Changelog

Now you can choose how long each AI agent keeps its data. Set a number of days, and anything older is deleted automatically.

This helps you keep only the data you need and meet privacy requirements such as GDPR.

You can define retention periods to auto-delete:

• Conversations, including their messages, attachments and analysis
• Contacts
• Exports

To set it up, open your AI agent and go to Settings > Data retention .

See what your customers talk about most and how they feel, with two new Advanced analytics features for AI agents.

Topics

AI automatically groups your conversations into topics and subtopics, such as "Billing > Refunds" or "Orders > Delivery times". There's nothing to set up: topics are auto-created from your conversations.

• See the most discussed topics in the new Top topics card on the Chat analytics page.
• Filter conversations by topic.
• Add, rename, disable or delete topics in Settings > Insights, one at a time or in bulk.

Sentiment

AI detects whether each customer left the conversation feeling positive, neutral or negative.

• Track how sentiment changes over time in the new Sentiment card on the Chat analytics page.
• Filter conversations by sentiment, for example to quickly find and follow up with unhappy customers.

Both features are on by default for AI agents on plans that include Advanced analytics. You can turn either one off in Settings > Insights.

Conversations are analyzed from the day the feature is turned on.

1. Chain multiple requests in one HTTP action

• HTTP Request actions can now make several API requests in a row.
• Wildcard response mapping: pull values from lists with paths like items[*].name. Lists of simple values are joined with commas when saved to a text variable.
• Better debugging: the conversation debug panel now shows details for each request, with JSON formatted so it's easy to read.

2. More control over website training

• Dynamic website: turn this on for sites that load their content dynamically, for example with JavaScript, so a more advanced crawler is used to fetch the full page content.
• Remove page clutter: strips menus, ads, pop-ups and other clutter from scraped pages. Turn it off for any source where you want the raw page.
• Website crawler has been optimized to discover links faster and the limit has been increased to 50,000 pages.

3. Security

• Changing or resetting your password now signs you out on all other devices.
• Added extra protection against brute-force attempts on login, sign-up and password reset.

4. Other improvements and fixes

• Added better support to the chat widget for right-to-left languages like Arabic and Hebrew.
• In AI action input parameters, you can now clear a parameter's variable, and setting up conditions is easier.
• The floating chat window no longer goes off the top of the screen on short browser windows or sites with a large bottom offset.
• Fixed an issue with billing where increasing one add-on and decreasing another in the same update could undo the increase at renewal.
• Inbox assignment now only lists teammates who have access to that agent.
• FAQ sources with empty text now show a clear error instead of failing.

You can now message WhatsApp customers outside the 24-hour reply window using message templates approved by Meta.

What's a message template?

It's a pre-written message that Meta approves in advance, like an order update, an appointment reminder, or a friendly "just checking in." You can personalize each one with details like the customer's name or order number, and add an image and buttons.

With templates, you can message customers outside the 24-hour window using the pre-approved messages.

Create your first template

1. Go to Deploy , then click Templates on the WhatsApp card.

2. Click + New template , give it a name, and write your template.

3. Click Submit for review . Meta usually approves templates within 24 hours.

Send it whenever you need

Once approved, send a template from any WhatsApp conversation, even after the 24-hour window has closed. Or add it to a workflow to send follow-ups automatically. When your customer replies, you can chat as usual again.

Learn how to create templates →

We’ve added four new models from OpenAI, Anthropic, and Google, giving you more choice across quality, speed, and cost.

• GPT-6.1 Sol : OpenAI’s newest Sol model, with near-Astra performance at a much lower cost. Strong reasoning and usage for complex, multi-step conversations.
• Claude Sonnet 5.5 : Anthropic’s best balance of speed and intelligence. Strong reasoning and instruction following for detailed answers.
• Claude Haiku 5.5 : Anthropic’s fastest model, built for high-volume agents. Handles everyday questions quickly and at a low cost.
• Gemini 4 Argon : Google’s flagship reasoning model. Built for complex questions and multi-step tasks.

All models are now available in paid plans. Head to your agent’s AI settings to try them and find the right balance of quality, speed, and cost for your needs.

1. AI reply exclusion rules

You can now prevent AI Agents from replying to messages that match specific words, phrases, or patterns.

• Create multiple rules with channel controls.
• Skip reply, pause AI immediately, or reply before pausing.
• Resume AI manually from the conversation or automatically after a set time.
• View exclusion and resume events in the conversation timeline.

Configure this under Settings → AI .

2. Configurable link behavior

You can now choose whether links in web widget messages, citations, and sources open in a new tab or the current tab. Configure this under Web Widget Settings → Configure .

3. Knowledge base improvements

• Added “Last synchronized” to source details.
• Fixed automatic recovery of stuck resyncs.
• Next auto-sync dates now update only after successful synchronization.

4. Conversation events are now shown inline

Important conversation updates, such as priority changes, assignments, and status changes, now appear directly in the conversation timeline, making it easier to understand what happened and when.

When more than one agent has the same conversation open, you can now see who else is there before you send a reply.

Above the composer, a pill shows teammates who are viewing, typing a reply, or writing a private note. Notes use a yellow pill, so you can tell them apart from a customer-facing reply. You can still send if you need to jump in.

In the sidebar, an "Also here" field will appear in Chat information when teammates are viewing or typing in the same conversation.

AI Agents can now let customers do bookings through Google Calendar or Calendly without leaving the conversation.

Connect Google Calendar to offer times based on your availability while automatically excluding busy periods. Bookings can include calendar invitations and Google Meet links.

For Calendly , add an event link to display its booking widget in the chat. Name and email can be prefilled, and booking information can be saved to variables for later use.

We’ve added three new features to help your team organize conversations and respond faster.

1. Conversation labels

Create custom labels, apply multiple labels to a conversation, and filter your inbox to quickly find related chats.

2. Live visitor indicator

See whether a visitor is currently online directly from the conversation, so you know when it’s the best time to reply.

3. Team mentions

Mention teammates in private notes using @name . They’ll receive an email notification with a link to the conversation, making collaboration and handoffs easier.