Gydr v3.3.3

Fixed

• WhatsApp messages from every sender now arrive — a message whose sender identity was delivered in a newer format could be dropped before your assistant saw it. It is now understood and answered like any other message.

Changed

• Shopify products answer like catalog products — when your assistant finds items in your Shopify storefront, it presents them the same way it presents your own catalog: grounded in the store's data and shown as product cards.
• An honest “I don't know” starts with what you do have — when a visitor asks about a detail your catalog doesn't record for a product, the reply now opens with the product itself before saying which detail is missing.
• Stock is shown as available or not — Shopify searches no longer ask for an exact stock count, so they work for every store regardless of the access it granted.

Fixed

• “How many products do you have?” gets a number — your assistant now answers catalog-count questions with the actual total instead of saying it couldn't find what you were looking for.
• No invented descriptions — for an item whose only details are a name and a price, your assistant shows exactly that and no longer fills in ingredients, benefits or other details it was never given.
• No product pitch on top of a caution — when your assistant advises a visitor to consult a professional, it no longer places a product card above that advice.
• WhatsApp messages on a Business-app number now arrive — a number connected by keeping it on your WhatsApp Business app could have its incoming messages silently dropped. They now reach your assistant.
• Shopify searches no longer fail on stores without inventory access — a store that hadn't granted stock-level access returned no products at all.

What's New

Changed

• More accurate answers about your catalog — when an item has no description, your assistant now gives its name and price only, rather than filling in details it doesn't have.
• A better read on what a visitor is asking — mentioning a personal circumstance mid-conversation, like an allergy or a budget, now gets an answer instead of switching into product recommendations.
• Product cards appear when you'd expect them — when a visitor asks to see or buy something, not when they ask a general question.
• Clearer Data Form editing — Console now says exactly what a refused save was refused for, and mentions the question-key freeze once rather than on every question.

Fixed

• First answers on a new form no longer fail — submitting the first answer could return an error; it now saves correctly.
• Multi-page forms continue after contact details — submitting your contact details no longer dropped the next page of a form.
• A first-time visitor's answer counts — an answer given before a conversation has started is now recognised by a session-start form.

Breaking Changes

• Note for operators: this release changes what a Console sign-in token can carry. Once deployed, it should not be rolled back on its own — doing so can briefly sign Console users out until the rollback is complete. If a rollback is ever needed, roll it back together with the deployment layer that mints Console sessions.

What's New

Changed

• One report per conversation — if you already have an open report on a conversation, filing again simply opens the one you have instead of creating a duplicate. Previously it was easy to file the same issue twice without realizing.

Fixed

• The chatbot no longer gets stuck asking for your location — when you asked for your nearest store or branch, the chatbot could show you a location box that it then refused to accept, telling you the form was no longer active and asking again, sometimes several times in a row, with no way out except ending the chat. The location box now works the first time.
• The conversation screen is readable on a phone again — reading a conversation on a small screen had become cramped: the date broke across several lines, the visitor's name was pushed off the edge, and the buttons ran past the side of the screen. Report and Export now sit on their own row, and everything else has room to read.

What's New

Fixed

• Invented prices and phone numbers — a reply could state a price or a hotline number that appears nowhere in your knowledge base. Your chatbot now checks any figure it is about to quote and won't state one it cannot support.
• A blank screen when switching accounts — the Console could fail to load after signing in or moving between accounts.

Changed

• Branch questions with a named city — asking about a city where you have no outlet now gets your nearest outlets, not a request for the visitor's location.
• "Is there another branch?" with a known location — the chatbot now looks further afield (up to 100 km) before saying nothing is nearby, keeps any branch-type filter the visitor asked for, and says so plainly when none is within range.

Fixed

• A policy question could get a "couldn't find that" reply — when the chatbot re-checked an answer, its second attempt could end with nothing to say. It now rewrites from what it already found, so the answer is given.
• Branch lists no longer appear during a re-check — the re-check step can no longer run a store search of its own, which is what produced the unrequested outlet lists.
• The location form no longer re-appears right after it is submitted.
• Replies stay in the visitor's language — a foreign brand, wine or place name inside an answer no longer causes the reply to be rewritten, and a conversation in one language no longer switches to a closely related one (for example Italian to Malay) on a shared word.
• Privacy in the re-check step — a visitor's own contact details are never used as reference material when a reply is re-checked.

Fixed

• A branch question with a store-type hint still skipped the location ask — when the search narrowed to "stores" or "clinics" but the visitor had named no place, outlets from an arbitrary area could still be listed. Narrowing by type no longer counts as knowing where the visitor is; the location form is shown first.
• A location the visitor typed long ago was reused without asking — outlets near a previously typed location could be listed on a new visit. A typed location is now confirmed once via the prefilled form before it is used; a location supplied by your site or app is still used directly.

What's New

Changed

• Malay and Indonesian are told apart reliably — the two languages share many words, and a Malay question could be answered in Indonesian or the reverse. Your chatbot now reads the words that only one of them uses and replies in the right one. A question that is genuinely identical in both languages follows the language the conversation is already in.

Fixed

• Replies switching language on a place name — a visitor chatting in English who typed something like "I'm at around Kepong" could receive the whole reply in Malay, because the town name outweighed the sentence around it. A reply now changes language only when the visitor's message is actually written in that language. A genuine switch still takes effect immediately, in every direction.

