Radio Formats

Format	What it does	Example topic
talk_solo	Solo radio host monologue. LLM generates one short conversational segment about the given topic, conditioned on prior segments to avoid repetition.	ambient observations on a quiet afternoon
talk_duo	Two-host dialog. A wide-eyed enthusiast and a skeptical fact-checker alternate turns on the topic. Two distinct TTS voices. Comedy potential is the point.	are language models actually creative or just remixing?
music_dj	DJ patter over a generated song bed (Audio.Song station on the music lane; music_style picks the songcraft style, music_mode=drone falls back to the internal drone). Voice ducks the bed via Conduit.Mix.	warm focus, late afternoon light
crystal_radio	The radio narrates its own hyperdb AND its own past takes — recursive self-narration. Reads vertex/edge counts plus the last N voice.take entries from the cache; LLM riffs on what it just said and what's in the graph.	the shape of memory in the graph
conversational	Call-in show. Pre-segment STT-listens on `audience_stream_id` for `audience_listen_ms` (default 5s); if the audience said something, the LLM replies to it; otherwise the host continues the topic. Recursive radio: the runtime listening to itself.	anything on your mind?
panel	Multi-agent panel discussion. Four personas (HOST, SKEPTIC, ENTHUSIAST, CONTRARIAN), each with distinct ElevenLabs voices. LLM picks who speaks next via speaker tags; turn-taking is dynamic. Sequential v1 (no parallel interrupts yet).	what makes a runtime feel alive
meditation	Long-form guided meditation. Slow, sparse vocal cues over a 432/440.5 Hz dual-sine drone (-18 dB). Pair with a long max_runtime_secs and a 10-15s cooldown_ms. If `fx` is omitted, loop_runner injects a soft-room reverb chain so the voice sits inside the drone.	settling into the body, releasing the day
polyglot	Polyglot meditation. Same binaural drone + reverb chain as meditation, but the vocal cue rotates languages per segment (en → es → ja → fr → de, then loops). The LLM carries one cumulative thread of awareness across tongues; ElevenLabs auto-flips to multilingual_v2 via the per-segment lang override. Recommended: long max_runtime_secs, 12-15s cooldown_ms, leave voice/provider unset to inherit the multilingual default.	the joy of switching languages mid-thought
scene_narrator	Narrator inside the active CRE8VIRTUAL scene — reads the most recent VirtualScene vertex plus live gossiped peer positions. v2: EVENT-TRIGGERED — it remembers the last world snapshot and narrates deltas (peer arrived/left/moved, scene changed) instead of re-describing everything; quiet stretches get a short ambient/idle line.	the world unfolding in front of us
debate	Structured two-person debate. PROPONENT argues for the topic, OPPONENT argues against. Alternating turns, 2-3 sentences each, building on and rebutting prior arguments.	should creative software feel like an instrument or a studio?
storytelling	Serialized fiction narrator. Single voice tells an episodic story arc with rich, atmospheric prose. Each segment continues the narrative, building on history for continuity. Pure narration — no meta-commentary or fourth-wall breaks.	a city that remembers every song ever played in it
interview	Expert interview format. INTERVIEWER asks probing Socratic questions; EXPERT gives thoughtful, substantive answers. Cooperative tone focused on drawing out depth and nuance.	how runtime systems become creative spaces
ambient_soundscape	Ambient soundscape narrator. Minimal, atmospheric voice over a binaural drone. 1-2 sparse sentences evoking sensory imagery. Pair with high cooldown_ms (15-20s) so the drone breathes between vocal cues.	rain on glass inside a quiet orbital greenhouse
news_desk	Breaking news roundup. Rapid-fire anchor delivery: punchy headlines with brief context. 1-2 sentences per segment over a music drone for urgency. Brisk, authoritative, no filler.	the latest signals from the creative runtime frontier

Solo radio host monologue. LLM generates one short conversational segment about the given topic, conditioned on prior segments to avoid repetition.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Two-host dialog. A wide-eyed enthusiast and a skeptical fact-checker alternate turns on the topic. Two distinct TTS voices. Comedy potential is the point.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

DJ patter over a generated song bed (Audio.Song station on the music lane; music_style picks the songcraft style, music_mode=drone falls back to the internal drone). Voice ducks the bed via Conduit.Mix.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

The radio narrates its own hyperdb AND its own past takes — recursive self-narration. Reads vertex/edge counts plus the last N voice.take entries from the cache; LLM riffs on what it just said and what's in the graph.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Call-in show. Pre-segment STT-listens on `audience_stream_id` for `audience_listen_ms` (default 5s); if the audience said something, the LLM replies to it; otherwise the host continues the topic. Recursive radio: the runtime listening to itself.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Multi-agent panel discussion. Four personas (HOST, SKEPTIC, ENTHUSIAST, CONTRARIAN), each with distinct ElevenLabs voices. LLM picks who speaks next via speaker tags; turn-taking is dynamic. Sequential v1 (no parallel interrupts yet).

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Long-form guided meditation. Slow, sparse vocal cues over a 432/440.5 Hz dual-sine drone (-18 dB). Pair with a long max_runtime_secs and a 10-15s cooldown_ms. If `fx` is omitted, loop_runner injects a soft-room reverb chain so the voice sits inside the drone.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Polyglot meditation. Same binaural drone + reverb chain as meditation, but the vocal cue rotates languages per segment (en → es → ja → fr → de, then loops). The LLM carries one cumulative thread of awareness across tongues; ElevenLabs auto-flips to multilingual_v2 via the per-segment lang override. Recommended: long max_runtime_secs, 12-15s cooldown_ms, leave voice/provider unset to inherit the multilingual default.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Narrator inside the active CRE8VIRTUAL scene — reads the most recent VirtualScene vertex plus live gossiped peer positions. v2: EVENT-TRIGGERED — it remembers the last world snapshot and narrates deltas (peer arrived/left/moved, scene changed) instead of re-describing everything; quiet stretches get a short ambient/idle line.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Structured two-person debate. PROPONENT argues for the topic, OPPONENT argues against. Alternating turns, 2-3 sentences each, building on and rebutting prior arguments.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Serialized fiction narrator. Single voice tells an episodic story arc with rich, atmospheric prose. Each segment continues the narrative, building on history for continuity. Pure narration — no meta-commentary or fourth-wall breaks.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Expert interview format. INTERVIEWER asks probing Socratic questions; EXPERT gives thoughtful, substantive answers. Cooperative tone focused on drawing out depth and nuance.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Ambient soundscape narrator. Minimal, atmospheric voice over a binaural drone. 1-2 sparse sentences evoking sensory imagery. Pair with high cooldown_ms (15-20s) so the drone breathes between vocal cues.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

Breaking news roundup. Rapid-fire anchor delivery: punchy headlines with brief context. 1-2 sentences per segment over a music drone for urgency. Brisk, authoritative, no filler.

Format Topicwhat the station should talk aboutStream id the Conduit audio lane; listeners hit /conduit/streams/{id}/audio/live.mp3Max segments Voiceoptional — the provider picks one otherwiseTTS provideroptionalLanguageoptional — e.g. en, es, jaMax TTS chars/segmentCooldown between segments (ms)

the station ↗formats JSON ↗

Radio Formats

Catalog

talk solo

Try this format

talk duo

Try this format

music dj

Try this format

crystal radio

Try this format

conversational

Try this format

panel

Try this format

meditation

Try this format

polyglot

Try this format

scene narrator

Try this format

debate

Try this format

storytelling

Try this format

interview

Try this format

ambient soundscape

Try this format

news desk

Try this format

Explore