> ## Documentation Index
> Fetch the complete documentation index at: https://docs.typewise.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Product updates

> Highlights and improvements across the platform

<Update
  label="August 20 2026"
  tags={["Improvements", "Bug Fixes"]}
  rss={{
title: "Faster AI replies",
description:
"Plus Knowledge Hub URL checks, smarter autocorrect, and fixes across email threading, channel switching, and Twilio setup."
}}
>
  ## Improvements

  * **Faster AI replies.** Every AI draft passes a review step before it reaches a customer, and roughly a quarter of drafts were being sent back for a rewrite over rules the drafting step had never been given. Aligning the two removes that redundant rewrite, so replies land sooner.
  * **Knowledge Hub warns about unreachable live URLs.** When you enter a URL that cannot be reached, the form now tells you before you save, instead of accepting a source that fails on every lookup. Templates containing a placeholder variable are skipped.
  * **Autocorrect stays out of the way.** Editing an AI-drafted reply now only corrects the words you type yourself, and words already used in the conversation, such as customer names, order codes, and product terms, are left exactly as written.

  ## Bug Fixes

  * Fixed inbound emails with an unrelated subject being threaded into an existing ticket, which merged two separate customer requests into one.
  * Fixed chats switching to email the moment a human agent replied, which split the conversation across two channels. The switch is now confirmed with the customer instead of forced.
  * Fixed the internal ticket title being used as the customer's email subject after a chat moved to email.
  * Fixed action cards showing an unpublished draft action instead of the published one the agent actually ran.
  * Fixed Bring Your Own Twilio setup hanging while loading senders, and a false "multiple senders" error caused by Twilio's WhatsApp Sandbox.
  * Fixed custom MCP tools with nested parameters reaching the agent without their inner fields, which left it guessing parameter names and missing documented filters.
  * Fixed integration knowledge sources advertising no parameters, which cost a discarded first attempt on every call.
  * Fixed AI Operator edits stripping bold, italics, links, and list numbering from saved instructions.
  * Fixed Conversation Simulator scores that left out knowledge base searches and fallback specialist activity.
  * Fixed stale conversation details persisting after a long conversation was summarized.
  * Improved knowledge base search quality by no longer indexing our own chat widget icon from scraped pages.
</Update>

<Update
  label="August 14 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Signatures & BYO Twilio",
description:
"Per-agent signature details that resolve automatically, connect your own Twilio account for WhatsApp, broader language detection, plus fixes to live URL knowledge sources and Form API submissions."
}}
>
  ## Personalized email signatures

  Signatures previously supported automatic translation and variable highlighting, but the only details available were first name, last name, and company. Anything else identifying the sender had to be typed into the template by hand.

  Administrators can now define the information their agents carry, in their own vocabulary, under Settings → Organization → User Properties. Each agent fills in their own values and photo on their profile, and one signature template resolves them per sender. A new agent completes their profile and the existing template covers them with no admin work.

  ## Bring your own Twilio

  You can now connect your own Twilio account for WhatsApp, so the number, the account, and the billing stay with you. Setup takes a single screen with your account SID and auth token. The existing Typewise-managed connection continues to work unchanged for everyone already using it.

  ## Language detection across all supported languages

  When a request arrives without a declared language, detection now runs across every language Typewise supports rather than a narrower preset list. Workspaces handling mixed-language traffic get more reliable classification without configuring a list upfront.

  ## Improvements

  * Simulator tests that use the verify email action now run to completion instead of ending in a handoff.
  * Clearer error reporting when the AI produces an empty reply, instead of a generic unknown error.

