Agent Experiences Release Notes

August, 2026

New Features

• Pro Agent switch mid-conversation: Admins can now flip a live conversation onto the Pro Agent without starting over. The switch is one-way, carries the prior transcript across so the Pro Agent picks up in context, and the toggle locks once Pro is engaged so a subsequent reload cannot drop the chat back to the standard agent.

• Bulk download of a Filespace as a zip: A "Download as zip" action on any Filespace streams every file in the space as a single archive — one click to grab an entire Filespace's contents without picking files individually.

• Full-viewport-width option for maximized embedded widgets: Embedders can opt into fullWidthOnMaximize on floating-launcher and inline-trigger embed configurations so the widget expands to the full viewport width — not just its default max width — when a user maximizes it.

Improvements

• Cleaner CTA loader in product-selection chats: The "recommended products" CTA placeholder no longer flashes speculatively after every answer. The loader now appears only when the backend signals a CTA is actually coming, so factual and support turns finish cleanly with no dangling placeholder.

• Embed widget: privacy hardening: Embedded agent widgets now honor Do Not Track and Global Privacy Control (GPC) signals from the visitor's browser, materially cut background analytics traffic on host pages, and emit an explicit widget_open signal so open-tracking is robust even when session capture is off.

Bug Fixes

• Embed widget: CSP-friendly reCAPTCHA: Restored the reCAPTCHA Enterprise script host so that embeds on pages whose Content-Security-Policy frame-src allowed only google.com can generate valid reCAPTCHA tokens again. Chat requests on those pages had been failing intermittently.

• Widget typeahead resilience: Fixed a widget-side tight loop where a transient reCAPTCHA failure could burst-refire the product-metadata typeahead call. Typeahead now degrades quietly to empty on failure and stops the request storm.

• Diagrams render correctly inside embedded widgets: Fixed the shadow-DOM styling and label-clipping issues that could cause agent-drawn Mermaid diagrams to render blank or clipped inside the embed. Diagrams in copilot answers now display correctly across embed hosts.

July, 2026

New Features

• Historical attachments are now clickable: PDFs, images, and documents shared in past conversations can be reopened from the conversation view — useful for revisiting context and reviewing what the agent was responding to.

• Chrome extension: side-panel mode: Dock the AI Agent inside Chrome's native side panel so it sits alongside the page you're working on, rather than as a floating widget.

• Chrome extension: strict-CSP site support: The Chrome extension now runs on sites with restrictive Content Security Policies (for example, in-house dashboards and other CSP-locked pages), thanks to a new CSP-proof widget shell.

• AI-usage disclaimer consent gate on the embed widget: Customers can require end users to accept an AI-usage disclaimer as a first-run screen before starting a conversation, with a customer-configurable HTML body per hub.

• Chat history persists in the Chrome extension across sites and sessions: The extension now uses a browser-storage bridge so conversation history reliably reappears on returning visits, working around per-site localStorage partitioning.

Improvements

• Chrome extension polish: A cluster of Chrome-extension fixes: the widget no longer appears inside OAuth login popup windows, inline-trigger clicks are consistently responsive, and the settings page holds up cleanly at narrow devtools-docked widths.

• Outline star icon restored for unstarred conversations: In conversation history, unstarred conversations show an outline star and starred conversations show a filled star, matching the earlier visual pattern.

• Widget accessibility (WCAG 2.2 AA quick wins): Keyboard and screen-reader activation on the widget launcher, Escape-to-close, ARIA-live streaming of the answer, and focus restoration when the widget closes.

• Chrome extension opens directly into the side panel: Clicking the extension toolbar icon now opens the side panel directly, replacing the previous popup step.

• Widget mounts reliably on managed networks and in restricted regions: A verification-scoring fix now reports the correct client IP (instead of our load balancer's IP), plus a rescue path for managed corporate networks and a verification host reachable from restricted regions. Non-retriable verification rejections now surface a clear toast instead of a silent block.

• "Was this helpful?" feedback on interactive widget answers: The feedback affordance previously only on text answers is now also available on chart, bill-of-materials, and diagram widget outputs.

• Search widget can spawn an on-page conversation child: Integrator improvement — a Search widget can now open a conversation window pinned to the same page.

