Drafts now find more of the useful help articles for a question. ReplyRabbit combines keyword and meaning-based search more carefully, keeps related articles together, and carries several relevant passages from each source into the draft.
If an AI provider cuts off a draft or image description at its output limit, ReplyRabbit now records the cause clearly instead of failing silently or treating an incomplete description as finished.
On phones, Recent conversations and Similar conversations now share a compact switcher, which leaves more room for the conversation itself while keeping both lists close by.
Fixed a setting that made ReplyRabbit hold on to search data for about 60 days instead of the intended hour. Nothing was broken by it, but your searches now pick up changes to your docs and knowledge base much sooner.
Suggested docs in the Next actions checklist are much more relevant. Before, a ticket could pick up an article just because it shared a common word with the customer’s message, like "card" or "credit". Now an article has to actually be about the subject, and if nothing fits, nothing is suggested.
Smart search over your docs had quietly stopped running once a docs site grew past a certain size, falling back to plain keyword matching without telling you. It now handles far larger docs sites, and if it ever does hit the limit, the settings page says so instead of staying silent.
The Magento connector arrow now lines up with the other conversation sidebar sections on phones, and dark mode no longer shows bright gaps between those sections.
New Effort setting for drafts. Pick how hard your AI model thinks, from Low to Very high, or leave it on the provider default. Higher effort means slower, more expensive, more careful drafts.
When your AI account runs out of credits or rejects its key, the message now names the provider and where to fix it, and the settings page has a Top up credits link for it.
Opening a long ticket no longer blocks the Draft reply button. The issue-tracker panel now checks your connected tracker once per ticket instead of probing every tracker for every note.
If you do hit a limit, the button now says so and asks you to try again shortly, instead of a blank failure.
When a customer asks for a code or offer and you sent them a related one before, the draft now names it exactly and says what it covered, then offers to confirm it still applies or set up an equivalent, instead of only promising to check.
One earlier reply with a very strong match no longer crowds out the others. Drafts now always weigh several past replies to the customer, so the one holding the actual code still gets through.
When an earlier reply to a customer is quoted in a draft, ReplyRabbit now picks the part of that reply that actually mentions the coupon, price or offer, instead of the opening lines.
Up to four earlier replies fit in the draft context, so a code sent a while ago is less likely to be crowded out.
Drafts now look further back through your past replies to a customer, around 120 replies instead of 40, so an offer from a couple of years ago still turns up.
When a customer asks about a coupon, discount or refund, replies mentioning those come first, even if the email is mostly signature and product blurb.
Drafts now remember what you told each customer before. When someone asks about a discount code, a price or a promise from an earlier ticket, ReplyRabbit looks back through your past replies to that customer and uses your actual words. Available on Pro, and you can turn it off per mailbox.
Only replies you actually sent are used, never unsent drafts or the customer's own messages, and card numbers, passwords and licence keys are removed before anything is looked up.
A message asking a question or making a request is no longer skipped as "No reply needed". If the AI misjudges one, ReplyRabbit asks again and drafts a reply anyway.
The "No reply needed" note now looks right in dark mode.
Drafts now put things in their own words instead of repeating the phrases the customer used. Product names, version numbers, order and tracking numbers and other precise details are still quoted exactly.
If a draft still echoes the customer too closely, ReplyRabbit quietly asks for a reworded version before showing it to you.
Every fold-away arrow in the conversation view is now the same size. The arrows on the Next actions and ticket panels match the sidebar section arrows instead of running a touch larger.
Pick your AI model from a list instead of typing it. The draft-model and embedding-model settings are now dropdowns per provider, with a "Custom" option if you need to enter a specific name.
Each model field now shows a "Get your API key" and a "Browse models" link that point at the right page for whichever provider you have selected, plus an "About these models" note explaining each option.
OpenRouter can now be used for embeddings (semantic search), so if your drafts run through OpenRouter you can keep everything on one key.
Every conversation-sidebar section now shares one heading style (size, weight, colour and caret), so ReplyRabbit signals, the store connector and FreeScout's own panels all line up in light mode, dark mode, desktop and mobile.
Each section remembers whether you left it open or closed. Your choice sticks per user across page refreshes and sign-ins, so the sidebar comes back the way you had it.
Dark mode gives the whole sidebar one calm near-black background instead of a mix of shades.
"Similar conversations" (the Semantic Search panel) is now off by default and can be switched on from Semantic Search settings.
The consistent sidebar panel headings now carry through to mobile and dark mode: same size, colour, alignment and carets on a phone as on the desktop, with no more stray blue or mismatched indents.
