Agentic podcast generation, end-to-end

Turn raw ideas into a polished, listenable episode.

Resound orchestrates an agentic pipeline to plan a show, write a script, and (optionally) generate speech with OpenAI TTS — then saves everything to output/.

Generate an episode Browse episodes

Agentic scripting

The system plans structure, generates sections, and iterates to produce a coherent script — not just a single “dump” response.

Configurable voices

Speakers are defined in config/config_openai.json (voice model + voice id).

Artifacts included

Every run saves content, section JSON, and final audio (when enabled) inside output/<episode>/.

How it works

1) Input
You provide an episode name + text (or a topic). Resound treats it as source material.
2) Agentic planning
Agents analyze the content, choose a structure, and produce a multi-segment script.
3) Script → Audio (optional)
With TTS enabled, each segment is voiced using OpenAI and stitched into a single episode file.
4) Output
Results land in output/<episode>/ for inspection, iteration, and playback.

Why this is different

Instead of a one-shot chat response, Resound is an orchestration layer: it enforces a show format, manages shared context across steps, and persists artifacts so you can debug, refine, and regenerate.

Fast mode

Generate script only (cheap + quick) and review before spending on TTS.

Full mode

Generate audio and get a playable episode you can share immediately.

Quick setup

Install Python deps: pip install -e .
Set OPENAI_API_KEY (shell or .env).
Start this UI: cd web && npm install && npm run dev

Note: this demo runs the CLI on the server, so it’s intended for local use.