What we shipped this week
New features, improvements, and fixes — the same day they land in production. Subscribe to the RSS feed or get notified by webhook.
Recent changes
Real-time emotion steering
You can now drive the emotion of synthesized speech mid-stream — pass emotion: 'warm' | 'excited' | 'calm' on each chunk and the prosody adapts instantly. Available on all voice.stream calls.
Avatar render latency -35%
Rewrote the scheduler to batch render jobs by model variant. Median time-to-first-frame dropped from 4.2s to 2.7s for 1080p output.
Webhook signature edge case
Fixed a bug where webhooks containing non-ASCII characters could produce an invalid HMAC signature, causing verification failures on some clients. All signatures are now computed over the raw UTF-8 body.
Hindi & Arabic voice models
Shipped native-quality voice models for Hindi and Arabic, bringing the library to 30 supported languages. Both include dialect variants.
Go SDK 1.0
Promoted the Go SDK to v1.0 with a stable API surface, full streaming support, and context-aware cancellation.
Agents tool-calling
Agents can now call your tools. Define a function schema, pass it to agents.converse, and the model will emit structured calls you can execute and feed back.
Build on what we just shipped.
Free to start — 10,000 credits every month, no credit card required.
SOC 2 Type II · GDPR · HIPAA-ready · No vendor lock-in