The conversation sidebar panels now share one consistent heading style, so ReplyRabbit and FreeScout's own panels (Previous Conversations, Attachments, and more) line up instead of mixing colours and sizes.
You can now move your ReplyRabbit licence to another site yourself. The trash icon next to the licence on the Modules page (and "Remove" on the licence page) releases it here so you can activate it on a different domain.
Licence keys for other products (like the new Semantic Search module) no longer unlock ReplyRabbit by mistake. Paste them into their own settings page instead.
New from the same team: Semantic Search for FreeScout, a separate module that makes multi-word searches work and finds conversations by meaning. See replyrabbit.app/semantic-search.
Fixed Rewrite so it polishes the reply you wrote instead of answering it. It now knows the text is your outgoing message to the customer and keeps your wording, questions, and tone pointed the right way.
The Rewrite button sits at the end of the editor toolbar again on the latest FreeScout, instead of floating in the gap before the send buttons.
When ReplyRabbit decides a message needs no reply, it now says so on the conversation even if an earlier next-actions checklist is still sitting in the thread. Before, those conversations went quiet with no draft and no explanation.
Drafts now act on a long gap, not just notice it. Replying to a months-old thread, the draft asks whether the customer's plans went ahead and how they turned out, instead of wishing them luck with something that has long since happened.
Drafts now know what day it is. Replying to a message from months ago no longer reads as if it arrived this morning.
After a long gap, drafts treat what a customer told you about their plans and timings as possibly out of date, and ask where things stand instead of assuming.
An old quote, discount or trial from earlier in the thread won't be repeated as if it were still open.
If your AI provider starts failing, you'll now see it right away: a notice on your mailbox settings page and an entry in your helpdesk's activity log, instead of the problem going unnoticed.
Crawl up to 5 documentation sites per mailbox, each with its own page limit and sitemap support, so drafts can draw on platform docs as well as your own.
Large docs sites now build up in the index across daily crawls instead of being re-sliced each day.
Next actions can include links to the docs pages the draft drew on, so you can verify sources before replying.
New docs gap signal spots questions that a public docs or help article could have answered, and tags the conversation so you can turn it into content.
External tools can now list conversation signals in bulk through the MCP connection.
The sidebar now works properly on phones: the signals panel folds away neatly (tap to open), and any notable signals still show as little colored icons in its title.
Next actions tidied up on small screens too: the edit buttons sit in the title row, and the whole checklist can be folded away with a tap.
Suggestions now say why: drafts recommend each offering with a brief reason it would help that customer, then make your offer in its own clear sentence.
Special offers now sound like part of the conversation: drafts phrase your offer in their own words while keeping your exact terms, so the percentage, code, and conditions never change but the sentence flows.
Sales-moment suggestions now read like a natural closing note: the offer gets a proper lead-in instead of pasted quotes, and the suggestion sits in its own tidy paragraph.
Cancellation replies stay focused on the cancellation, without uninvited setup or teardown steps.
Sales-moment suggestions now show up when they should: at a real cancellation or pre-sales moment, drafts include a relevant suggestion by default and only hold back when the customer is upset or it would feel out of place.
Drafts can now spot a natural sales moment, like a cancellation or a pre-sales question, and gently mention other offerings that fit, right after sorting out the customer's request.
You decide what gets suggested: list your offerings in your own words, add a special offer, and it is always quoted exactly as you wrote it, never made up or changed.
Built-in good manners: no suggestions for upset customers, nothing pitched that the customer already owns, and each customer hears about an offer at most once a month (you pick the gap).
Conversations where a customer is cancelling now get their own tag, so you can spot and review them at a glance.
The teach ReplyRabbit menu now says it skips drafts for an email address, not a sender, matching how the rule really works when one person writes from several addresses.
The teach ReplyRabbit menu now shows exactly which email address you're skipping, and reassures you that drafts still arrive when you turn off action suggestions.
The menu is easier to read and easier to use from the keyboard, with clearer text and a visible focus outline.
You can now teach ReplyRabbit from the next actions box: hover the remove button and tell it to skip action suggestions for that email for good, or to also stay quiet for that sender.
Emails you teach it about keep a small note and an 'Actions off' tag in your list, so you always know why nothing was suggested.
Asking ReplyRabbit to skip a sender now takes effect right away, even if sender skipping had never been switched on before.
When AI analysis can't run, the signals panel now explains why instead of staying blank, and ReplyRabbit stops retrying an analysis that can't succeed.
A new setup overview sits at the top of settings, so you can see at a glance if anything still needs attention, like a missing AI model. One click jumps you to the right section.