Bug Fixes

• Multi-turn image and PDF attachments stick across the conversation: Fixed a regression where attachments shared in an earlier turn were dropped in later turns; the agent now correctly references them throughout the conversation.

• Embed widget conversation history is no longer silently wiped: Closed a production bug where a hub's per-user conversation history could get lost due to shared-storage partitioning.

June, 2026

New Features

• Inline citations on Pro Agent answers: Pro Agent now renders inline citations to the supporting sources alongside its answers — on parity with the citation behavior on our core conversational agent.

• Discord forum opt-in greeting card: On Discord agents configured in opt-in forum mode, new forum posts now get a friendly opt-in greeting card with explicit buttons and a per-hub-configurable prompt before the agent replies. Pairs with the new delayed-reply opt-in mode on the Agent Platform.

• Interactive widget answers: Beyond static diagrams, agents can now generate interactive widget-style visuals as a new answer format — automatically picked when a static diagram would be too cramped to convey the information. Live across the embedded copilot, the admin dashboard, and the Chrome extension. Rolling out broadly.

• HubSpot live-chat handover: Agent handover can now drop users into a HubSpot inline form or directly into a HubSpot live conversation thread inside the chat. Returning users pick up the same HubSpot conversation across sessions, so the chat is one continuous thread. Initial rollout to pilot customers.

• "Visualize this" for bills of materials and component lists: The agent now proactively offers an interactive visual when an answer is a bill of materials or a set of components that make up a system — instead of flattening it into a markdown table and waiting for the user to ask "make me a diagram". Comparison of alternatives still renders as a table.

• Multiple-choice questions: single vs multiple selection with persistence: MCQ widgets emitted by the Guided Explorer and Product Selection agents now render as radio buttons for single-select and checkboxes for multi-select, driven by the agent's intent. User selections and any additional note are persisted with the conversation and re-hydrated in read-only state when the conversation is reopened or refreshed.

• File attachments in agent conversations: Users can now attach files — PDFs, images, and other documents — to messages in the long-horizon agent runtime. The agent reads them mid-conversation and results survive session resume. Attachments are also available to Admin users in Converse (previously Super Admin only).

• Embedded widget: enterprise host compatibility: The embedded widget now degrades gracefully inside restrictive host environments so it can be shipped in more places:

  • Suppresses spurious error toasts on catalog requests behind corporate SSL-inspection proxies (e.g. Zscaler); falls back to a clean empty state.
  • No longer collides with host-page AMD/UMD loaders (e.g. Magento) — the widget's internal gesture handling has been rewritten to avoid global namespace pollution.
  • Salesforce Lightning Web Security and storage-blocking consent managers are handled at the Embed SDK layer — see Core Platform.

• New onLoad(ref) integrator callback: Embed SDK's onLoad callback now receives the widget ref so integrators can call ref.open(), close(), update() etc. directly from the load callback, without a separate lookup step.

Improvements

• "Pro Agent" branding: The long-horizon / multi-step agent mode introduced in May is now consistently branded as Pro Agent across both the user-facing conversation toggle and the hub configuration selector (previously - "Forge Agent").

• Mermaid diagrams in dark mode and full-fit zoom: Diagram toolbars and scrollbars are now theme-aware in dark mode, diagrams open centered by default, and the zoom-out floor has been removed so wide diagrams can be sized down to a true "whole diagram in one view" — addressing the previous behavior where wide diagrams could appear cut off with no way to zoom out far enough.

• Friendlier conversation error states: Conversation errors are more useful for end users — a clear "internal error" message with a request ID surfaces on backend 500s (so users can report it back to support), benign 401 "token expired" toasts are silenced, an explicit "please reload" block appears on embedded / whitelabel chat when reCAPTCHA scoring fails, and a handful of small UX issues (SSR copy-icon shift, streaming-textarea behavior, empty-chat styling) are resolved.

• Quieter reCAPTCHA in embedded chat: Transient reCAPTCHA token failures now auto-retry transparently in the background, so end users no longer see false "verification failed" errors during normal conversations.

• Clearer "Interactive" hint on widget answers: When an answer is rendered as an interactive widget, the message now shows a small "Interactive" chip in the title row and a "Click or hover to explore this diagram" footer hint. Both are gated on whether the widget actually has interactions, so static visuals stay clean.

