"I built this because I have 47 unread articles in my bookmarks right now that I'm not going back to, newsletter subs I don't read, and I wanted to create a better way to consume content while on the go."
Transform any article, URL, or topic into personalized podcast-style audio content in under 2 minutes. Perfect for commuting, working out, or multitasking.
- Professional Anchor - CNN-style news delivery
- Dynamic Duo - Two hosts discussing topics conversationally
- Pirate News - Ahoy! News with a nautical twist
- Late Night Host - Comedy-infused commentary
- Valley Girl - Trendy, casual delivery
- Gen Z - Internet-native, meme-aware style
- Professor - Academic, educational approach
- Coffee Chat - Intimate, friend-to-friend conversation
- Casual - Relaxed, everyday tone
- Topic to audio in under 2 minutes
- Live article fetching with intelligent source prioritization
- Multi-source content synthesis and summarization
- Smart Content Curation - Firecrawl + domain filtering
- Advanced Summarization - GPT-4 powered content analysis
- Multi-Style Script Generation - Adapts tone for each voice personality
- Conversation Generation - Natural back-and-forth dialogue for duo modes
- Browse audio content created by other users
- Share your generated podcasts
- Email subscriptions for daily personalized briefings
- Responsive design across all devices
This project showcases a sophisticated multi-agent workflow using the Inkeep Agents SDK:
graph LR
A[π° Ingest Agent] --> B[βοΈ Summarizer Agent]
B --> C[π€ TTS Agent]
C --> D[πΎ Catalog Agent]
A --> E[Firecrawl Tool]
B --> F[OpenAI Tool]
C --> G[ElevenLabs Tool]
D --> H[Convex Tool]
-
π Ingest Agent (
src/agents/agents.ts:5-11)- Uses
firecrawlToolfor intelligent web scraping - Domain filtering and content prioritization
- Batch processing with fallback mechanisms
- Uses
-
π Summarizer Agent (
src/agents/agents.ts:14-20)- Uses
summarizeTool(OpenAI GPT-4) - Context-aware content analysis
- Multi-style script generation for different personalities
- Uses
-
π΅ TTS Agent (
src/agents/agents.ts:23-29)- Uses
elevenLabsToolfor voice synthesis - Supports conversation mode with multiple speakers
- Professional-quality audio generation
- Uses
-
ποΈ Catalog Agent (
src/agents/agents.ts:32-38)- Uses
convexStoreToolfor storage and metadata - Social features and discovery
- Analytics and user preferences
- Uses
export const podcastGraph = agentGraph({
id: "tldr-podcast-graph",
name: "TLDR Podcast Graph",
description: "Ingest β Summarize β TTS β Store",
defaultAgent: ingestAgent,
agents: () => [ingestAgent, summarizerAgent, ttsAgent, catalogAgent],
});- AI-generated natural dialogue between multiple speakers
- Speaker personality injection
- Sequential voice synthesis with different ElevenLabs models
- Domain prioritization (Reuters, BBC, TechCrunch over Reddit, Twitter)
- Batch scraping with intelligent fallbacks
- Content cleaning and filtering
- OpenAI GPT-4 - Advanced summarization, script generation, dialogue creation
- Inkeep Agents SDK - Multi-agent orchestration, tool composition, analytics
- ElevenLabs - Premium voice synthesis with 9+ personalities
- Next.js 15 - App Router, API routes, server components
- Convex - Real-time database, file storage, mutations
- TypeScript - Full type safety across the entire stack
- Firecrawl - Intelligent web scraping with domain filtering
- Resend - Professional email delivery and scheduling
- Autumn.js - Subscription management and payments
- Vercel - Edge deployment and serverless functions
- Tailwind CSS - Modern design system
npm installCreate a .env.local file:
# Core APIs
OPENAI_API_KEY=your_openai_key
ELEVENLABS_API_KEY=your_elevenlabs_key
FIRECRAWL_API_KEY=your_firecrawl_key
# Inkeep Analytics (optional)
INKEEP_API_KEY=your_inkeep_key
# Database & Storage
CONVEX_DEPLOYMENT=your_convex_deployment
NEXT_PUBLIC_CONVEX_URL=your_convex_url
# Email & Subscriptions
RESEND_API_KEY=your_resend_key
AUTUMN_API_KEY=your_autumn_keynpm run devOpen http://localhost:3000 to see the result.
// Input: Any article URL
const url = "https://techcrunch.com/2024/01/15/ai-breakthrough";
// Output: Personalized audio in your chosen style
// Duration: ~2 minutes processing time// Input: Any topic
const topic = "latest developments in quantum computing";
// Output: Multi-source synthesis + audio generation
// Sources: Automatically curated from top news outlets// Two-host dialogue with distinct personalities
// Alex (analytical) + Sam (curious) discussing the topic
// Natural interruptions, questions, and reactions- Multi-Agent Architecture - Proper agent separation and coordination
- Tool Composition - Custom tools integrated seamlessly
- Analytics Integration - Real-time metrics via Inkeep Analytics MCP
- Credential Management - Secure API handling
- Advanced Prompt Engineering - Context-specific prompts for different personalities
- Content Intelligence - Smart summarization with quality filtering
- Dynamic Style Adaptation - AI adapts writing for 9 different voice personalities
- Follow-up Processing - Real-time responses to user queries
- Error Handling - Graceful degradation and retry mechanisms
- Performance - Optimized for speed with concurrent processing
- Scalability - Built for real users with proper database design
- User Experience - Polished UI/UX with real-time feedback
- Busy Professionals - Stay informed during commutes
- Fitness Enthusiasts - Learn while working out
- Multitaskers - Consume content while doing other activities
- Accessibility - Audio-first approach for various needs
- Content Creators - Generate audio content from any source
This project demonstrates advanced AI agent orchestration and could be extended with:
- Additional voice personalities
- More content sources
- Enhanced analytics dashboard
- Mobile app development
- API for third-party integrations
MIT License - Feel free to use this as inspiration for your own AI agent projects!
Built with π using Inkeep Agents SDK and OpenAI - Showcasing the future of AI-powered content transformation.