Subscribe to release notes

Background Noise Cancellation for Voice

Ada Voice AI Agents now filter out background noise before speech detection, so conversations stay on track even when end users are calling from noisy environments.

• Cleaner input, stronger outcomes — Background sounds like traffic, crowds, and office chatter are filtered out before they reach speech detection and transcription, preventing mis-triggers and transcript errors.
• Built for real-world conditions — Whether your end users are calling from busy streets, shared offices, or vehicles, your AI Agent handles it without breaking conversational flow.
• Higher containment, more confidence in Voice — Fewer noise-driven interruptions mean better automated resolution rates and higher CSAT at scale.

Background noise cancellation is enabled by default for all Voice AI Agents. If needed, you can disable it on the Ada dashboard under Config > CHANNELS > Voice > Configuration.

For more details, see Voice onboarding.

Test AI Agent replaces Coaching Preview

The Preview button in the Coaching form has been replaced with Test AI Agent.

• Save your Coaching first, then click Test AI Agent to open a test conversation and try it out.
• Test AI Agent works with all Coaching behaviors. Preview was limited to Send a message–type Coaching only.

Learn more in the Coaching tools documentation.

Testing at Scale now supports Voice conversations

Your AI Agent’s Testing at Scale capabilities now extend to the Voice channel so you can extend the same quality assurance coverage to Voice that you already have for Web Chat and Email!

If you have the voice channel set up on your instance, you can create voice channel specific test cases, simulate voice conversations, review pass/fail results, and listen to an audio playback of the simulated transcript to hear how your AI Agent will sound when responding.

To get started, create a new test case and select Voice as the channel.

For more details, see Testing at Scale.

Lightbox viewer for images in handoffs

Images shared during handoff conversations now open in an inline lightbox viewer instead of a new browser tab, keeping agents in context during active handoffs.

Highlights

• Lightbox view – Clicking any image in a handoff conversation opens it in an inline overlay.
• Carousel navigation – Scroll through multiple images in the conversation without closing the viewer.
• Download and print – Click the image within the lightbox to access download and print options.
• Quick dismiss – Press ESC or click the close button to exit the lightbox.

Scope

This update applies to all handoff types where images are displayed by the system. Images shared as links are not affected.

Handoffs API now generally available

The Handoffs API is now generally available. It enables custom handoff integrations that connect an AI Agent to any live agent platform or internal system, without requiring Ada Solutions engineering.

API endpoints

Five REST endpoints are available for managing conversations during an active handoff:

• Send message (POST /v2/conversations/{conversation_id}/messages/): Send messages from a human agent to the end user during a handoff.
• Get messages (GET /v2/conversations/{conversation_id}/messages/): Retrieve the full conversation transcript with pagination. Supports text, file, and link content types.
• Upload attachment (POST /v2/conversations/{conversation_id}/attachments/): Upload files up to 50 MB to a conversation via presigned URL. Returns a download URL valid for 7 days.
• Update conversation (PATCH /v2/conversations/{conversation_id}/): Update conversation metadata. Supports string, boolean, integer, and float values.
• End handoff (POST /v2/conversations/{conversation_id}/end-handoff/): End the handoff and return control to the AI Agent.

Webhook events

Two webhook events support real-time integration during handoffs:

• v1.conversation.message: Fires on every message during an active handoff, delivering text, file, or link content in real time.
• v1.conversation.handoff.ended: Fires when a handoff completes, enabling cleanup and state sync in external systems.

Integration features

• handoff_integration field: Identifies which integration is active on a conversation. Use it with Svix Channels to filter webhook deliveries so each integration only receives its own events.
• ai_agent_domain field: Included in all webhook events. Identifies which AI Agent environment the event originated from (for example, acme.ada.support or acme.eu.ada.support). Essential for multi-Agent or multi-region setups.
• HTTP Request Block trigger: Start a handoff by configuring an HTTP Request Block in the AI Agent’s automation. The block calls the integration’s endpoint and signals the handoff start.
• Svix Channels filtering: Filter webhook deliveries by handoff_integration value so each integration only receives relevant events.

Get started

• Conversations API reference for endpoint details, authentication, error codes, rate limits, and Svix webhook verification.
• Creating a custom handoff: A step-by-step developer guide covering the full integration lifecycle, from configuring the handoff trigger to sending messages, handling file attachments, and returning control to the AI Agent.
• ada-handoffs-api-demo: An open-source reference app demonstrating the full handoff lifecycle, including agent messaging, file uploads, transcript rendering, and handoff termination.