Suggested segment rules now have a 'No reply needed' switch, so you can mark a suggestion as skip-the-draft before adding it.
Suggestions are smarter about what you already have: a rule matching a domain now hides equivalent suggestions for that same domain, however it was written.
Suggestion cards are more compact, with the reason and match details tucked neatly beside the fields.
Collapsed settings sections now fold up cleanly, with no stray buttons or leftover space at the bottom.
Marketing blasts and security bulletins from other vendors no longer get flagged as sales opportunities. Only a customer's own interest in your products lights up that signal now.
All your model choices now live together in one AI Provider section: drafts, search indexing, mailbox previews, and attachment reading.
Mailbox previews and attachment reading can now run on a faster, cheaper model, or even a different provider, without changing what writes your drafts.
Save a key once and other sections using the same provider share it automatically, with a clear note showing where it came from.
If you remove a key that other sections were sharing, ReplyRabbit warns you right away instead of failing quietly later.
ReplyRabbit is now much better at spotting phishing emails that pretend to share a document or warn about your account, and you can choose to move anything it flags straight to your Spam folder.
Fixed your provider settings after a failed draft? ReplyRabbit now notices and quietly retries recent conversations that were left waiting, so you don't have to chase them one by one.
The next-actions checklist keeps each item short and to the point, so you can scan it at a glance.
The actions badge in your inbox list now shows a little checkbox with a progress count, and the tick turns green when everything is done.
'Find suggestions' for segment rules can now look back up to 90 days, and if something goes wrong it tells you the real reason instead of a generic error.
The settings page got a tidy-up: connected badges line up across sections, Integrations sits in its own card like everything else, and the API key helper now links every provider we support, including Z.AI.
List previews now only refresh your open conversations, so closed and spam ones no longer use up the refresh.
Fetching your website now fills in your social profiles and a short business description too, not just one field, and the new fields sit neatly under the Fetch button.
When your AI provider runs low on credits or hits an account snag, ReplyRabbit now tells you exactly what happened and links straight to where you can fix it, instead of a generic connection error.
You can now see a 30-day breakdown of your AI usage by model, with a rough cost estimate, right from your settings.
The AI model list now includes our newest models, and OpenRouter always uses its true 'auto' option instead of an old pinned model.
The 'sync tags now' button and the tag column grey out when tag syncing isn't turned on, so it's clear there's nothing to click yet.
Signal icons in your inbox list now match the style of your settings page, with a hover to reveal the full label.
First-words previews no longer show old AI summaries or leftover email signatures, so they stay clean and to the point.
Segment rules can now combine more than one condition with AND or OR, so you can match a customer by company and email together, or either one, instead of just one condition at a time. Available on Team.
Priority and mood signals now show a different icon for each level, so critical looks different from high, and very upset looks different from upset, at a glance.
Signal icons in your inbox list are bigger, better aligned, and easier to read in night mode.
ReplyRabbit no longer writes an AI summary for spam, and you can now turn off summaries for 'no reply needed' messages too, so it only spends effort on the ones that matter.
Summaries and previews now skip tracking links, unsubscribe footers, and email signatures, so they show what the message is actually about.
You can tell ReplyRabbit your business website, social profiles, and a short description, so drafts can mention them accurately instead of guessing.
Drafts now answer multi-part questions with a clear list, instead of running everything together in one paragraph.
'Find suggestions' now keeps a fair balance of results, so 'no reply needed' patterns no longer crowd out suggestions for your most-replied contacts and your AI-drafted rules.
It also stops offering a rule you already saved, and groups all the 'no reply needed' suggestions into one tidy summary you can expand and pick from.
Your saved 'no reply needed' rules now sit in a compact, collapsible table that is closed by default, so the rules section stays short.
No-reply rules are now always checked before your other segment rules, so a sender you marked 'no reply needed' is skipped even if they would also match a drafting rule.
Customer segment rules now support a 'No reply needed' outcome, so ReplyRabbit skips drafting entirely for senders like automated vendor notices or a customer's own monitoring bots, instead of just shaping the tone of a reply. Available on Team.
'Find suggestions' now looks at your real conversation history, not just a recent sample, so it can propose senders who rarely get a reply and your most-replied-to contacts, alongside the existing AI-drafted suggestions.
Email matching in segment rules is unified into one option: match by the full address or the domain, with equals, contains, begins with, and ends with all available.
Your AI signal examples now match the style you pick. Choose Emoji or your own text and the whole row updates to show how each signal really looks in your inbox.