Fixed

• Branch suggestions for visitors who haven't said where they are — a question like "can I exchange this at another branch?" or "which outlet should I go to?" with no place named could be answered with a store in the wrong city. Your chatbot now shows the location form (or asks for a city on messaging channels) and only names outlets once it knows where the visitor is. A visitor who has already shared a location gets the outlets nearest to it, and a visitor who names a city gets that city's outlets as before.

Fixed

• Instagram replies — a message sent to your connected Instagram account was accepted but dropped before your chatbot saw it, so nothing came back. Instagram conversations now get a reply like Messenger and WhatsApp.

What's New

Changed

• Permission changes apply instantly — changing a teammate's role or set, removing them, or suspending an account takes effect on their very next request. Nobody is asked to sign in again.
• Suspended accounts get a clear screen — a member of a suspended account can still sign in and sees an explanation, instead of a failed login.
• Partner staff leave with the partner — when a partner's membership on a managed account ends, the seats it granted to partner staff end with it.
• Faster sign-in and account switching — signing in and switching between accounts is noticeably quicker, and Console pages respond faster after idle periods.
• Deleting a channel releases its Page or number — removing a Messenger, Instagram or WhatsApp channel lets that Page or number be connected elsewhere, unless another of your channels still uses it.
• New WhatsApp numbers are ready to send immediately — a number onboarded through the guided WhatsApp setup is registered for you; no extra step before the first message.
• No more automatic follow-up message on WhatsApp — Gydr no longer sends a templated nudge when a conversation goes quiet.
• Opted-out WhatsApp recipients fail clearly — a message to someone who has blocked or opted out is marked failed straight away instead of waiting to be retried.

Fixed

• Admins can manage Viewer seats — the row menu on a Viewer teammate was empty for Admins; it now offers the same actions the account allows.
• Publish stays reachable on a saved draft — after saving a widget draft, the Publish action could disappear from the editor bar. It now stays available.
• Facebook Messenger and Instagram connect again — the connect pop-up finished but the Console showed nothing, or reported that the authorisation could not be completed. Both flows now connect end to end.
• Instagram replies on a Page-linked account — replies were rejected for some Instagram accounts linked to a Facebook Page; they now send.
• Channel settings edits keep the rest of your settings — saving one field on a channel no longer clears the others.
• Console header no longer overlaps page content on some screen sizes, and documentation links in the Console open the right page.

What's New

Changed

• Simpler partner types — the partner program now has two types: Managed Service Partner and Technology Partner. Existing partners have been transitioned automatically; no action needed.
• Clearer partner credit management — allocating credits to a managed account now shows a preview of the result, asks for confirmation before moving anything, and explains exactly why a transfer can't proceed.

Fixed

• Form errors now name the field — a rejected save tells you which field was refused and why, instead of a generic “failed to save” message.
• Chat attachments could silently fail — an upload processing error could discard a visitor's attachment without any sign of failure. Uploads now recover reliably.
• Occasional undelivered replies — in rare cases a generated reply failed to reach the visitor's chat window. Delivery is now verified end to end.
• Usage charts could under-report — daily usage on the credit consumption chart could show less than the monthly total under heavy traffic. The gap now closes automatically.
• Partner credit reclaim is precisely bounded — a partner reclaiming credits from a managed account can only reclaim what they originally granted.
• Integration setup no longer flashes the wrong form — a connect step now shows a loading indicator while its configuration is being fetched.

Fixed

• Replies stay in your visitor's language — in rare cases a conversation could receive an answer in a language nobody in the chat had used. Your assistant now consistently answers in the language your visitor is actually writing in.
• Consistent quality across every language we support — answers in languages such as Hindi, Bengali, Tamil, Hebrew, Greek, Burmese, Khmer and Amharic are now treated exactly like every other supported language, so they arrive as written instead of occasionally being reworded or slipping into English.
• Language settings on your knowledge base are always recognised — if you record your knowledge base's content language using a short two-letter code, that setting is now read correctly.

What's New

Changed

• Consistent controls everywhere — every filter bar, action cluster, and button in the Console follows one layout grammar and one control height, on every screen size, with buttons sized correctly on mobile.
• Usage in your own calendar — the credit chart buckets by your business day, so daily and monthly totals agree and a full day means a full day.
• Usage History loads as one view — both of its data sources are read together instead of in sequence, ending the timeouts on wide date ranges.
• Accurate dashboard figures — visitor counts in the language panel and handoff rates are now measured against one consistent population, and every series is bucketed in your own calendar.
• Bounded dashboard ranges — the date range is capped to a span that returns promptly rather than appearing to hang.
• Import errors that point at the problem — a rejected knowledge CSV names the row and column that failed.
• Validation errors that name the field — a rejected form says which field and why, instead of "Failed to save."
• Complete audit log pages — a filtered audit log fills the page instead of truncating it.
• Channel deletion is clean — removing a channel detaches it from your live widget and from your saved configurations, rather than leaving a broken reference behind.
• Configuration screens match — channels and plugins share one configuration layout and a real loading state, and a field offering a fixed set of choices renders as a choice rather than a masked secret.

