What we shipped. When we shipped it.

Three more Indian languages live on voice — bringing the total to twelve. Full code-switching support included for English-mixed conversations. Built natively, not via translation layer.

A new generic HTTP step means you can connect AIVA workflows to any REST API — no waiting for us to build official integrations. Configurable auth, custom headers, response parsing.

Inference pipeline rewrite shaves 40ms off the round-trip. Most noticeable on long conversations where token-by-token streaming compounds. Engineering write-up.

Build conversation-triggered actions without writing code. Drag-and-drop steps, configure each one with a form, test with one click. Most teams ship their first workflow in under 30 minutes.

• 200+ built-in integrations (Stripe, Zendesk, Slack, Shopify, Salesforce, more)
• Generic HTTP step for anything else
• Conditional branches and retry logic
• Version history — roll back to any prior save

The conversation-volume chart now shows per-hour breakdowns. Especially useful for ops teams trying to staff shifts. Export to CSV is one click.

Fixed a rare bug where the widget would show “connecting…” indefinitely on flaky 3G networks. aiva.js@2.2.3 resolves it.

The Next.js SDK now supports Vercel Edge runtime and Cloudflare Workers. Same useAiva() hook, no changes needed. Edge guide.

The webhook signature header is now X-Aiva-Signature (was X-Signature). The old header still works for 90 days but will be removed in v2.4 (April 30). Migration guide.

Enterprise customers can now sign in via SSO. Tested integrations with Okta, Auth0, Microsoft Entra ID, and Google Workspace. SCIM provisioning included.

Emoji now render correctly on Jio and Airtel SMS. Previously they'd come through as ? on certain feature phones. Encoding fix in our Twilio adapter.

Conversation timestamps were showing in UTC for some Indian customers. Fixed — they now correctly default to Asia/Kolkata.

Configure exactly when AIVA should hand off to a human. If the customer mentions specific keywords, or the conversation goes past N turns, or a workflow returns failure — escalate to Slack, email, or a phone agent.

Connect your help docs to AIVA in three clicks. Auto-sync on changes, with diffs shown in your dashboard. Notion was supported earlier; this adds Google Docs and Confluence.

A 180ms gap would occur when AIVA transferred a call to a human agent. Fixed with smoother handoff buffering. Most customers won't notice the difference, but power users on quality-graded headsets will.

Live KPIs, conversation feed, intent breakdowns — refreshed every 2 seconds. Custom dashboards, drag widgets, save layouts, share with your team.

Annual plans now save you 2 months' worth on every tier. Existing monthly subscribers can switch in one click — we'll prorate the difference.

Five Indian languages live on voice on day one of 2026. Full code-switching support, native scripts, regional accent recognition. Why we built this first.

Audited by an independent firm. Enterprise customers can request the report under NDA. Type II in progress, target Q2 2026.

EU customers' data now stays in the EU (Frankfurt, eu-central-1). US customers' data stays in the US (Virginia, us-east-1). India default remains Mumbai. Configurable per-tenant.

Fixed a Cumulative Layout Shift issue affecting Lighthouse scores on iOS Safari. The widget's mount no longer pushes page content during initial load.

Configure refund rules in plain English: “if customer asks for a refund under ₹5,000, process it via Stripe.” AIVA understands the intent, runs the refund, replies. 1.8s end-to-end.

Complete platform rebuild with three goals: faster, more languages, less infrastructure. Voice latency cut in half. Frankfurt and Virginia regions live. SDK rebuilt with native streaming. Full rebuild story.

• Voice latency: 600ms → 280ms (2.1× faster)
• Languages: 3 → 5 (with 7 more shipping in 2026)
• Regions: 1 → 3 (Mumbai · Frankfurt · Virginia)
• SDK: streaming-first, type-safe across all targets