  ## Bug Fixes

  * Fixed expired Salesforce sessions failing permanently instead of refreshing.
  * Fixed live URL knowledge sources failing to fetch pages that redirect.
  * Fixed German webform submissions receiving English ticket titles and summaries.
  * Fixed Form API follow-up submissions losing their variables when merged into an existing ticket.
  * Fixed links in AI replies being rewritten into broken URLs, which caused empty handoffs.
  * Fixed the AI agent escalating to a human when a tool parameter could be resolved automatically.
  * Fixed chat tickets still showing as snoozed after being closed.
  * Fixed empty image attachments breaking the conversation.
</Update>

<Update
  label="August 6 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "AI file attachments",
description: "Plus support for your own OAuth clients on custom MCP integrations, cleaner dictation, and a batch of reliability fixes."
}}
>
  ## The AI can now send file attachments

  Customers often need a manual, a form, or a document. Attachments returned by your integrations, or files a human colleague adds when answering an AI request, can now be sent to customers across chat, email, and WhatsApp. The AI reads each file as context and only shares the ones it explicitly selects, and attachments are preserved when a reply goes through human approval.

  ## Improvements

  * Custom MCP integrations can now use your own OAuth client, for providers such as Google Workspace that require pre-registered credentials.
  * Dictation in the AI Operator now transcribes after you finish recording, producing clean, punctuated text without filler words.
  * Building on selective publishing from the last release: publish and discard are now a single control, so you review, publish, and clean up pending changes in one place.

  ## Bug Fixes

  * Customers now receive a single wait message when the AI checks with a colleague several times in one turn, instead of repeated on-hold replies.
  * AI email signatures now also appear in conversations synced to your CRM and in handoff emails.
  * Fixed product cards occasionally rendering with broken image links or as plain text; they now use exactly the links your tools return.
  * The AI can no longer tell a customer it handed the conversation off without the handoff actually happening.
  * Fixed AI runs that could fail part-way through longer troubleshooting conversations instead of completing or handing off cleanly.
  * Fixed a failure that could stop the AI on very long conversations.
  * Fixed an error that could interrupt AI processing while a request to a human colleague was pending.
  * Ending a chat in the widget now takes effect immediately, even on a slow connection.
  * Edits to a newly created ticket view are no longer lost when the tab closes right after.
  * The reply composer no longer gets cut off at the bottom of the window with long drafts.
  * Voice input in the AI Operator now reports microphone problems instead of failing silently, and switches automatically when your microphone changes.
  * The AI Operator can now see system actions such as handoff and email verification, and no longer removes existing mentions of them when editing instructions.
  * Conversation Simulator: long test lists and test-run conversations now scroll, layouts render correctly on wide screens, a new chat can start without an email address, and simulator handoffs no longer light up the inbox notification dot.
  * Suggestion dropdowns in editors now close when you click outside them.
</Update>

<Update
  label="July 30 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Publishing & Verification",
description: "Plus clearer context-variable events, a readable specialist breakdown, and fixes for email reopens and chat delivery."
}}
>
  ## Choose exactly what goes live when you publish

  Publishing used to push every pending change at once, which meant an unfinished edit could go live alongside the one you wanted. You can now review your unpublished configuration changes, select the ones that should go live, and discard the rest. Changes that depend on each other are grouped and published together, so nothing ships half-finished.

  ## Verify a customer's email address up front

  The AI used to ask for email verification only at the point an action needed a trusted address, which interrupted the conversation late. You can now have it verify the address at the start instead, so the rest of the conversation runs on a confirmed identity and later actions don't stall.

  ## Improvements

  * Taking over a ticket now keeps the dialog open until the takeover completes, so a failed takeover is no longer reported as successful.
  * The resolution breakdown in Insights Hub no longer includes the "No Specialist" bar, which makes the remaining specialists readable.
  * The conversation timeline now reports the actual reason a context variable was not extracted instead of guessing.
  * The timeline says a context variable was added the first time it is filled, rather than updated.