Fixed

• Asking for a person no longer ends the chat — a request for a human hands the conversation to your team and keeps it open; only a genuine goodbye closes it. Saying goodbye no longer opens a ticket, either.
• A handover never reads as a hang-up — a conversation being transferred no longer says goodbye on its way to your agent.
• You are always told what happened — every handover outcome is announced to the customer, and if it cannot go through they are given your alternative contacts rather than silence.
• No promises we cannot keep — a widget with no help desk connected never offers to fetch a human.
• The conversation belongs to your agent — once a person is handling it, nothing else can end it out from under them, and closing it on your side ends the session properly and tells the customer.
• A customer message during handover always reaches someone — a message sent while transferring is never quietly dropped.
• Similarly-named products — questions about one product in a family of near-identical names now resolve to the right one, or ask a short clarifying question instead of guessing.
• Nearby-location answers — "what is near me" questions complete in one turn, name the right places, and the locations shown always match the ones named in the reply.
• Product cards match the message — the cards below a reply never show more, fewer, or different products than the reply itself names.
• Fewer irrelevant results — weak matches are no longer presented as though they were answers, and searches with filters applied still tolerate a reasonable misspelling.
• The assistant only offers what it actually has — it no longer mentions capabilities that were withdrawn for that turn, and the response settings configured on a widget now reach the conversation.
• No repeated buttons on messaging channels — a button from a finished conversation can no longer restart it, and tapping one no longer produces a fresh copy of itself.
• Taps are shown and understood — a tapped option appears in the conversation as what the customer chose, and a form is never dumped as raw data.
• One turn at a time, in order — messages arriving in quick succession are answered in the order they were sent, once each, and a platform re-delivering the same message no longer produces a second reply.
• Reactions are not messages — an emoji reaction no longer triggers a reply.
• The assistant collects what it needs itself — when a detail is required before your team can be brought in, the assistant asks for it rather than blaming your agents for being unavailable.
• Your customer's name is filled from their platform profile, so your team sees who they are talking to.
• A reply already paid for is never discarded — transient platform failures are retried rather than thrown away.
• Usage figures repair themselves — gaps between usage records are detected and repaired, so your chart and your bill agree.
• Rejections come back as rejections — an invalid request returns a clear client error with the field named, rather than a generic failure.

What's New

Changed

• Visitors are ranked by their last message, not their last visit — someone who just wrote to you now sits above someone who merely reopened the chat window. The column is now labelled "Last Activity."
• Only real conversations appear in Visitors — opening the chat widget without typing anything no longer creates an entry, so the list reflects people you actually talked to.
• The "Referrer" field is gone from visitor details — it could not be captured reliably and showed a value that misdescribed where the visitor came from. Nothing replaces it.
• In-chat forms use the whole width available — questions sit beside their answers, prompts line up with the choices next to them, and the layout no longer changes based on how many options a question happens to have.
• Enter moves you through a form — visitors can complete each step from the keyboard. Confirmations that end a conversation deliberately don't respond to Enter, and put the emphasis on going back rather than on ending.
• Forms show a receipt after submitting — instead of a single line of acknowledgement, the visitor sees what they sent.
• Handing over to a person explains itself — the confirmation now tells the visitor their recent messages travel with them, and the transfer options show the name you gave each team rather than an internal label.

Fixed

• Half-star ratings can be submitted — picking 3.5 stars is now accepted rather than rejected.
• The browser tab icon follows the widget you're viewing — switching between widgets no longer leaves the previous one's logo on the tab, and a widget without a logo no longer inherits another's.
• Typing in Chinese, Japanese or Korean no longer sends half a word — pressing Enter to confirm a character no longer sends the message mid-composition, in both the chat box and in forms.
• Long words in a form question stay readable — an order reference or a long link in a question is no longer cut off past the edge of the panel.
• A question isn't shown twice — a single-question form no longer repeats the same sentence as both its heading and its question.

What's New

Changed

• Dynamic forms use the full width of the chat — questions, choices, sliders and ratings now lay out for the space they actually have, instead of being squeezed into a narrow column.
• Faster replies — answers arrive quicker, most noticeably on questions that draw heavily on your knowledge base.
• More accessible settings switches — switch rows are now a full tap target and read their description to screen readers.

Fixed

• The product carousel no longer contradicts the reply — when your assistant says it can't find a match, it no longer shows a row of unrelated products underneath. What appears is what the reply actually recommends.
• Answers stay honest when the match is weak — rather than presenting a loosely related item as though it answered the question, your assistant now says it doesn't have that information.
• Disclaimer settings survive a save — the switches keep their state when you save and publish.
• Forms no longer re-ask what you've already answered — including when the follow-up is worded differently.
• Form answers now pull from your knowledge base — answering a form question that needs information from your content now retrieves it.
• Replies in non-Latin scripts are no longer dropped — answers in Chinese, Malay and other non-Latin scripts are delivered correctly.
• OpenAPI import no longer leaves a blank first endpoint — the seeded row now fills in properly.

What's New

Changed

• Richer answers — your chatbot now shows as many relevant results as the question genuinely calls for, instead of settling on a fixed few. Ask about a category with many matches and you'll see more of them.
• Up to 10 product recommendations — product answers now surface up to ten items, and the recommendation panel always shows exactly the products the reply talks about — no extras, none missing.
• Larger REST API connections — a REST API connection now supports up to 100 endpoints and a bigger configuration, so you can bring more of your service into one connection.

Fixed

• Widget domain restrictions apply correctly — if you've restricted your widget to specific domains, that restriction is now enforced against the page the widget is embedded on. Clearing the list restores unrestricted access as expected.
• Recommendation panel matches the reply — the panel no longer shows products the reply never mentioned, or leaves out one it did.
• Accurate API responses — when a request is rejected, the API now returns the matching error status instead of a success response carrying an error inside it.