• Embedded widget: correct brand colors without explicit theme configuration: The widget now falls back to a sensible default theme when the host page has not configured one, so brand accent colors render correctly out of the box.

• Embedded widget: product-catalog carousel sizing in narrow containers: Carousel slide count is now derived from the widget's container width instead of the host viewport, fixing over-sized carousels in modals and narrow mounts.

• Product Selection agent: MCQ direction copy: The agent no longer refers to a multiple-choice widget as being "above" the message when it actually renders below.

Bug Fixes

• More reliable embedded widget authentication: Resolved two further reCAPTCHA token failure modes on the embedded widget — concurrent token requests no longer collide onto the same token, and tokens minted on pages that host the agent inside a shadow DOM now carry the correct hostname. Follow-on to the widget auth hardening shipped in May.

• Chrome extension launcher restored: The Rapidflare Chrome extension's floating launcher button is back, now riding the unified embed SDK and free of the legacy script-injection issues that broke it on strict-CSP pages.

• Reasoning replay on past conversations: Opening a past conversation now reliably preserves the agent's reasoning steps and pipeline events, so users see the full trail that produced the answer instead of a stripped-down view.

• Pro Agent completion-email links resolve to the right conversation: Conversation links attached to Pro Agent completion emails (and the inbound-email reply path) were dropping the customer slug and hitting a stale path segment, producing a URL that did not resolve. The links now route to the correct dashboard conversation page.

• Promoted-answer image previews restored: Document images inside promoted answers replayed from history are once again shown as previews, matching live-streamed answers.

May, 2026

New Features

• Pro agent mode: Agents can now run multi-step, longer-running tasks. The conversation view streams tool events live and surfaces a dedicated artifacts panel, so users can watch the agent's work as it happens and inspect produced artifacts in place. This support is in early preview and rolling out to specific customers.

• Accessory and family-based product narrowing: For product discovery agents with a product catalog, the agent now recommends accessory product types alongside primary matches and surfaces product families with descriptions during the narrowing phase — guiding users toward the right product faster. Currently enabled for select customers.

• Redesigned call-to-action experience: A reworked CTA rendering layer ships against a new taxonomy-shaped agent response, giving guided-product and sales agents a more deliberate, design-led handoff into next steps (compare, explore, configure) instead of generic links.

• Private multi-turn Discord conversations: A new /private-ask slash command opens an ephemeral, multi-turn conversation with the agent inside Discord channels — letting users have a private back-and-forth with the agent without cluttering the public thread.

Improvements

• Conversation export quality: HTML and Markdown exports of conversations now correctly preserve tables, heading hierarchy and message-list formatting, producing cleaner artifacts for sharing and archival.

• Embedded widget polish: A round of fixes for embedded agent deployments — customer logos load reliably again on embedded widgets (was blocked by a security header in some configurations), single-theme logo fallback when only one of light or dark is configured, the search-mode loading skeleton now matches the container size and shows the resolved placeholder, and a misleading "check your internet" overlay was removed so real server errors surface to users.

• Feedback hover and popup polish: Faster hover response, smoother behavior when clicking the trigger right as the popup appears, a Firefox reliability fix, and removing a thumbs-down reaction no longer shows the success toast.

• Slack bot polish: Tighter feedback chiclets, thread-aware context handling, and a fix for a rare duplicate-reply path where the bot could post the same answer up to three times. Slack conversations are noticeably cleaner and more reliable.

• Clickable mailto: and tel: links in agent answers: Phone numbers and email addresses rendered inside agent responses now navigate as expected when clicked, instead of being treated as inert text.

• Richer agent answer rendering in product selection: The product selection agent now renders inline images and structured assets in its responses at parity with our core conversational agent, producing more visual and on-brand answers during product discovery.

• Conversation exports carry hub and date range in the filename: Exported conversation histories now include the hub name and date range in their default filename, making downloaded reports easier to organize and share.

Bug Fixes

• Share-link conversation viewer restored: Conversation share links and the read-only conversation viewer now render correctly again, after a regression introduced by the platform's message-history migration. End users who open a shared conversation will once again see the full transcript instead of an empty viewer.

