VoiceThere 0.2.0
September 17, 2026 · release
@voicethere/agent 0.7.11 · @voicethere/cli 0.14.1 · @voicethere/client 0.8.0
VoiceThere 0.2.0 is the second production cut of the hosted platform. This post ties together the features that shipped since 0.1.0 and points to the docs you need.
Dashboard: app.voicethere.io
Code tab
Edit agent TypeScript in the dashboard without a local upload for every tweak. The Code tab has a file tree, Monaco editor, save, typecheck, compile, and deploy. Seeded templates load source on create. New projects get a root package.json with npm scripts for local verify and upload.
IntelliSense for @voicethere/agent and Node loads as soon as you open the editor. Typecheck refreshes dependency typings from package.json. Public npm semver packages resolve on compile with no toggle.
Per-plan caps: Free 10 MB source / 10 MB typings pack; Budget and Budget+ 25 MB each; Advanced and Ultimate 50 MB each.
The Recording consent template shows how to ask for recording permission on session connect and pause capture around PII. See Conversation recording and the recording-consent template post.
- Dashboard code editor
- Template workspaces with package.json
- Default IntelliSense
- Source size limits
- Typings pack limits
Spatial audio, mix mute, clip playback
Voice+Data sessions support stereo mix groups, live poses, and per-client TTS placement. Mute globally or per listener. Inspect mix state with getClientMixStatus. Play sound effects and notification clips with play(); they mix with microphones and TTS but are not speech synthesis.
Three live showcases on the marketing site demo orbit, soundboard, and proximity room modes. Put on stereo headphones and try them.
- Positional mix · Sept 2 post
- Mute and STT overrides
- Clip playback · Sept 7 post
- Spatial audio showcases · Showcase post
Spoken language ID and inbound RNNoise
Whisper tiny can emit a user_language event so your agent can switch prompts or TTS voice. Tune languageId.enabled and minSpeechMs from the dashboard. Inbound RNNoise is on by default for voice runners; turn it off per project when CPU is tight.
- Spoken language identification · Dashboard settings
- Inbound RNNoise default on
- Advanced voice settings
Embed widget designer
Each project has a Widget tab to pick presets, set brand colors, and preview the floating launcher. Your client API key stays on your page. @voicethere/client 0.8.0 adds createVoiceThereWidgetAsync for production embeds.
Publishing appearance JSON to the VoiceThere CDN (so sites load theme changes without editing host HTML) is rolling out with this cut. It is not on production yet. When it lands, the Access snippet will pass configUrl after Deploy.
npm and CLI patches
@voicethere/agent 0.7.11 sends the first spatial pose before the greeting speak() in the spatial-showcase template. @voicethere/cli 0.14.1 appends the last deploy job error when deploy --wait times out and the control plane recorded one.
Reliability
Agent speech no longer chops on a fixed 20 ms on/off gate. That reduces clipped first words on short replies, especially when spoken language identification runs on the same utterance.
LLM agent discovery
Production serves spec-compliant /llms.txt and /llms-full.txt so coding agents find docs, OpenAPI, sessions, webhooks, and npm packages without scraping the site.
curl -sS https://app.voicethere.io/llms.txt | head
CLI defaults to production
The CLI targets the production API by default: https://app.voicethere.io/api/v1. Override with --api-base or VOICETHERE_API_BASE when you need another host.
npm install -g @voicethere/cli
npm install @voicethere/agent @voicethere/client
voicethere login
See Quickstart and npm packages.