What's New

Changed

• A refreshed Gydr logo — new brand mark across the website, console, and emails.
• Handoff is offered only when someone can answer — If your help desk integration reports no agents online, the chatbot no longer shows a connect button. Instead it shares the alternative contacts you configured, so your visitors are never handed a button that leads nowhere.
• Clearer import results — Items that could not be saved during a website import are now reported back to you instead of silently disappearing from the count.
• Failed imports keep their work — When part of an import cannot be recovered, the extracted results are retained rather than cleared straight away.
• Consistent Integration API usage recording — Read requests to the Integration API now record credit usage in line with the documented per-request billing. Previously some read requests reported usage that was never actually recorded.
• More useful API error responses — Conversation API failures now come back with a specific status and message naming what went wrong, instead of a generic error.

Fixed

• Public API requests were rejected — API keys were being turned away with an authorization error on the Knowledge and Conversation endpoints. Fixed.
• A single bad page could destroy an entire website import — One document that failed validation would take the whole crawl down with it. Imports now save everything they successfully extracted.
• Model reasoning could appear in chatbot replies — When running on your own model keys, some models' internal reasoning text could reach visitors. It is now removed in every form we have seen.
• Rating a conversation could fail with an unexplained error — Rating a conversation that had not yet ended now returns a clear message, and is never charged for.
• Switching accounts could get stuck — An unsaved-changes prompt could trap you on the previous account's page. Account switching now completes cleanly.
• Emails showed a broken logo image — The header image in Gydr-branded emails failed to load. Fixed.
• Test page buttons behaved unexpectedly — Clicking a connect-to-human or end-session button on the widget test page no longer sends a stray message or spends credit; it explains what would happen for a real visitor instead.
• Knowledge Base icon alignment — The icon sat off-centre on the collapsed console sidebar.

Fixed

• Custom domain certificate errors now name the real cause — When a certificate couldn't be issued for your domain, the setup screen blamed your DNS record even when it was correct. It now identifies the actual reason and tells you exactly which record to add to fix it.
• Removing a custom domain — Removing a domain could fail with a generic "unexpected error" and leave it in place — the very step you're asked to take before trying again. Removal now completes, and on the rare occasion something does go wrong, the message explains what.
• Category filters no longer hide FAQ answers — Filtering your FAQs by category could exclude valid answers instead of narrowing to them. Filters now focus your results without dropping matching answers.
• Clearer answers when your content is close but not exact — When your knowledge base held something related but not the specific detail a visitor asked for, your chatbot could imply the related item was the answer. It now says plainly when the exact detail isn't there.

What's New

Changed

• The Knowledge Base, rebuilt — Knowledge is now split into Library and Records so you can tell your knowledge bases apart from what's inside them at a glance. Content editors open in a wide side panel instead of a cramped modal, with field markers showing which fields your chatbot actually reads.
• Form answers now appear in the transcript — When a visitor answers an in-chat form, their answer shows as a proper message bubble in the conversation, in live chat and in history alike.
• Clearer team roles — The Team settings page now explains what each role can do from a single source, so what you read is always what's enforced.
• Filter visitors by widget — The Visitors inbox can be narrowed to a single widget.
• Rewritten documentation — The documentation site has been rebuilt end to end: corrected facts, cleaner navigation, working in-page links, and credit costs collected in one place.
• Every mutating button now shows progress — Buttons that save, delete, or publish show a real in-flight state so you're never left wondering whether a click registered.

Fixed

• Widgets that wouldn't open — Widgets whose skills referenced a plugin could fail to load with a "Widget not found" error. They now open normally.
• The mobile chat panel — On phones, opening the chat could scroll the page underneath, expose the host page behind the panel, or leave the panel misaligned after the keyboard appeared. The panel now stays put and the page behind it stays locked.
• More reliable handoffs to a human — Requests to speak with a person now consistently produce a handoff, and stale handoffs no longer linger on a conversation.
• More accurate answers when information is missing — Your chatbot is better at saying it doesn't know rather than filling the gap, and it no longer discards a correct answer that was already grounded in your content.
• Console focus and scrolling — Menus now move focus to the right place, the space bar no longer scrolls the page from an image uploader, and navigating between pages no longer leaves the page unable to scroll.

What's New

Changed

• Per-skill resource lanes in Widget Studio — Each attached skill now gets its own lane holding exactly its knowledge bases and plugins, instead of one shared column of your whole catalog. Shared resources are shown in every lane that uses them, and lanes fold to keep large setups readable.
• Safer editing in Widget Studio — The canvas now edits only the widget you're looking at. Account-wide resources (skills' knowledge wiring, plugins, and more) are shown read-only, with each panel linking to the page that owns it — so a change in one widget's Studio can no longer affect your other widgets.
• Clearer message flow on the canvas — The conversation loop now enters and exits the core at fixed points, with a tidier layout for the visitor, channel, and response paths.

Fixed

• Knowledge writes through the Integration API — Creating or updating knowledge items with an API key could fail with a server error. Writes now complete reliably, and validation errors name the exact field that needs fixing.
• Widget Studio layout drift — Stale hand-dragged positions are discarded when the canvas layout is updated, so the diagram always renders its intended shape.

Changed

• Hardened widget security — Strengthened the protections around the embedded widget's handling of visitor information and custom branding assets, backed by an expanded automated test suite that guards every supported embedding method.

