Conversation Manager Reference

Field values for the live conversation test page.

Test Page

Conversation

AI Profile
ai_default, ai_fast, ai_local_balanced, ai_deepseek, or ai_high.
Language
auto, en, es, or pt.
Locale
Use en-US, en-GB, es-MX, es-ES, or pt-BR.

Voices

Use a single Kokoro voice id or a weighted mix.

af_heart
em_alex
ef_dora:0.7;em_alex:0.3
English US
af_heart, am_michael, af_sky, am_puck
English GB
bf_emma, bm_george, bf_lily, bm_fable
Spanish
ef_dora, em_alex, em_santa
Brazilian Portuguese
pf_dora, pm_alex, pm_santa
Other
ff_siwis, hf_alpha, hm_omega, hf_beta, hm_psi, if_sara, im_nicola

Timing

Speed
Speech speed multiplier. Typical range is 0.8 to 1.2.
Pause ms
Silence between streamed sentence chunks. Default is 50.

History Export

Ephemeral conv-mgr instances expose one authenticated export endpoint for client-side archival.

GET /conversations/{conversation_id}/export
GET /conversations/{conversation_id}/export?include_events=false
Includes
Conversation metadata, full transcript, all messages, all turns, and per-turn metrics.
Events
Included by default. Use include_events=false for a smaller transcript-focused payload.

Text Chat

Typed turns use the same session context, recent history, AI profile, metrics, and export path as voice turns. A session can mix typed and voice turns.

POST /conversations/{conversation_id}/turns
{
  "input": { "type": "text", "text": "What should I check first?" },
  "output": { "audio": false },
  "language": "en",
  "locale": "en-US",
  "voice": "af_heart"
}
Default
output.audio is false for this HTTP endpoint.
Spoken Output
Set output.audio to true to receive response audio artifacts. The test page's Speak response checkbox sends this flag.

AI Context

Use the page's Context Markdown field for caller-owned instructions and background. API clients should send the same content as a context string.

POST /conversations
{ "context": "## Role\nProvide concise safety guidance.\n\n## Constraints\n- Stay near the exit." }

PUT /conversations/{conversation_id}/context
{ "context": "## Background\nThe loading dock is beside pump four." }
Text
Text and Markdown are supplied as caller-authored context. Other JSON values are rendered generically without an application schema.
Turn Override
Text and audio turn payloads may also include context. That context applies to the current AI turn without replacing stored session context.
Export
The decoded session context is included under conversation.context.

Creature Effect

Set Effect to creature. Leave it as none to disable.

Aliases
creature, monster, beast, demon
Intensity
0.0 to 1.0. Default is 0.75.
Pitch
Semitones from -12 to 0. More negative is deeper.
Advanced
The server also accepts echo_delay_ms, echo_decay, growl, and grit.

Examples

Language: es
Locale: es-MX
Voice / Mix: ef_dora:0.65;em_alex:0.35
Speed: 1.05
Pause ms: 220
Effect: creature
Intensity: 0.25
Pitch: -3
Language: pt
Locale: pt-BR
Voice / Mix: pf_dora:0.65;pm_alex:0.35
Speed: 1.05
Pause ms: 180