  ## Bug Fixes

  * Building on automatic email threading from the last release: when a customer replies to an automatically closed ticket, the AI now picks the conversation back up instead of leaving it waiting in the human queue.
  * Messages a customer sends in quick succession in the chat widget are all delivered.
  * The chat widget scrolls to the newest message again.
  * A knowledge source that has not finished loading no longer interrupts a conversation; the AI re-discovers its tools and continues.
  * A brief connection problem while loading a workspace's configuration no longer leaves a ticket stuck in an error state.
  * Inbound email from an address previously used in the Conversation Simulator now creates a real ticket instead of joining the simulated conversation.
  * Partial handoffs no longer arrive empty or repeat the customer's own message.
  * Salesforce knowledge articles without content are skipped instead of being indexed as title-only entries that can outrank the real answer.
  * Handoffs and syncs now set the expected ticket status in Salesforce.
  * The Conversation Simulator now grades the reply the customer actually received, so a run can no longer pass on a draft that message review rejected.
  * Simulator runs that correctly escalated to a human are no longer graded as failures.
  * A replaced or broken integration no longer keeps the "AI integration not responding" banner on, and broken connections now appear in the integrations list so they can be removed.
  * Connecting an MCP server whose sign-in lives on a different domain, such as Calendly, now completes instead of failing at the last step.
  * Long spam explanations now wrap in the tooltip instead of stretching across the screen.
  * "No options found" can no longer be picked as a value in dropdowns.
</Update>

<Update
  label="July 23 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Mobile agent app",
description: "Plus email threading into closed tickets, action-approval controls, and a batch of reliability fixes."
}}
>
  ## The agent workspace now works on mobile

  You can run the inbox and answer tickets from your phone. Drill into a ticket from the list, read it, open customer details, and send a reply, with the composer and keyboard staying out of the way. The desktop experience is unchanged.

  ## Improvements

  * Automatic email threading now merges a follow-up into a recently closed ticket and reopens it when it belongs to the same conversation, so replies that arrive after a ticket closes stay in one place.
  * Clearer Gmail forwarding setup: a dedicated guide explains where Google's verification email arrives, so first-time setup doesn't get stuck.
  * The AI Operator can now turn human approval on or off for actions such as Update Customer Details.

  ## Bug Fixes

  * You can now close a ticket even after it reaches the conversation-length limit.
  * Form submissions with an unsupported attachment now create the ticket and hand it off to an agent instead of failing.
  * The AI no longer leaks internal summary text into long conversations.
  * AI replies no longer include a duplicate email signature.
  * MCP connections no longer drop during token refresh.
  * Choice options shown by the AI Operator now render their links correctly.
  * Onboarding no longer freezes because of a leftover session.
</Update>

<Update
  label="July 16 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Ratings & Auto-merge",
description: "Plus reusable actions with pre-filled inputs, context variables filled from the conversation, and a round of email fixes."
}}
>
  ## Conversation ratings

  Agents and admins can now rate finished conversations from 1 to 5 stars, for both AI and human handling. Ratings appear in the conversation timeline, so teams can review quality and spot conversations worth learning from.

  ## Automatic ticket merging

  New incoming emails are checked against your existing tickets and merged into the right conversation instead of opening a duplicate. Threads stay complete even when customers write in again.

  ## Multiple instances of the same action

  Add the same action or lookup several times with different names and pre-filled inputs, for example one case-creation action per queue or one send-email action per inbox. The AI picks the right one from a clear name instead of guessing parameter values.

  ## Context variables filled from the conversation

  Building on Context Variables: the AI now fills them with details customers share during the conversation, and agents can add or edit variables manually in the ticket sidebar.

  ## Improvements

  * When the AI is missing information, it now asks a human colleague for input instead of handing off the entire conversation.
  * Team and workspace changes are now logged in the ticket timeline, like assignment events.
  * The AI Operator now introduces itself as Nova.
  * Building on forward-recipient suggestions: suggested addresses can now be dismissed, for example ones with a typo. Forwarding to a dismissed address brings it back.
  * Mentions and forwarded replies can be marked as solved to clear them from the inbox; solved threads can be reopened.
  * Forward detection now selects the customer's address automatically instead of also offering your own sending address.