• Embedded widget polish — citation badge, mobile sheet, floating popup, dismiss: Resolved four customer-visible widget bugs: the citation badge no longer flickers during streaming responses, the mobile bottom-sheet handles body margins correctly, floating popups with a custom trigger now show as expected, and overlays dismiss cleanly when the widget is closed via the embed-widget-closed event.

April, 2026

New Features

• "Visualize this" follow-up action: Users can trigger diagram generation as a follow-up in conversations, with improved Mermaid rendering stability and interactive pan/zoom.

• Quote from AI answers: Select text from an AI response to ask a follow-up or give targeted feedback on a specific passage.

• Longer AI responses: Output token limits raised reducing truncated answers in complex conversations.

Improvements

• Smoother quote-and-ask in embeds: Reworked selection handling and AI overview handoff for more reliable quote-from-answer workflows in embedded agent deployments.

Bug Fixes

• Embedded "Compare Products" links fixed on custom domains: Compare-product CTAs in shadow-embed experiences now correctly resolve to the customer's whitelabel domain instead of broken URLs.

• Proposal flow resilience: Proposals no longer error when recommendation payloads are incomplete

January, 2026

New Features

• Inline Citations: References and citations are now displayed inline within the AI's response text, making it easier to see the source of information directly in context rather than as a separate list at the end.

• Promoted Answers: AI Agent now supports promoted answers, allowing admins to configure specific responses for frequently asked questions that take precedence over AI-generated responses.

• Notes on Conversations: Admins can now add notes to conversations in the conversation history, enabling teams to annotate and track important context about customer interactions.

• Discord Forum Auto Reply: AI Agent now automatically responds to new posts in Discord forum channels, providing instant answers without requiring the /ask command. Enable this feature by using /install with the forum_auto_reply option in any forum channel.

• Enhanced Feedback Flow: The feedback comment popover now opens automatically when users click thumbs up or thumbs down, encouraging more detailed feedback collection.

Improvements

• User Invitation System: Admins can now invite users via email directly from the dashboard, streamlining team onboarding.

• Conversation History Export Enhancements: Exported conversation history reports now include improved formatting with centered text alignment and better Excel report structure.

• Copy Message Improvements: Copy-to-clipboard functionality now includes an option to copy formatted text as HTML for better pasting into documents like Google Docs.

December, 2025

New Features

• Discord Integration: Rapidflare AI Agent is now available on Discord. You can interact with your AI Agent directly in Discord channels.
  • /install [reply_type] - Activate the agent with your preferred response mode (public, ephemeral, or thread)
  • /uninstall - Deactivate the agent from a channel
  • /ask - Ask questions to the AI Agent
  • Feedback buttons for thumbs up/down on responses
  • See the Discord User Guide and Discord Deployment Guide for details.

Improvements

• Improved performance for repeat visits to the agent widget by introducing caching for certain API calls. This enhancement ensures that returning users experience faster load times and a more responsive interaction with the agent.

October, 2025

New Features

• Thinking Aloud / Reasoning Display: AI Agent now displays its reasoning process in real-time, showing users the steps the AI is taking to formulate an answer. This transparency helps users understand how conclusions are reached.

• Web Search Integration: AI Agent can now search the web for real-time information when needed, enabling responses that include current information beyond the indexed knowledge base.

Improvements

• Mobile Dashboard Responsiveness: Significant improvements to dashboard responsiveness on mobile devices, ensuring a better experience across all screen sizes.

• Product Comparison Tables: Enhanced comparison table rendering with carousel support for multiple product images.

September, 2025

We have introduced new breaking changes in a major version update to 3.0.0. This version contains several important security improvements and bugfixes.

3.0.0

Bug Fixes

• Fixed issue where in some slow network scenarios, customers would briefly see Rapidflare's logo in the launcher icon, which would then be replaced with the customer's icon shortly after.