Fixed

• Visitor details preserved exactly — Visitor names, emails, phone numbers, and company names containing spaces or hyphens were being altered before reaching your conversations (for example, a two-word name arriving joined together). Details you pass to the widget now come through exactly as entered.
• Widget recovers from network hiccups — A brief connection blip while your page loaded could previously leave the chat widget missing until the visitor refreshed. The widget now retries automatically and loads normally.
• No more endless loading spinner — If the chat genuinely can't load (for example, a restrictive site configuration), visitors now see a clear "Chat is currently unavailable" message instead of a spinner that never resolves.
• Chat loads in page-builder previews — The widget could get stuck loading when your page ran inside certain preview or sandboxed environments (such as site-builder preview panes). It now starts correctly there too.
• Custom launcher icons fall back gracefully — An invalid custom icon or an image that fails to load now shows the default chat icon instead of a broken graphic.

Fixed

• Embedded chatbot not appearing — Websites embedding the chatbot with the standard script snippet (the tag with your API key) could fail to show the chat bubble or chat window. The snippet now activates reliably again. Integrations that start the chatbot from their own code were not affected.
• Unnecessary download prompt in the Console on mobile — On some mobile browsers, opening the Console could trigger a small, harmless file-download prompt that overlapped the support chat bubble. Payment-security scripts now load only on billing pages, removing the prompt everywhere else while keeping payment fraud protection fully intact.

Fixed

• Embedded widget on older Safari and iOS — On Safari and iOS versions below 16.4, the embedded chat widget could fail to load and appear blank after the v2.5.1 update. The widget now loads correctly on these browsers.
• More accurate replies to combined requests — When a visitor's message combines a question with an action in the same turn, the assistant now verifies its answer more reliably instead of over-correcting it.
• Account switching visuals — When switching between accounts in the Console, the loading overlay now shows the look of the account you're switching to, not the one you're leaving.

Fixed

• Forms no longer arrive with the answer attached — When your chatbot asks a question with a form (a clarifying choice, a location confirmation, a booking detail), it now waits for the visitor's response instead of sometimes answering in the same message. Asking and answering are now cleanly separated turns.
• Product prices follow the visitor's locale — Prices shown in chat now format according to each visitor's own regional settings.
• Cleaner replies — Removed rare cases where internal notation could appear inside a chat reply.

What's New

Changed

• Redesigned knowledge editing — The forms for adding and editing products, FAQs, documents, and locations are now multi-column workbenches, making it faster and easier to fill in and review your knowledge base content.

Fixed

• Global outlet questions answer correctly — When a visitor asks a general question about all your locations (like which outlet has a particular feature), your chatbot no longer narrows the answer to a single saved location.
• Deleted knowledge is fully removed — Items you delete from your knowledge base, including in bulk, are now cleared from search right away, so removed content no longer appears in answers.
• More accurate, better-grounded answers — A round of answer-quality improvements: your chatbot stays closer to your knowledge base, matches product requests more precisely, and won't add unsolicited disclaimers.

What's New

Changed

• Improved Credit Efficiency — Your credits now go further, with more chat messages and AI-powered responses delivered per credit.
• Clearer Payment Summaries — Plan confirmation and upgrade modals now show a more detailed breakdown of what you'll be charged and when.

Fixed

• Partner Pricing Display — Partner discount pricing now shows correctly only on the primary account, with centered discount cards and banners, and updated discount text across all supported languages.
• Custom Plan Banner — Custom plans now appear as a hero banner above the standard tier options, and the "current plan" badge updates correctly after switching plans.
• Mid-Cycle Plan Changes — Proration during mid-cycle plan changes now uses invoice-level adjustments for accurate billing.

What's New

Fixed

• More accurate chat responses — Your chatbot now correctly declines with "I don't have that information" when it genuinely cannot find an answer, instead of guessing or making something up. It no longer claims your knowledge base is missing information when it actually has it. And it stays focused on your content — it won't volunteer unsolicited professional-disclaimer boilerplate unless your own knowledge base already includes that language.
• Product editing: clearer errors and unblocked legacy items — When saving a product fails, you now see exactly which field needs attention and what the character limit is, rather than a generic error. Products written before character limits were enforced can now be edited again without being blocked by their own legacy content.
• Plugin configurations restored — Some plugin configurations that were set up before early July appeared to vanish silently. They have been recovered and are now visible again in your plugin settings.
• Form behavior improvements — When your chatbot creates more questions than fits in one form, the overflow now carries over to the next form instead of being lost. Earlier answers no longer interfere with the chatbot authoring the next form's questions.

Breaking Changes

• BYOM is now Growth+ — Bring-your-own-model is available on Growth+ plans and above. If your account is on a Starter plan or below using BYOM, you will need to upgrade to continue using this feature. BYOM access stops automatically on downgrade, and your chatbot will continue answering using Gydr's built-in model.

What's New

Changed

• Goodbyes recognised in any language — when a visitor says they're done, the end-of-chat confirmation now appears regardless of the language they're chatting in.

Fixed

• Replies match your visitor's language — your chatbot no longer drifts into the language your knowledge base happens to be written in. A visitor who writes in Malay gets a reply in Malay, even when your content is in English.
• No more stray conversation labels — replies occasionally began with leftover formatting like a speaker label. That's gone, in every language.
• Renaming a chatbot stays in your draft — editing a saved configuration's name or description no longer changes your live chatbot. Your published chatbot keeps its name until you publish the change.
• Access restored correctly after a payment issue — custom-plan accounts now regain their full credits and features as soon as payment is resolved.
• Partner-managed accounts stay active — accounts remain usable after a partner is removed from them.
• Cancelling a custom plan works on every base plan — cancellation no longer gets stuck for some accounts.