  ## Bug Fixes

  * Tickets with unsupported attachments such as ZIP files are processed normally instead of being handed off immediately.
  * Incoming emails without threading headers are processed normally again.
  * Tickets synced to Freshdesk stay closed and show replies in the right order.
  * Signature and body images are no longer cut off in outgoing emails.
  * Forwarded emails with very short text no longer pick a signature in the wrong language.
  * "Start new chat" in the chat widget now opens a new conversation even when all previous tickets are closed.
  * Integrations can be reconnected without first changing a field.
  * Fixed the priority filter closing after the first selection; several values can now be picked in one go.
  * Tool calls now appear correctly in the Conversation Simulator.
</Update>

<Update
  label="July 9 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Auto-unassign & channel",
description: "Plus Salesforce reply threading, Saved Views fixes, and cleaner AI email replies."
}}
>
  ## Automatic unassignment on human agent reply

  When an agent replies to an email or WhatsApp ticket, Typewise can now unassign them automatically. A customer's later reply no longer waits in the queue of an agent who has gone offline, so urgent messages get picked up faster.

  ## Show channel on every ticket

  Tickets now show which channel each message arrived on (email, chat, or WhatsApp), and messages are presented consistently across channels. Anyone reviewing a ticket can see where it came from at a glance.

  ## Improvements

  * When reconnecting an integration, you can review its stored configuration, including the Salesforce instance URL, to confirm it points to the right environment.
  * Self-hosted MCP connections can now change their URL on reconnect without deleting and rebuilding their tools, actions, and knowledge.
  * Supported attachment types, including ZIP files, are now forwarded to Salesforce when a ticket is handed off.
  * Building on Saved Views from the last release: the AI operator can now work with saved views, and returning to an unsaved, edited view keeps its full edit state.

  ## Bug Fixes

  * Fixed search returning no results when started from a page other than the first in All Tickets; it now returns to the first page and shows matches.
  * Customer replies to Salesforce-handled emails now attach to the existing case instead of opening a new one.
  * Fixed an AI timeout message caused by a delay before the AI backend started processing.
  * Internal reasoning no longer appears before the customer text in AI email replies.
  * AI replies no longer describe the company from an outside-observer viewpoint and speak as the company instead.
  * Manual actions no longer fail on tickets whose field names contain accented characters.
  * Saved Views now keep their filter when saved, and inline renames are committed.
</Update>

<Update
  label="July 2 2026"
  tags={["Improvements", "Bug Fixes"]}
  rss={{
title: "Saved Views for Tickets",
description: "Plus inline ticket editing, forward-recipient suggestions, and a batch of filter and sorting fixes."
}}
>
  ## Improvements

  * **Saved Views for All Tickets**: create, edit, and delete custom views of your ticket list. Views come with a redesigned filter panel, column reordering that persists per view, a Priority filter, and quick "No assignee" and "Me" filters.
  * Edit a ticket's priority, team, and assignee inline from the list, without opening it.
  * Email forwarding suggests recipient addresses you've used before.
  * Search within the integration action and tool picker.
  * Manual actions and lookups now let you mark individual fields as optional instead of always required.
  * Tickets handled by the AI now show as "Unassigned" instead of "AI".

  ## Bug Fixes

  * A ticket now leaves "waiting for customer" once the customer replies to a snoozed ticket.
  * "Is not" filters now include tickets with an empty assignee, team, or resolution status.
  * You can sort All Tickets by team, assignee, and channel type.
  * Fixed misaligned column headers in All Tickets.
  * The per-view export button is now labeled as Excel (.xlsx) to match the file it produces.
  * Resetting a filter no longer leaves a stale search field behind.
  * Integration credential changes now take effect immediately, instead of intermittently failing until a restart.
</Update>

<Update
  label="June 25 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Branded email footers",
description: "Plus better forwarded-email threads, a steadier reply editor, and agent reliability fixes."
}}
>
  ## Branded email footers

  You can now embed your own banner or logo image in the email footer, so automated replies stay consistent with your brand on every customer touchpoint.

  ## Improvements

  * View the full email thread and expand quoted text for forwarded messages.
  * Forwarded-email replies are now easier to tell apart in the conversation view.
  * The reply editor caps its height for long messages and keeps the action and send bar always visible.
  * Update conversation variables mid-conversation through the chat SDK.