Improvements

• In addition to our conversational agent experience, our widget now supports rendering our brand new product selection agent. To enable this feature, please contact support@rapidflare.ai and we will work with your team and to create a product catalog for your agent.
• API Keys are now safe to embed in your user-visible frontend website code bundles. Our new API keys are protected by reCAPTCHA and you can configure it to work on only specific domains that you choose to allow-list. Adding domains is optional but highly recommended. You can read more about our updated API key management described here
• Existing API keys continue to work as before without any action needed on your end. However, we strongly recommend that you generate a new API key from our admin dashboard under Hubs » Configure » Security with your trusted domains, and update to the latest version of our JavaScript widget in your web app as described here.

July, 2025

New Features

• Rapid Sales Agent: A new conversational experience for product discovery. Our newest agent asks clarifying questions to understand user requirements and guides them through finding the right products through an interactive multi-choice question flow.

• Widget Lifecycle Events: New event notifications when the embedded widget opens or closes, allowing your application to respond to widget state changes.

• Widget Fullscreen/Resize Controls: New maximize and minimize buttons on the embedded widget, giving users control over the widget size.

Improvements

• Starter Prompts Auto-Generation: Admins can now auto-generate starter prompts customized to their products and product families directly from the configuration screen.

• Copy Link to Message: Users can now copy a direct link to a specific message within a conversation for easy sharing.

May, 2025

New Features

• Human Agent Handoff: Users can now request to be transferred to a human support agent during a conversation. When enabled, a "Talk to Human" option appears, allowing seamless handoff to live support.

  • This feature is disabled by default. Contact support@rapidflare.ai to enable it for your organization.

• Whitelabel Domain Support: AI Agent can now be deployed on your own custom domain, providing a fully branded experience for your users.

Improvements

• Reduced Font Sizes: Improved readability with optimized font sizes across the chat interface based on customer feedback.

• Scroll to Bottom Button: A new button appears when users scroll up in a conversation, allowing them to quickly return to the latest messages.

March, 2025

New Features

• Chrome Extension: Rapidflare AI Agent is now available as a Chrome browser extension, allowing users to access the AI Agent from any webpage.

• Interactive Quiz Module: New quiz feature for training and assessment scenarios (available for select customers).

Improvements

• Safari Clipboard Support: Fixed copy-to-clipboard functionality on Safari browsers.

• Vimeo Video Support: Related content now supports Vimeo videos in addition to YouTube.

Dec 6, 2024

New Features

• AI generated titles for conversations to better represent and access past interactions
• Categorized starter prompts configuration. Now you can group starter prompts by their category names and also provide a simple help description which will be shown to users on hover.
• Comparison tables and other artifacts provided by AI Agent as part of it's answer can be downloaded as PDFs. This allows you to share the artifact with your end customers, or use as part of other document collections and reports.

Improvements

• For embedded AI Agents, we have even more options for customizing the launcher icon. Now you can provide your own HTML element as the launch trigger, and thus customize it's placement, styling. With this fine grained control, you can pretty much blend the Rapidflare AI Agent into your web apps such that it's no longer a standalone element, but rather an intrinsic pop up.
• When using our launcher, you can now pre-set the launcher's initial position to one of the four corners of the screen.

Bug Fixes

• Browser URL updates when clicking on recent conversations in Your Conversations

Oct 1, 2024

1.8.2

Bug Fixes

• Fix a bug that might trigger numerous retries to conversation stream API calls, causing API to reject new conversation requests.

Sep 5, 2024

New features

• You can now download a conversation as HTML, Markdown, PDF, JSON or Excel from the AI Agent conversation screen. This is similar to the admin dashboard conversation history export feature.
• New search field in the conversation history table in the Filter by user menu to search for specific conversations by one or more users.
• Conversation history table view now has a new Mixed feedback filter in addition to the Good feedback, Bad feedback, and No feedback filters. The Mixed feedback filter shows all conversations with a mix of both good and bad feedback.

Aug 24, 2024

New Features

• New sharing features for admins to enable users to share conversations.
  • Sharing Conversations: You can share a conversation through URL, email, HubSpot, or with anyone. Recipients of the shared link can view the conversation.
• Conversation history
  • Now enables export to PDF. This allows viewing all the conversations, messages exchanged and feedback on conversations in PDF format.

Aug 9, 2024

We are catching up on our July release notes in this early August release notes update.

