Fix/elevenlabs multi stream input#25

Open

Jacob-Lasky wants to merge 25 commits intomainfrom

fix/elevenlabs-multi-stream-input

Commits on Feb 26, 2026

feat: Phase 3 - demo JSON configs (Dubai RE, Hey Saga, Deepgram, Hey Manny stub)
Jake Lasky
committed
docs: mark Phase 3 (Demo JSON Configs) as Complete in STATE.md
Jake Lasky
committed
docs(01-01): complete Phase 1 Backend JSON Config System plan

Jake Lasky
and
claude
committed
feat: Phase 2 - frontend redesign with Deepgram design system
Jake Lasky
committed
feat(04-integration): wire config_id to VoiceAgent and update frontend session start
Jake Lasky
committed
feat: Phase 4 - integration wiring config selection to VoiceAgent
Jake Lasky
committed

Commits on Feb 27, 2026

fix: create AudioContext during user gesture to unblock Chrome autoplay

Jake Lasky
and
claude
committed
fix: force WebSocket transport to fix binary audio on Fly.io

Jake Lasky
and
claude
committed
fix: deepgram auto-selected first, builder language/voice model sync

Jake Lasky
and
claude
committed
fix(partial): agentName substitution in system prompt + loop policy fix

Jake Lasky
and
claude
committed
Fix stop/restart: close WS and track thread for clean restarts

Jake Lasky
and
claude
committed
Use {{agentName}} in all config greetings and system prompts

Jake Lasky
and
claude
committed
Fix agentName substitution and implement hotword detection via function call

Jake Lasky
and
claude
committed
Fix hotword matching: tolerate STT punctuation between words

Jake Lasky
and
claude
committed
Add conversational continuity after hotword activation

Jake Lasky
and
claude
committed
Add close_hotword_session for LLM-controlled conversation end

Jake Lasky
and
claude
committed

Commits on Mar 2, 2026

Disable Hey Manny demo, set Dubai voice to Pandora

Jake Lasky
and
claude
committed
Add ElevenLabs TTS support and Tagalog language for non-English STT

Jake Lasky
and
claude
committed
Set ElevenLabs Tagalog voice ID for bpo-tagalog config

Jake Lasky
and
claude
committed
Fix ElevenLabs speak provider structure per Deepgram VA API spec

Jake Lasky
and
claude
committed
Remove language_code from ElevenLabs provider; let voice auto-detect

Jake Lasky
and
claude
committed
Add model_id as URL query param for ElevenLabs reconnections

Jake Lasky
and
claude
committed
Add voice_id to ElevenLabs provider; revert URL query param change

Jake Lasky
and
claude
committed
Fix ElevenLabs BYO TTS: use multi-stream-input endpoint, QOL UI improvements

Jake Lasky
and
claude
committed
chore: ignore .claude/ and .planning/ directories
Jake Lasky
committed