Built for people who actually listen.

PodPast.ai is a semantic search and transcription platform that ingests YouTube videos and podcast RSS feeds, generates speaker-attributed transcripts, and exposes them to Claude (and other LLMs) via a Model Context Protocol integration. The product exists to solve one specific problem: knowledge workers who consume 10–15 hours of long-form audio per week cannot recall, locate, or cite what they heard.

The target user is what Edison Research calls the "Super Listener" — a person averaging 11.2 hours per week across podcasts, YouTube lectures, and long-form interviews. This person is a researcher who can't find the economist who made an argument about capital formation in a podcast three months ago. A journalist who needs to verify exactly what a source said in a 2023 interview. A founder who knows they heard something specific about go-to-market strategy but can't remember which pod. PodPast.ai gives these users a persistent, queryable record of everything they've ever listened to — with citations and timestamps that point back to the source.

Technical architecture

The backend is a FastAPI application deployed on Railway at api.podpast.ai. Transcript storage and vector search run on Supabase PostgreSQL with the pgvector extension. Audio-only episodes that lack a pre-existing caption track are processed by Deepgram nova-2, which streams the audio URL directly without downloading or storing any audio files. Transcript text is chunked into overlapping 150–200 word segments and embedded using OpenAI's text-embedding-3-small model at 256 dimensions. Vector similarity search uses pgvector's approximate nearest neighbour index (IVFFlat) with a re-ranking pass that weights keyword overlap and recency. The MCP server is a TypeScript package published on npm (@podpast/mcp-server) that communicates with Claude Desktop via stdio transport. The marketing site and user-facing web interface are built on Next.js 15 App Router and deployed on Vercel.

Entity facts

Founded

2025

Headquarters

United States

API endpoint

api.podpast.ai (Railway)

Marketing site

podpast.ai (Vercel)

Transcription

Deepgram nova-2

Vector DB

Supabase + pgvector

LLM integration

Claude Sonnet via MCP

Embedding model

text-embedding-3-small (256d)

Contact

General: hello@podpast.ai
Support: support@podpast.ai
Privacy: privacy@podpast.ai
DMCA: dmca@podpast.ai