New Features

• New read out loud mode for AI Agent answers. Click on the loud speaker icon.
• Voice input method to have your questions transcribed in vs. typed in.
• Related content (videos, learning, etc.) now presented in a tab on the right side.
• Quick and handy access to your personal conversation history (see tab on the right side)
• Better typing + reading experience. Typing input box remains pinned to the bottom of the screen, while content scrolls naturally in the content pane.
• Hovering on references provides inline previews if available (e.g. web links)

Improvements

• All of these new experiences work in both full screen mode on non-mobile screens, or in a mobile responsive format.
• Overall UI performance improvements provide a more snappy AI Agent experience

July, 2024

We are moving away from versioned AI Agent widgets to versioned AI Agent wrapper iframes. Release notes from this point onwards will be dated release notes for what's new in the content of our AI Agent widget, as well as any updates to versioned AI Agent wrapper releases.

2.0.0

• Embedded AI Agent is now wrapped by an iframe. Embedding through JavaScript works still the same.
• New features include ability to reposition chat bubble in any 4 corner of the screen, better responsive handling and better sandboxing.

June 21, 2024

1.8.1

Bug Fixes

• Starring conversations via now works reliably, fixing previous reliability issues.

1.8.0

Improvements

• New conversation icon changed to to better represent a "refresh" or a "restart". The previous icon was confusing and unclear.

1.7.5

Improvements

• In xs responsive breakpoints, feedback comment modal is now relatively sized bettero and doesn't fill screen.

1.7.4

Bug Fixes

• In xs responsive breakpoints, fix control bar layout issues.
• Reference links are now truncated 40 characters or less to additionally handle layout issues in xs breakpoints.

June 7, 2024

1.7.3

Bug Fixes

• Fixed issue where viewing a past conversation did not render related documents as it did during the live conversation.

1.7.2

Improvements

• Increase text size on xs breakpoints

1.7.1

Improvements

• Better layout for videos presented as related content

1.7.0

Improvements

• Better handling for xs (400px) responsive breakpoints.
• Reference links are now truncated 50 characters or less so they don't overflow xs breakpoints
• Remove unnecessary padding in frameless mode

May 24, 2024

1.6.0

New Features

• Optional, frameless mode for embedding in mobile web views

Improvements

• Reduce font size of "version" string in title in the AI Agent header

1.5.4

Improvements

• Increase size of logo to in launcher icon to 24px in small responsive width (640px)

1.5.3

Bug Fixes

• Sandbox AI Agent's CSS to avoid impacting host site styles

May 10, 2024

1.5.1

Bug Fixes

• More robust conversation link sharing behavior, taking into account embedded vs. AI Agent dashboard based sharing behaviors.

1.5.0

New Features

• AI Agent now presents related links (Videos, Learning modules)
  • Note: This feature is enabled only for customers for whom we are ingesting such content.

1.4.2

Bug Fixes

• Fixed issue where message feedback existence was not being logged against conversations

1.4.1

New Features

• AI Agent now supports Advanced AI models.

  • Our Advanced models follow instructions better, are more robust at answering questions and in general more powerful for a variety of tasks, with greater world knowledge.
  • Contact us at support@rapidflare.ai if you would like to set your enterprise to this more advanced model.

  Note: These models will take longer to generate responses.

• Responses from the AI Agent will now be always streamed as they are being generated. Since advanced models take longer to complete this will enable humans to start processing the AI Agent's responses as they are being generated.

1.3.0

New Features

• You can type multiline messages to the AI Agent (press Shift+Enter) and the input box will resize to give you more room to type as well.

1.2.7

Improvements

• AI Agent now logs hostname from which conversations originate to facilitate better analytics.

Apr 26, 2024

You can pin to major and minor release streams in addition to specific and latest versions. See the documentation here.

1.2.6

New Features

• Copy support for messages
• Download support for conversations
• AI Agent now honors additional guidelines. Admins can configure additional guidelines in the AI Agent Configuration page.
• AI Agent can now be optionally powered by an advanced AI model that is better at following guidelines and produces even more accurate responses.
  • Self service for this feature will be coming soon.

Improvements

• Web widget retains conversations when reopened after being minimized.
• More responsive and improved layout on mobile screens and small laptop screens.
• Conversation sharing (via dashboard links) now enabled

Bug Fixes

• Fix for internal errors in cases where feedback or references are empty