Epist.ai Audio RAG Platform

[!TIP] "Read" your audio. Transform podcasts, meetings, and voice notes into a queryable Second Brain with RAG.

🔗 Quick Links

• Documentation Site
• Quick Start Guide
• API Reference
• MCP Guide
• Contributing

🚀 New: Model Context Protocol (MCP)

Connect Claude Desktop directly to your knowledge base using our official MCP Server:

pip install epist-mcp-server

Read the MCP Guide →

Introduction

Epist.ai is a next-generation system that treats audio as a first-class citizen, enabling seamless ingestion, understanding, and retrieval of information from meetings, voice notes, and ambient conversations.

It provides a production-ready API for audio ingestion, transcription, and semantic search, built on modern RAG (Retrieval-Augmented Generation) principles.

🧠 Core Concepts

The Audio RAG Pipeline

1. Ingestion: Audio/Video files are streamed to GCS.
2. Transcription: Fireworks AI (Whisper V3 Turbo) generates high-fidelity transcripts with speaker diarization.
3. Semantic Chunking: Transcripts are intelligently split to preserve context.
4. Vectorization: Text chunks are converted to embeddings using OpenAI's text-embedding-3-small.
5. Hybrid Retrieval: Search combines BM25 keyword matching with Vector similarity for maximum precision.

Platform vs. Tenant

• Platform (This Repo): The core engine providing the API, storage, and processing logic.
• Tenant (External): Applications like podcast-rag-app that consume the API to provide specific user experiences.

Features

• Audio Upload: Stream large audio files directly to Google Cloud Storage.
• RSS Ingestion: Ingest entire podcast feeds automatically with metadata extraction.
• Automatic Transcription: Background transcription with speaker diarization.
• Hybrid Search: Combine semantic meaning (Vector) and exact keywords (Full-Text) using Reciprocal Rank Fusion (RRF).
• Multi-Modal Support: Support for .mp4, .mov, .mp3, .wav.
• 📊 Observability: "Glass Box" tracing system to visualize every pipeline step.
• 💬 Interactive Chat: Chat with your audio content using citations that link directly to timestamps.
• MCP Integration: Connect your audio knowledge base directly to Claude.

Tech Stack

• Backend: FastAPI (Python 3.11+)
• Database: PostgreSQL 15 + pgvector
• ASR: Fireworks AI (Whisper V3 Turbo)
• Embeddings: OpenAI text-embedding-3-small
• RAG Framework: LangChain
• Infrastructure: GCP (Cloud Run, Cloud SQL, Cloud Storage, Cloud Tasks)
• Frontend: Next.js 15 (React)

Quick Start

This project uses uv for dependency management.

# Install dependencies
uv sync

# Run development server
uv run fastapi dev src/main.py

Frontend Development

cd apps/web
npm install
npm run dev

Project Structure

Changelog

v1.2.0 (Jan 05, 2026)

• RSS Ingestion: New generic platform feature to ingest entire podcast feeds.
• Enhanced Metadata: Automated extraction of author, description, and images.

v1.1.0 (Dec 27, 2025)

• Stability & Observability: Integrated Sentry and migrated to Cloud Scheduler.
• Security & Performance: Added TrustedHost, GZip, and production security headers.
• Tiered Sync: Automated feed synchronization with tier-based refresh intervals.

v0.1.0 (Dec 08, 2025)

• Core Platform: Initial release with Audio RAG and Vector Search.

Contributing

We welcome contributions! Please see our Contributing Guide for more information.

License

This project is open source under the MIT License.

Support

If you have questions or need assistance, please contact us at admin@epist.ai.

Epist.ai - Transforming audio into actionable knowledge.