Pre-greeting context and sensitive metadata for End Users API

Two new capabilities are available on the End Users API for setting user context and passing sensitive values through Ada’s public APIs.

Create end users before a conversation

A new POST /v2/end-users/ endpoint allows creating an end user with profile data (language, metadata) before starting a conversation. When POST /v2/conversations/ is called with the returned end_user_id, the AI Agent has full user context from the first turn. This eliminates order-of-operations issues where metadata was unavailable at greeting time.

Available for custom channel (Conversations API) integrations only.

Sensitive metadata

A new sensitive_metadata field is available on both POST /v2/end-users/ and PATCH /v2/end-users/:id, nested inside the profile object with a fields sub-object. Values passed through this field are encrypted at rest, redacted from the dashboard, excluded from LLM context, and automatically deleted after 24 hours. The field is write-only and never returned in API responses. Maximum 20 key-value pairs per request.

Use sensitive_metadata for auth tokens, session IDs, or personally identifiable information. This matches the security contract of the Chat SDK’s setSensitiveMetaFields().

• sensitive_metadata on POST /v2/end-users/: available for custom channels only.
• sensitive_metadata on PATCH /v2/end-users/:id: available for all channels.

For integration patterns and examples, see the End Users API developer guide.

Updated Zendesk Guide integration

The Zendesk Guide knowledge integration has been updated with new capabilities.

New features

• Article tags: Import Zendesk article labels as tags in Ada to make filtering and segmenting knowledge content easier.
• Private articles: Sync articles that are only accessible to authenticated Zendesk users by providing a login email during setup.
• Multiple help centers: Connect more than one Zendesk help center to a single AI Agent.

Migration

If you connected your Zendesk help center before March 25, 2026, the existing integration continues to work. Migration is required only to access the new features above.

See Import articles from Zendesk Guide for setup and migration instructions.

Amazon Connect Voice integration improvements

The Amazon Connect voice integration now supports two new capabilities:

• Pass metadata from Amazon Connect to Ada — The Ada-Get-Call-ID Lambda can now be used to send metadata from Amazon Connect to Ada at the beginning of a conversation. Values sent will be available in Ada as purple metavariables.

• More clearly identify if a call should continue in Amazon Connect — The Ada-Get-Transfer-Data Lambda response now includes a should_continue field. If the value of this field is true, that means that Ada escalated the call to a human agent and the call should continue in Amazon Connect. If the value of this field is false, that means that either Ada or the end user ended the call without escalation and the call should be ended in Amazon Connect.

Both changes are backwards compatible. Existing Lambda implementations using GET requests continue to work without modification.

To learn more, see Integrate Voice with Amazon Connect.

Connect Zendesk with Global OAuth

Zendesk Handoff integrations now use a centralized OAuth connection instead of manual API token entry. Connect each of your Zendesk subdomains once from Config > Apps > Zendesk, and the connection is shared across all Zendesk Handoff integrations except Zendesk Chat.

Why this is changing

Zendesk is deprecating API token authentication for all integrations. Previously, connecting Ada to Zendesk required generating an API key in Zendesk and entering it manually. That key had no expiration and required manual rotation.

With OAuth, you authorize Ada to access your Zendesk account through Zendesk’s secure login flow. The connection stays active automatically in the background with no recurring reauthorization and no keys to manage. Access can be revoked at any time from your Zendesk admin settings. Existing configuration such as webhooks and triggers carries over.

Accounts that need to migrate will receive detailed instructions via email. Existing API token connections remain fully functional during migration. Nothing will stop working while you transition to OAuth. All accounts are expected to complete migration by the end of May 2026. For additional guidance, contact your Customer Success Consultant (CSC).

Learn more in Zendesk Handoffs.

Updated Salesforce Knowledge integration

The Salesforce Knowledge integration has been updated to use External Client Apps with the Client Credentials OAuth flow. The new integration adds support for multiple Salesforce Knowledge connections, filtering synced articles by data category or locale, and automatically syncing Salesforce data categories as tags in Ada.

Customers who connected their Salesforce knowledge base before March 17, 2026 can continue using the existing integration. See Import articles from Salesforce Knowledge for migration steps.