Keypilot

AI-powered SEO article generator that transforms keyword CSVs into publication-ready articles through a 5-step LLM pipeline with real-time streaming.

Live Demo · Report Bug · Request Feature

---

Overview

Keypilot takes a CSV of keywords (up to 500), runs them through five sequential LLM steps, and streams each step's output in real time — from raw keywords to a polished, SEO-optimized article ready for publishing.

Each step uses a purpose-selected model and temperature tuned for the task:

Step	Task	Model	Provider
1. Cluster	Group keywords into 3–7 topical clusters	Llama 3.1 8B	Cerebras
2. Outline	Generate a structured article outline	Llama 3.3 70B	Groq
3. Draft	Write the full article from the outline	Llama 3.3 70B	Groq
4. Refine	Edit for flow, readability, and consistency	GPT-OSS 120B	Cerebras
5. Polish	Final grammar, formatting, and keyword pass	GPT-OSS 120B	OpenRouter

Features

• Multi-provider LLM pipeline — Different models optimized for each step
• Real-time streaming — SSE-based token streaming with live UI updates
• Drag-and-drop CSV upload — With automatic deduplication and preview
• Configurable output — Article type, length (500–4000 words), tone, style, and audience
• Step-by-step visibility — Auto-advancing tabs showing each pipeline stage
• One-click export — Copy to clipboard or download as markdown
• Dark mode — Full dark mode support
• Cancellation — Graceful pipeline abort at any point

Getting Started

Prerequisites

• Node.js 18+
• API keys for Groq, Cerebras, and OpenRouter

Setup

git clone https://github.com/pradhankukiran/keypilot.git
cd keypilot
npm install

Create a .env.local file:

GROQ_API_KEY=your_groq_key
CEREBRAS_API_KEY=your_cerebras_key
OPENROUTER_API_KEY=your_openrouter_key

Run

npm run dev

Open http://localhost:3000.

Configuration Options

Option	Description	Examples
Article Type	Format of the generated article	How-to guide, listicle, comparison, ultimate guide, product roundup
Target Length	Approximate word count	500 – 4,000 words
Tone	Voice of the article	Professional, casual, authoritative
Style	Writing approach	Data-driven, narrative, conversational
Target Audience	Intended readers	Small business owners, beginner developers

Tech Stack

Layer	Technology
Framework	Next.js 16 (App Router, Turbopack)
UI	React 19, Tailwind CSS v4
Language	TypeScript 5 (strict mode)
AI	Vercel AI SDK v6 (streamText)
LLM Providers	Cerebras, Groq, OpenRouter
Markdown	react-markdown + remark-gfm
CSV Parsing	PapaParse
Streaming	Server-Sent Events (SSE)
Deployment	Vercel (serverless)

Project Structure

src/
├── app/
│   ├── page.tsx                 # Main UI
│   ├── layout.tsx               # Root layout
│   └── api/generate/route.ts    # SSE streaming endpoint
├── components/
│   ├── PipelineForm.tsx         # CSV upload + config form
│   ├── StepTabs.tsx             # Pipeline progress tabs
│   ├── ArticleOutput.tsx        # Final article display
│   ├── MarkdownRenderer.tsx     # Prose markdown renderer
│   └── CsvPreview.tsx           # Keyword preview chips
├── hooks/
│   └── usePipeline.ts           # Pipeline state management
└── lib/
    ├── pipeline.ts              # Pipeline orchestration
    ├── prompts.ts               # System prompts per step
    ├── providers.ts             # LLM provider config
    ├── stream-utils.ts          # SSE utilities
    ├── csv-parser.ts            # CSV parsing + dedup
    └── steps/                   # Individual step implementations
        ├── step1-cluster.ts
        ├── step2-outline.ts
        ├── step3-draft.ts
        ├── step4-refine.ts
        └── step5-polish.ts

License

This project is open source and available under the MIT License.