Release notes

The AI Agent now records why it hands a conversation to a human agent. It records a reason_code from a fixed list of ten values, or null when it records no code. It also records a reasoning text when one is available.

The v2 Conversations endpoint of the Data Export API now returns a handoffs array on each conversation. Each entry is a handoff attempt with its outcome_status, reason_code and reasoning. The get_conversation MCP tool returns reason_code and reasoning.text on its handoff entries. The new HANDOFFREASON filter in get_available_filters filters conversations by reason code.

A Voice Proactive Outreach that runs a Playbook can now start every call with an Opening line: the exact words that the end user hears the moment they answer.

The line does not depend on how the Playbook begins, and it plays before the Playbook’s first step is ready. Write it once in the default language of your AI Agent. Include your business’s registered name, along with any AI disclosures or recording notices required by law. Every call then opens with those exact words before the Playbook runs.

A new Keypad entry timeout (seconds) setting is on the Voice Configuration tab. It controls how long your AI Agent waits after the last key press before it uses the digits. Set it between 2 and 7 seconds. Pressing # submits the digits at once.

The timeout applies wherever an end user enters a value on the phone keypad. That includes Ask steps in Playbooks, Action inputs, and keypad capture in Greetings, Handoffs, and Processes. A shorter timeout makes keypad entry feel faster. A longer timeout gives end users more time between keys.

Admins can now save the Zendesk Messaging signing key ID and shared secret in the dashboard. Open Config > AI AGENT > Handoffs, then Configure beside Zendesk Messaging. Before this change, the Ada team added the key for you.

The Embed2 web chat now supports the Zendesk JWT. After you save the key, the web chat requests the JWT from zdChatterAuthCallback, as the Messaging SDK does. Ada uses the verified details to find or create the Zendesk user at the Handoff. See Use a Zendesk JWT.

The File Upload block now accepts .avi, .har, .json, .log and .neon files. To let end users upload these files, add the extensions to Accepted File Types in the block.

Existing File Upload blocks do not change. A new file type shows in the upload window only after you add it to the block.

The Enable image uploads to your AI Agent toggle is now off by default. An AI Agent that moves to the Messaging SDK does not accept images until you turn the toggle on. AI Agents already on the Messaging SDK keep their current setting.

To let end users attach images, go to Config > CHANNELS > Messaging. On the Appearance tab, open the General drawer. Enable Enable image uploads to your AI Agent. For more information, see Images.

The Messaging Android SDK opens the system chooser when an end user attaches a file. When the chat asks for images, the chooser offers the camera, shows only image files, and lets the end user pick several images at once. The chat still accepts up to three images in one message.

AdaMessagingActivity and AdaMessagingDialog use the chooser with no change. An app that embeds AdaMessagingView and sets filePickerCallback can call the new createChooserIntent and onChooserResult methods. See File uploads.

A change set can now hold up to 20 entities. The previous limit was 10. See the edit_agent_behavior limits.

The limit of 20 active change sets per Agent is unchanged. A change that would add a twenty-first entity to a change set still returns an error. Remove a change, or put it in another change set.

edit_agent_config now accepts entity_type='variable' with the create operation. An assistant that builds a Playbook can create the global variable it needs without leaving the MCP session. The variable serves a set, ask or if_else step, or a tool output binding.

A variable takes a name, a description and an optional type (text, number or yes/no, default text). Only global scope is supported. Rename and delete stay on the Variables page. Both change every Playbook and rule that references the variable.

The Voice Proactive Outreach report now shows the AR rate and the CSAT score of the conversations your outbound calls started. Both appear in the tiles at the top of the report, in the table of Proactive Outreach items, and in the CSV download.

• AR rate is the percentage of engaged conversations whose latest automated resolution assessment is Resolved, out of those with an assessment.
• CSAT is the percentage of engaged conversations whose latest CSAT response is positive, out of those that received one.

The Amazon Connect handoff can now access your AWS account through an IAM role secured with an External ID, so no long-lived access keys are shared. IAM role access is the recommended option for new setups. Existing setups that use an IAM user keep working.

You can also forward Amazon Connect queue events to the integration. The queue a chat entered, and when, are then recorded on the conversation as meta variables. This step is optional.

Text inputs and variables now offer Speech and keypad and Keypad only in their Voice capture settings. This applies to Action inputs and to Playbook Ask steps.

Use a Text input for digit strings that can start with 0, such as the last four digits of a card, an expiration date, or a one-time code. A Number input drops leading zeros, so “0430” is saved as 430.

Availability rules on knowledge sources shipped over the last two weeks for web import sources, Knowledge API sources and the sources that an Ada App creates. This note covers the changes that came with that work and have no release note of their own.

Rename and delete a Knowledge API source from the dashboard

Playbooks now have a NOTE step. AI Managers can record the intent of a flow next to the steps it explains. For example, a note can say why a branch exists or what the values of a variable mean.

What’s new

A NOTE step is an annotation for the people who edit the Playbook.

The v2 Conversations endpoint now returns a conversation’s Topic even when the conversation has no classified Intent.

Before, a conversation with a real Topic and no classified Intent returned an empty classifications array, so the Topic was missing. Now classifications includes the Topic, and its intents field is an empty array. A conversation with a real Topic and a real Intent returns the same result as before.

Fixes

A Variable reference in a Playbook’s General Guidelines now reads as the variable’s name instead of raw template text. Before this change, a reference such as {{ variable:... }} could reach the end user as written. General Guidelines are guidance, so a reference names the variable; to use a variable’s value, put it in a SEND, ASK, or SET step.

Proactive Outreach, formerly called Proactives, now has its own page at Config > AI AGENT > Outreach. The page replaces the Proactives tab in Config > CHANNELS > Chat/Messaging. Your existing Proactive Outreach items show on the new page and work as before.

We’ve also added the ability for your Voice AI Agent to proactively reach out to your end users with Voice Proactive Outreach. Your AI Agent calls an end user by phone, and the Agent runs a linked Playbook when a person answers. Start a call in one of two ways:

The Snapshot API now exports and imports the image redaction setting, in the image_redactions section. The section holds the 2 image types that you can let your AI Agent read: government_id and other_credential. The Agent always withholds payment card images, so the section does not include them.

• Export: the snapshot includes both image types. true means that the Agent withholds the image. false means that the Agent reads it.
• Import: the import replaces the setting on the target Agent. If the section does not include an image type, the target Agent withholds that type. If the import does not include the section, the setting does not change.
• Validate: a mode: validate run names each image type that the import changes.
• Audit log: if the import changes the setting, it writes one Updated row for Image Redactions.

When your AI Agent withholds a sensitive image, its default reply now names the document type, for example “your card” or “your passport or ID”. The Agent then gives one next step. For a payment card, it may ask for the last 4 digits as text. For an identity document or other credential, it offers a human agent when a handoff is available. When no handoff is available, it says that human support is not available and keeps helping. The Agent still never asks for a full card number or a document number.

You can now choose which types of sensitive image your AI Agent withholds. Two toggles under Image redactions cover government ID documents and other credential documents. Both are enabled by default, which matches the previous behavior. Disable a type when your end users must send that document to complete a flow, such as a bank-transfer confirmation.