Voice input now generally available.
Speak to the widget in any of 30+ languages. Whisper-grade accuracy, 200ms time-to-first-token, included on Professional and above.
Custom AI models on Business.
Bring your own fine-tuned model — we support OpenAI fine-tunes, Anthropic on-the-fly steering, and any OpenAI-compatible endpoint. Useful when your domain is genuinely unusual.
Widget now 12kb gzipped.
Trimmed 8kb off the runtime by lazy-loading the rich text renderer. The widget now beats most analytics scripts on weight.
Knowledge auto-refresh.
Crawled sources now re-index on a schedule you set (hourly / daily / weekly). Diffs show up in the dashboard so you can see what changed and why.
EU-only routing for LLM inference.
All inference now routes through EU-hosted providers by default. Opt out per workspace if you don't care about data residency.
Live agent handoff got smarter.
The AI now detects 'I want to talk to a human' in 30+ languages and hands off cleanly — without needing a literal command.
Analytics dashboard redesign.
Faster, sharper, with intent classification baked in. The old version is still available for 90 days at /analytics/legacy.