What's New

Changed

• More accurate answers across your whole knowledge base — your chatbot now applies a consistent relevance standard to every knowledge type (products, FAQs, documents, and locations), so visitors get stronger, on-topic answers and fewer loosely-related results slip through.

Fixed

• Multilingual greetings routed correctly — fixed an issue where an opening greeting in some languages could be misrouted, so conversations now start smoothly regardless of the visitor's language.
• Steadier answer generation — resolved a rare case that could interrupt how the chatbot assembled a response, making replies more consistently reliable.

What's New

Changed

• Faster chatbot replies — conversations feel noticeably quicker, especially during back-and-forth exchanges.
• Redesigned widget editor — a clearer two-row layout separates your live chatbot from your in-progress edits, so you always know exactly what your visitors are seeing.

Fixed

• Test widget no longer gets stuck — the preview no longer hangs on "Thinking…" when a test message fails, so you can keep iterating.
• More reliable in-chat forms — forms your chatbot presents mid-conversation now recover cleanly, so visitors aren't left on a stalled step.
• Better answers across languages — your chatbot more reliably draws on your knowledge base when a visitor writes in a different language than your content.
• Steadier connections — the chatbot recovers gracefully from brief connectivity blips during a conversation.

What's New

Changed

• Test with your real integrations active — test conversations now run your connected plugins and integrations, so what you see matches what your visitors will experience.
• Faster lists and analytics — dashboard list and analytics views load noticeably quicker.
• Polished test surface — a cleaner, more usable layout across the widget test page, including a scrollable debug panel for longer conversations.

Fixed

• Personality applied more reliably — your chatbot now follows its configured personality more consistently.
• More relevant product suggestions — the chatbot is more selective about when it surfaces products, so suggestions stay on-topic.
• Better multilingual handling — improved language detection for conversations that mix Chinese, Japanese, or Korean characters with Latin text.

What's New

Changed

• Skill instructions followed as written — The instructions you give each skill are now applied exactly as you wrote them, so your chatbot behaves more predictably and stays on-message.

Fixed

• More reliable knowledge-base answers — Your chatbot now consistently draws on your FAQs, documents, and other knowledge whenever it's helping a customer, so fewer relevant answers slip through.
• Exact details across languages — When your chatbot replies in a different language than your knowledge base, technical details like prices, codes, and specifications now stay precise and unchanged.
• Product links preserved — Links to your products now reliably appear in your chatbot's answers.

What's New

Changed

• "Plugins" is the new home for your integrations — what used to be called MCP Tools is now Plugins, with redesigned, easier-to-navigate setup pages and brand logos for each connector. Your existing configuration carries over.
• More features on Starter — Bring Your Own Model and Plugins are now available starting on the Starter plan, and human-handoff integrations are now a Starter-and-above feature.
• More accurate, on-topic answers — the assistant now answers strictly from your knowledge base rather than general training, and declines honestly when the answer genuinely isn't there instead of guessing.
• Answers in your visitor's language — replies follow the visitor's language throughout a conversation, including switches mid-chat and questions asked in a different language than your content.
• Recognizes your own services — if your business sells technical or specialized services, the assistant now treats questions about them as in-scope and answers from your knowledge base instead of refusing.
• Cleaner catalog browsing — a broad "what do you have?" now lists your products cleanly instead of an awkward "no matches found" hedge.
• A cleaner chat experience — a refined thinking indicator while the assistant works, and a message composer that adapts between touch and desktop.
• Smarter website crawling — the crawler now renders modern, JavaScript-heavy pages before importing and keeps an item even when one field runs long, so more of your site's content makes it into your knowledge base.
• Personality builder polish — the Make-your-own personality flow is clearer, previews each style as a sample conversation, and validates before you save.
• Knowledge-base table refinements — column-aligned rows, per-type color accents, and save confirmations.
• Stronger compliance audit logging — broader, more consistent security audit records across account and billing actions.

Fixed

• Paid plans no longer get stuck on Free — fixed a payment-processing race that could leave a paid subscription showing as Free after an out-of-app or interrupted payment; affected accounts now self-heal automatically.
• Payments needing extra verification now complete — repaired checkout for cards that require an additional verification step from the bank.
• Custom personalities apply reliably — a saved custom personality (Growth and above) now correctly shapes your live chatbot's replies.
• Accurate billing allocations — fixed monthly credit-allocation drift for partner-managed accounts.
• Forms behave reliably — location requests, clarifying questions, and end-of-chat prompts now appear and submit dependably, never loop, and always offer a way out.
• Location is remembered — a visitor whose location is already known is no longer re-asked on every nearby-search.
• Messages arrive complete — the chatbox now waits for the full response, so the "thinking" indicator never gets stuck.
• Your content survives to the answer — longer product, FAQ, and document fields are no longer trimmed before the assistant uses them.

Fixed

• Your product AI Context Notes are honored — The AI Context Notes you add to products (usage tips, consumption guidance, caveats) are now consistently reflected in your chatbot's answers, so it follows your guidance instead of overriding it with generic information.
• Replies stay in your visitor's language — The chatbot no longer switches to the wrong language partway through a conversation; replies now consistently match the language your visitor is using.
• More reliable nearby-store results — When a visitor shares their location to find the closest store, the chatbot now dependably returns nearby stores instead of asking for their location again.
• Accurate store availability — The chatbot no longer incorrectly says a business has no physical stores (or is online-only) when store locations are available; it now surfaces the real locations.