The confidence signal now spells out what each level means, so High, Medium, and Low are clear at a glance.
GitHub and Jira conversations now show the same 'issue created' signal as Linear, so you can see at a glance when a conversation is tracked, whichever tool you use.
Two labels are clearer: a possible sale now reads 'Sales lead', and a message that needs a person now reads 'Human review'.
The AI summary preview is easier to follow. It shows a short 'Summarising' note while it writes, and always falls back to the first lines of the message if anything interrupts it.
Connecting an AI assistant is simpler. Settings now include a ready-to-paste prompt and a one-step downloadable skill you can use, and write access for assistants is a single clear admin-only setting.
Your AI signal settings are clearer at a glance. Every signal now previews exactly how it looks in your inbox, including each colour a signal can take (for example critical, high, normal, and low priority), and draft state and Linear links sit in the same table instead of a separate one.
Next steps in your drafts are smarter now. ReplyRabbit learns from the steps you keep or edit, only suggests the ones it is confident about, and writes them as things to do now instead of things to wait on.
Signal settings live in one place now: pick which signals show, choose how each icon looks (a colored icon, an emoji, or your own text) with a live preview, and set their FreeScout tags, all from a single table.
Your knowledge base, docs, and product catalogue now show a clear Connected badge once they are set up, so you can see at a glance what ReplyRabbit can draw on.
ReplyRabbit now includes MCP support for approved clients like Claude, Codex, and Cursor. Admins can enable `/replyrabbit/mcp`, create scoped bearer tokens, limit tools and mailboxes, and review metadata-only activity logs.
The first MCP tools expose controlled support intelligence: mailbox listing, conversation search and reads, cached summaries and analysis, generated reply text without sending, and knowledge search.
Added Jira Cloud support. You can now create Jira issues from FreeScout conversations, keep the linked issue visible in ReplyRabbit, update transitions, priority, assignee, and labels, and draft a customer follow-up when Jira moves the work to Done.
Mailbox integrations now use an issue tracker selector, so existing Linear and GitHub Issues mailboxes keep working while new mailboxes can choose Jira Cloud with their own site URL, account email, and encrypted API token.
Safety and reliability improvements so ReplyRabbit always treats the content in customer messages, knowledge base articles, and attachments as information, never as hidden instructions.
Added an optional heads-up notification when a Linear ticket is created automatically, so your team can review it.
Drafts now reply to whoever sent the latest message and greet them by name, so when a new person joins a thread the reply goes to them, not the original sender. ReplyRabbit also keeps the right people in the loop, copying in only the contacts that latest sender chose to include.
You can now set rules for different kinds of customers, so drafts match how you actually work with them. For example, flag a managed client and ReplyRabbit replies with a warm "we're on it, we'll follow up" instead of asking them to troubleshoot, and it writes the next steps as your team's to-dos rather than theirs. Available on Team.
Drafts now understand the images and PDFs your customers attach. When someone sends a screenshot of an error or a PDF invoice, ReplyRabbit reads it and works what it finds into the reply, so you are not starting from a blank page. Reading attachments is available on Team.
ReplyRabbit also never tells a customer their photo or file did not come through when it actually did. It always knows what was attached, on every plan.
Drafts now consider what each customer has actually bought. When your store connector is set up, ReplyRabbit focuses on the products they own and stops suggesting fixes for things they do not have
Drafts now read warmer and more human by default: they open with a friendly greeting using the customer's first name, use natural contractions, and steer clear of stiff, formal phrasing
When a draft answers a specific question, it shows a short version of that question in italics just above the answer, so the reply has clear context at a glance
Drafts now lean on the right source for each question. You can make your documentation the main source of truth and treat your help center as backup, or let ReplyRabbit pick whichever fits the question best
Bring your own docs. Point ReplyRabbit at your documentation site and it reads those pages too, so drafts can pull in the right guide and walk a customer through it, not just your help center articles
Drafts now explain the fix, not just the problem. When one of your help articles covers what a customer is asking about, the reply walks through the likely cause and the exact steps to sort it, in plain language, instead of only asking for more detail
Even sharper matching: drafts now home in on the exact passage in a help article that answers the question, so the right answer surfaces first even when it is buried deep in a long article
Drafts now find the right help article and product by what a customer means, not just the exact words they use, so the best answer surfaces even when someone describes a problem in their own way
You can switch this smarter matching on for Pro and Team mailboxes from your settings
Drafts now find more relevant help articles when a customer's question arrives inside a forwarded contact form, focusing on the question itself instead of the surrounding name and contact fields
