, and file shortcuts in a persistent inspector.
• PDF mode supports text-based PDFs only. Scanned or image-only PDFs are intentionally out of scope for this MVP.
• Source Q&A now uses a request-local HTTP MCP server so the model can list, search, and read grounded source snippets on demand.
• Generated MVP pages open inside Origami at /workspace/mvp/[artifactId] and are stored in browser-local storage so they survive refreshes and same-browser tabs.
• The AI only returns typed tool payloads. The UI stays deterministic and React-rendered.