keltokhy/writ-fm

What it solves

WRIT-FM is an autonomous, 24/7 AI-powered radio station. It automates the entire broadcasting process, from generating AI music and writing hosted talk breaks to rendering those scripts with text-to-speech (TTS) and managing the live stream via Icecast.

How it works

The system uses a combination of AI agents and streaming tools to maintain a continuous broadcast:

• Content Generation: AI agents (using Claude or Codex) write scripts for various segment types (deep dives, news analysis, interviews) based on a schedule. AI music is generated via ACE-Step/music-gen.server.
• Audio Rendering: Scripts are converted to audio using pluggable TTS engines like Kokoro, Chatterbox, or Voxtral.
• Streaming Stack: A custom CLI (writ) manages Icecast, ezstream, and API servers to feed the audio into a continuous stream.
• Autonomous Operation: An operator daemon runs a maintenance loop every 15 minutes to stock music, generate talk breaks, process listener messages, and maintain editorial continuity using a station ledger (memory).

Who it’s for

This is for developers and AI enthusiasts interested in autonomous agents, generative audio, and the intersection of AI-driven content creation and traditional radio broadcasting.

Highlights

• Multi-Station Support: Ability to run multiple isolated station instances (e.g., WRIT-FM, KLOD-FM) simultaneously.
• Dynamic Personalities: Five distinct AI hosts with unique voices, philosophies, and focus areas.
• Editorial Memory: A station ledger and "intent cards" allow the AI operator to maintain continuity across different runs.
• Interactive Elements: A listener daemon that polls for messages and generates spoken on-air responses.
• Flexible TTS: Support for multiple TTS engines including voice cloning (Chatterbox) and multilingual support (Voxtral).