Release notes

When you're creating Actions, we've added some examples to the UI to help you write your paths. For more information, see Use JMESPath to save a portion of a JSON API response.

We've made several improvements to Ada Voice:

We've moved a few things around in the Conversations view:

In bots that serve generative content from your knowledge base, in Actions, we've made it required to add a name for each API output. This was optional before, but we've determined that adding names allows the bot to integrate API data into responses more consistently. For more information, see Configure API calls in your generative bot with Actions.

In bots that serve generative content from your knowledge base, we've introduced the ability to configure API calls, so your bot can fetch information for your chatters on demand. For more information, see Configure API calls in your generative bot with Actions.

In our ongoing efforts to ensure the highest quality and accuracy of your chatbot's responses, we've decided to deprecate the Auto-training feature. A few key factors influenced this decision:

What this change means:

For more information on how to effectively train your bot, see Training best practices.

For Voice bots, we've improved the experience around giving the bot more time to process things like API calls. We've removed the artificial typing sound, and replaced it with periodic messages like "one moment" to fill any silence. For more information, see Minimize pauses while your bot performs backend actions.

In bots that serve generative content from your knowledge base, we've introduced the ability to guide bots' behavior by creating custom instructions for it to follow. For more information, see Customize your bot's generative content with Guidance.

For generative bots, which automatically generate responses to chatters using knowledge base content, you can now customize your bot's behavior with custom instructions called Guidance. For more information, see Customize your bot's generative content with Guidance.

After a handoff occurs through Zendesk Messaging, Zendesk Messaging tickets originating from your bot will now be tracked in Zendesk through the "messaging" channel source. We're making this change to better comply with Zendesk's new omnichannel routing and reporting capabilities that rely on the "messaging" channel source.

This change takes effect once a chatter starts a new chat session. For bots that have long persistence settings, you or your agents may wish to tell chatters to clear their caches to display the updated channel source. For more information, see Agent Experience.

In Voice bots, you can now add custom bot vocabulary, so your bot can more easily pick out terms that are specific to your brand or your industry. For more information, see Help your voice bot understand special terms.

We've been working on reducing latency for blocks that perform backend actions (e.g., the HTTP Request, Set Variable, Answer Utilities, or Actions blocks). You can now strategically build Answer flows to significantly reduce pauses when these blocks appear and improve caller experiences. For more information, see Minimize pauses while your bot performs backend actions.

We've added a Play button to blocks that contain spoken content, so you can test your content directly in the dashboard.

In the menu where you can see an Answer's parent Answers, we've added an icon where you can easily see whether the parent Answer has live content for messaging, voice, or both.

When you're editing an Answer and click Test Answer, the testing window now opens to the same modality you were editing, depending on whether you were in the Messaging or Phone tab.

The Messaging and Phone tabs now stick to the top of your screen, so it's easier to keep track of which modality you're editing content for.

We've introduced Ada's new free plan, which you can use to create bots that run entirely on generative AI. These bots write their own replies using content from Zendesk Guide, with no additional bot building necessary.

For more information, see our documentation, or sign up to get started with your own free plan bot.

In the Answer Utilities block, we've removed the 1024 character limit for the Render Templated Text action's output. Previously, if the output was too long, it would show 1021 characters followed by ... and would truncate the rest of the output.

For Voice bots, we've reduced the time it takes for the bot to respond when it encounters Set Variable and HTTP Request blocks.

We've updated our Zendesk Messaging handoff to align with Zendesk's best practices for bot partners. While there are no changes to the builder or end user experience, please note that Zendesk Messaging tickets originating from your bot will now be tracked in Zendesk through a new channel source, "Sunshine Conversations API," instead of "messaging." Your Zendesk administrator may want to update your triggers and automations in Zendesk to leverage this new channel.

You can read more about configuring Zendesk Messaging with Ada in our guide Configure and use Ada Glass for Zendesk Messaging.

We've updated the Fetch Chat Metadata block, so it now only saves the current conversation with a chatter. Previously, it saved every message from all conversations the chatter had with the bot, which resulted in handoffs containing very long transcripts and irrelevant content.

This change primarily impacts channels where chatter IDs don't change between chatter interactions with an Ada bot, which includes third party channels (Facebook Messenger, Twitter, Instagram, WhatsApp), Voice, and bots that have their chatter persistence settings set to "Never Forget."

We fixed a display bug that caused the text that chatters entered to overlap with the Send button and shift when the text wrapped to new lines.

Our bot invitation emails now have "Activate your Ada account" as the subject line instead of "Invitation," so they're easier to spot in your inbox.

We've refined the prompt we use for AI rephrasing to better filter out inappropriate content.

You can now use Sunshine Conversations Web as a channel. If you're interested, contact your Ada team for assistance.

We fixed an issue with our Zendesk Live Chat handoff that was causing chatters to get stuck in a waiting state.

We fixed an issue where images that Zendesk Live Chat agents sent appeared as links to chatters.

For new bots, we've stopped creating default quick replies.

We fixed a bug affecting time conversions for some time zones.

We fixed an issue where screen readers would read out entire URLs for widgets. Now, they read out a dedicated aria-label instead.

We've updated our Ada Widget SDK. See the changelog and how to access the latest version here: https://www.npmjs.com/package/@ada-support/ada-widget-sdk.

We've fixed an accessibility issue that prevented chatters using screen readers from easily focusing on the Cancel button to close iFrames containing custom web widgets.

We fixed a bug where Smart Search wasn't coming up for a client's Salesforce or Zendesk knowledge bases.

We fixed a bug where the dashboard would time out when updating folders for large bots.

We fixed a bug that was causing training groups to not display for some large bots.

We fixed a bug that allowed Ada to respond to new messages after an agent closed a ticket without having to see a passControl first.

Improve and Create Variations are now available across all languages. Use these GPT-powered features to rephrase your multilingual Answer content, and to provide variations, to improve the conversational tone of your Answers.

We have upgraded Improve and Create Variations to GPT-3.5-Turbo.

We have improved our knowledge base parsing system so it removes documents that have duplicate content.

Additionally, our knowledge base parsing system also includes URLs when indexing chunks.

We have improved our logging around Spot node availability/termination.

We have launched a new dev2-green EKS cluster, which now serves all traffic in our dev2 environment.

We have enabled Loft & DevSpace on our new green cluster.

When you drag a Builder Note into an Answer, the block now automatically opens in edit mode.

We fixed a bug that was preventing Quick Reply blocks from showing up when you searched for them.

If you have an Answer open in the dashboard and delete it, rather than staying in the deleted answer, you now go back to your Answers list.

We have added more information to the login page, so we can keep you posted with the latest Ada news.

We fixed a bug that caused some Answer training from the Improve tab to fail.