What's New

Changed

• Tidier widgets list with quick actions — Each widget now shows its agent avatar at a glance, and a new per-row menu puts common actions a single click away.
• Calmer end-of-chat for visitors — When a conversation times out, your chat widget now shows a single, clear "conversation ended" state right where your visitor was typing, with a one-tap way to start a fresh conversation.
• Interface polish — A round of accessibility, motion, and visual refinements across the console.

What's New

Changed

• Smoother sign-up and sign-in — Creating an account and signing in is now more reliable. If a sign-up gets interrupted partway through, the system automatically picks up where you left off the next time you return, so you're no longer left stuck mid-process.
• Refreshed Audit Log — The Audit Log has a cleaner, easier-to-scan layout with clearer event details, making it simpler to review account activity at a glance.

Fixed

• Audit Log filtering — Filtering the Audit Log by action now works correctly and returns the matching events instead of coming up empty.
• External links in chat — Links shared in the chat window now open properly in a new tab instead of failing to load.

Changed

• Consistent slide-out panels — every slide-out panel now opens at the same comfortable width, so editing your widgets, skills, and knowledge feels uniform across the console.
• Easier knowledge base navigation — the "Add" button now sits right where you expect it: pinned to the right on desktop and full-width on mobile for an easy tap.
• Better mobile layouts — console pages now fill your screen properly on mobile, with cleaner, more consistent spacing throughout.
• Automatic cleanup of abandoned invitations — team invitations that are never accepted are now tidied up automatically, freeing that email address to be invited again later.

Fixed

• Reliable team invitations — inviting a teammate no longer fails partway or leaves an invite stuck in limbo; invitations now go through cleanly, every time.
• No more layout shift — opening a panel or overlay no longer nudges the navigation bar as the scrollbar appears and disappears.
• Cleaner crawl-job cards on mobile — website crawl-job details now wrap and reflow neatly on smaller screens.

Changed

• More natural responses to multi-topic and location questions — When a visitor asks about several things at once, or follows up a product question with "where can I find this?", the bot now handles the conversation more smoothly without unnecessary clarifying questions or incorrect off-topic refusals.
• Better language matching — When a visitor writes primarily in Chinese, Malay, or another non-Latin script, the bot now replies in that language even if their message mixes in a few English words.
• Faster responses on single-topic questions — Straightforward queries no longer trigger extra processing, so simple chats feel snappier and use fewer credits per turn.
• Complete UI translations — Console settings and the homepage chat demo now display correctly in all supported languages (English, Malay, and Chinese).

Fixed

• Website Import recovers from temporary failures — The import status page used to get stuck on "Import job not found" after a transient hiccup and stop checking for progress. It now self-recovers and shows clear, friendly messages instead of raw provider error text. Large website imports also run with a more conservative concurrency to avoid hitting capacity limits.
• Human handoff replies preserved — When the bot transfers a conversation to your team, its response message is now kept intact instead of being replaced by a generic refusal.
• "Where can I get these?" correctly prompts for a location — Follow-up proximity questions after a product result now ask the visitor to share their location instead of replying that the topic is out of scope.
• More consistent product and FAQ answers — Resolved several edge cases where the bot repeated the same search multiple times or included referral suggestions that were not grounded in your knowledge base.

What's New

Changed

• A completely redesigned Console — a faster, enterprise-grade interface that's now fully mobile-friendly, with refreshed navigation and consistent, scannable list views across Knowledge Base, Skills, Widgets, and Integrations.
• A reworked Visitors inbox — visitor conversations are now a clean master-detail inbox, with uploaded attachments shown inline.
• Easier knowledge management — click any item to edit it, sync and select items right from the list, and get clear, plain-language messages when an item fails to sync.
• A streamlined widget editor — widget settings are now grouped into clear tabs, each saved independently so you only update what you change.
• A smarter website crawler — improved extraction of products, FAQs, documents, and store locations from your website.
• Configurable low-credit warnings — set the threshold for your credit-low banner in Settings.
• Greeting screen extras — add a line of additional information with clickable links to your widget's greeting.
• Refreshed Account and Profile settings, plus an updated privacy and cookie-consent experience aligned with the latest data-protection standards.
• Partner-managed accounts can now configure their own credit alerts and use the personality sandbox and other inherited plan features directly.
• Tighter knowledge isolation — each widget now draws only on the knowledge bases and skills you've assigned to it, enforced end to end.

Fixed

• Cleaner account switching — switching between accounts now fully refreshes your view, so you never see data from the account you just left.
• Product suggestions match the reply — product cards shown alongside a chat response now line up with what the chatbot actually recommended.
• More accurate location answers — the chatbot no longer says there are no stores when your locations can answer, and it now handles outlets that don't have a type set.
• Correct message order on reconnect — chat messages stay in the right order and no longer duplicate or shuffle when a conversation reconnects or a session ends.
• Sharper, more relevant answers — when there's no strong match, the chatbot no longer pads its reply with loosely-related items.
• No more accidental sends — the chat widget now blocks a message from being sent after a session has timed out, and shows immediate feedback while a file uploads so a slow connection never looks broken.
• Steadier reconnections — if the chat reconnects, your own messages no longer appear twice.
• Overlay polish — dialogs, slide-out panels, and their backdrops now layer and cover the screen correctly across the Console.
• Greeting reliability — your greeting's extra info is now saved correctly and the greeting shows even when no FAQs are configured.
• Clearer error messages — Settings flows now show errors right inside the dialog instead of failing silently.