When you send a ticket to Linear, it now gets a clear, developer-friendly title and drops into the right project on its own, so your team knows what needs doing at a glance
The conversation copied into a Linear ticket is now tidied and summarised, with signatures, contact-form clutter and empty space trimmed away, and long replies broken into lines you can actually read
Linear ticket summaries now use the name of the person who handled the conversation, so it reads like a proper handover
Your edits to the next actions list now help shape future suggestions, the same way your reply edits already do
Drafts find the right help article and product more reliably, even on forwarded emails, by focusing on what the customer actually asked rather than the forwarding clutter around it
Your product list now picks up each product's real name and description from your site, so drafts can talk about your products properly
Drafts now lean on what we have learned about a customer even when they only ever start fresh conversations, like forwarding things in, so replies to your regulars feel better informed from the first message
When someone forwards a message to you, like a contact form passed on by a colleague, your draft now reads the whole forwarded note instead of saying the details aren't showing
On a closed ticket, the next actions list now tells you why it's read-only, with a short note explaining that the boxes can't be ticked or edited because the conversation is closed
Fixed the Team analytics dashboard, which had stopped loading and showed an error page. It opens again and shows your adoption, draft quality, and token figures
Tidied up the style-examples page: removed a stray bit of text that was showing above the form, and dropped the extra confirmation checkbox so adding an example is quicker. The reminder to redact personal details before saving stays
Forwarded automated notifications, like a security or login alert with nothing for you to answer, now correctly get no draft, no next-actions checklist, and no Linear ticket. If the customer adds a real question about the alert, you still get a draft as usual
On a closed ticket, the 'next actions' summary now sits in a calm dark grey instead of brown, since a closed ticket no longer needs your attention
Drafts no longer add a second 'Thanks, [your name]' at the end. Your own signature is the only sign-off now, so replies finish cleanly without doubling up
Closing a conversation no longer hides its next actions checklist or its linked Linear ticket. They tuck into a tidy 'click to expand' panel instead, so you can open a closed ticket and check at a glance that everything on the list got done
Your conversation list now shows ReplyRabbit at a glance. Conversations with a draft show how many next actions are done plus a Draft tag, ones that needed no reply (including automated notifications) show a quiet 'No reply needed' tag, and you still see the real message next to it
The 'no reply needed' note is calmer now, in soft greys, since it doesn't need your attention
Fixed a problem where ticking an item in your next actions list could say it couldn't be saved. Your checkmarks now save reliably
The rabbit looks a touch bolder so it sits nicely alongside your other menu icons, and older 'no reply needed' notes now show the new rabbit and sit neatly across the row too
The ReplyRabbit rabbit has a fresh, friendlier look. The new bunny now shows up everywhere you see ReplyRabbit, including the 'no reply needed' note, which also sits more neatly across the row with its close button tucked to the right. The little rabbit in your mailbox menu now lines up tidily with the other icons too
When you click the rabbit to draft a reply yourself, it now always writes one, even on a thank-you, a closing note, or an automated notification where ReplyRabbit would normally hold back. If there really is nothing to reply to, it tells you so kindly instead of showing an error
The ReplyRabbit icon has a friendly new look, and the same little rabbit now appears on the reply toolbar and beside ReplyRabbit in your mailbox menu
When ReplyRabbit decides a message does not need a reply, it now leaves a small 'No reply needed' note where the draft would have been, so you can see at a glance that it made a choice instead of finding an empty space. One tap clears the note, and it goes away on its own the moment a real reply or draft is added
Short questions that happen to open with a thank-you (like 'Thanks, can I get a refund?') are no longer skipped by mistake, so you still get a draft when one is actually needed
ReplyRabbit now learns your writing style from the edits you make and uses your real replies as examples, so drafts sound more like you and need fewer changes over time
Drafts come out cleaner: your name, role and company fill in automatically, and signatures, legal footers and repeated quotes are stripped out before the AI reads them
Long-term customers get a warmer, more familiar tone, while unsolicited cold pitches are recognised and usually left without a draft
A new Learned style screen lets you review, edit and add the examples ReplyRabbit learns from
ReplyRabbit now reads the whole conversation before drafting, and quietly skips a reply to simple thank-yous and 'all sorted' messages, so you get a draft only when one's actually needed
Fixed the 'Fetch examples from recent conversations' picker: the popup no longer gets stuck behind an overlay, and it now reliably finds your answered conversations to learn from
ReplyRabbit no longer drafts replies or checklists for your own mailbox's mail. Auto-Bcc copies and other self-sent loops are now ignored, even after a conversation is reassigned to a customer