Breaking Changes

• Facebook Messenger output format — for customers using the Integration API with Facebook Messenger formatting, the carousel response now uses an elements array (previously cards), and each quick reply includes a content_type field, to match Facebook's current Send API. Update any code that reads the carousel response.
• Usage credits in the API response — the Integration API now reports credit usage broken into multiple categories (chat, retrieval, orchestration, tools, and crawler) instead of a single combined value. If your integration reads the credits field from the API response, update it to the new breakdown.

What's New

Changed

• Refreshed console navigation — A new left-edge sidebar stays compact as an icon rail and expands when you need it, with cleaner styling across the console and the Gydr logo in one consistent spot.
• More reliable chat product suggestions — Product cards shown alongside chat replies now appear consistently and are filtered to the closest matches, so visitors see fewer loosely-related items.
• More dependable chat session endings — Ending a chat now confirms before closing, so sessions wrap up cleanly every time.

Fixed

• Accurate "how many" answers — The chatbot no longer under-counts when a visitor asks about the size of your catalog or your number of stores.
• No more premature "I couldn't find that" — When your knowledge base has the answer, the chatbot leads with it instead of opening with a denial.
• Cleaner chat replies — Removed occasional stray formatting characters from chatbot responses.
• Smoother console on mobile — Fixed touch scrolling, pull-to-refresh, and a scroll-lock issue that could leave the page stuck.
• Slide-in panels open cleanly — Configuration panels now slide in smoothly, without bouncing or layering glitches.
• Chat session ending no longer freezes — Resolved a case where ending a session could stall.

What's New

Changed

• Better follow-up handling — Your chatbot now connects multi-step questions more reliably. Visitors who refer back to products or topics from earlier in the chat get coherent answers instead of generic ones.
• Answers your visitors' questions instead of deflecting — Your chatbot no longer redirects visitors to seek outside help when their question is one your knowledge base can answer. Disclaimers only appear if you've explicitly written them into your knowledge base or skill prompts.
• Smarter suggestions for higher-budget shoppers — When a visitor indicates they're shopping in a higher price range, your chatbot now leads with premium options instead of starting from the cheapest.
• Cleaner category handling — Category names that differ only in capitalization are now treated as the same category. No more split lists or confused follow-ups when your catalog has minor inconsistencies.
• Conversation API responses include a new format field — Every non-error response now tells you which response shape produced it. If your code validates response shapes strictly, add format to your parser.
• Developer tab: Response Content Format dropdown — Pick how each API key formats its Conversation API responses. Available when the Conversations: Chat scope is enabled, with a new Format column on your API keys table so you can see which key delivers which shape.
• Richer location answers across channels — Phone numbers, map links, and websites you've configured on your outlets now show up as tappable buttons on Facebook Messenger cards and as native location messages on WhatsApp.

Fixed

• Widget no longer crashes on reconnect — Reopening the chat or retrying after a credit-exhausted message used to throw an error. Visitors can now refresh and resume cleanly.
• Smoother first-message experience after idle periods — If your chatbot was unused for a long stretch, the first message back occasionally surfaced as a timeout error. That window is now significantly smaller — visitors get a response instead of a failure.

What's New

Changed

• Polished chatbox interactions — The greeting-to-chat transition uses a smooth curtain animation. Input text is vertically centered. Footer spacing is precise. Hero content scrolls on desktop. End Session moved up to the header menu for less clutter.
• Widget Theme and White Label are now separate — No more confusing overlap between how your widget looks and how your Console is branded. Two distinct features, cleanly handled.
• Console feels faster — We cut unnecessary network requests from account switching and trimmed overhead behind the scenes. The dashboard is lean and responsive.
• Smarter embed for your sites — The widget now loads with stricter security guarantees, and loading errors get reported back to your page so they're easier to debug.

Fixed

• Sessions finally stay clean — Stale ended sessions can no longer corrupt new ones. We've tightened how sessions expire so they reliably close instead of lingering.
• Handoffs work every time — No more duplicate handoffs per session. Visitor profiles sync to integrated help desks once, in the right order. Close-ticket notifications always fire, even when sessions end unexpectedly.
• Better AI responses — Your chatbot no longer sends empty replies. Period. Internal references to your knowledge sources no longer slip into the answer text. Product and location info stays consistent. Off-topic questions get a graceful "I can't help with that" instead of a confused guess.
• Billing and credits, sorted — Credit reconciliation works correctly across all tiers. Failed payments retry automatically. Purchased credit packs reconcile properly. Partner accounts always use the partner formula.
• Chatbox polish — Mobile typing no longer zooms in. Disabled input fields don't steal focus. The copy button plays nice with other bubbles. The greeting screen won't auto-navigate away before you're ready. Account switcher is perfectly centered.
• Help desk integration improvements — Visitor profiles sync once, in the right order. Ticket timestamps are consistently formatted.

Breaking Changes

• New chatbot response engine — Your chatbot now runs on a rebuilt response engine that delivers smarter, streaming replies. If you built custom integrations against the previous response engine, you may need to update them. The Conversation API surface is unchanged.
• Custom chatbox URL override is deprecated — The chatboxUrl override on the widget embed snippet is no longer supported. We've tightened how the chatbox loads to reduce the risk of compromised embeds. Remove the override from your embed code.