  ## Bug Fixes

  * Emails now send reliably even when the reference header is very long.
  * The AI agent recovers from "entity not found" tool errors instead of failing the whole reply.
  * The AI Specialist now uses information provided in the email subject line instead of asking again.
  * Conversation Simulator QA now evaluates against the correct workspace's knowledge.
  * Fixed false output-validation failures in simulator tests.
</Update>

<Update
  label="June 16 2026"
  tags={["Improvements", "Bug Fixes"]}
  rss={{
title: "Inbox and handoff improvements",
description: "Original-subject email handoffs, forwarded-email replies in the inbox, note attachments, and a batch of integration and UI fixes."
}}
>
  ## Improvements

  * Email handoffs and backups now keep the ticket's original subject line, so they still match the right case in your CRM.
  * Replies to forwarded emails now appear in the inbox instead of staying hidden.
  * Internal notes and replies now support file attachments.
  * After you create or publish a specialist, you're taken back to the specialist list.

  ## Bug Fixes

  * Fixed a bug that could remove a user from all of their teams.
  * Salesforce actions and lookups now match records by user ID instead of email, so they resolve the right customer.
  * Tickets created through the Form API now go to the correct workspace.
  * Fixed pagination controls that wouldn't respond until the page was refreshed.
  * The sidebar now collapses as soon as you click the toggle, instead of waiting until you move the cursor away.
  * Saving a knowledge source no longer flashes a stray confirmation dialog or briefly resets the form before closing.
  * Emails with a line break in the subject line no longer fail to send.
  * A partial handoff now stays partial instead of completing the whole handoff.
  * Fixed intermittent Knowledge Hub lookup failures.
  * The link button in the message editor opens the URL dialog again.
  * Dropdown menus now stay within the screen on small displays, so the lower options are reachable.
  * Selected options now show a clear blue outline, so it's obvious which one is active.
</Update>

<Update
  label="June 5 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Context Variables, plus Operator & Simulator refinements",
description: "The Variable Registry is reborn as Context Variables with Verified inputs, alongside a wave of AI Operator, Simulator, publishing, and CRM fixes."
}}
>
  ## Context Variables

  The Variable Registry is now **Context Variables** — a clearer screen with labelled columns and a per-row *Verified* checkbox. A set of well-known variables (`user_email`, `originUrl`, `name`, `subject`, plus system-managed `userId`, `crmTicketId`, and `forwarded_email_recipient`) is available automatically. On Actions, *Secure inputs* are now **Verified inputs**: pick a Verified variable or a static value, and the AI agent cannot override, substitute, or collect it from the conversation. Variable names may also contain spaces, and unfilled variables are no longer passed to the AI.

  ## Improvements

  * Move tickets between workspaces, preserving shared variables and applying the target workspace's configuration
  * Search tickets by a partial or truncated ticket ID
  * The ticket timeline now shows who snoozed a ticket and until when
  * Line-based diff view for more precise change review
  * Simulator: tools referenced in global guidelines can now be mocked, explanation text renders markdown, and tests require at least one channel to be selected

  ## Bug Fixes

  * Chat SDK now isolates its network requests, so scripts on the host page (e.g. Shopify, Magento) can no longer interfere with the widget
  * AI Operator: fixed an unclickable button behind the simulator drawer, focus conflicts with the diff view, an internal tool name leaking into chat, and messaging an old chat opening the wrong conversation
  * Publishing: the diff view can no longer get stuck in an unrecoverable lost state
  * Salesforce email sending no longer fails on malformed sender addresses, and Freshdesk synced tickets now receive the correct closed status
  * Responsiveness, translation, and ticket-view performance fixes, plus clearer messaging after the AI hands a conversation to a human
</Update>

<Update
  label="May 21 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Forwarding & Spam Rules",
description: "Plus clearer inbox date tooltips"
}}
>
  ## Forward tickets to teams outside Typewise

  CS agents can now forward any ticket to an external email address, with replies coming back into the platform as internal notes and the original ticket auto-snoozed. The forward composer includes an emoji picker, editable recipient pills, threaded history, and an "include history" toggle.

  ## Adjustable spam rules

  The spam classifier now reads per-client free-text rules and your company description from configuration, so admins can tune filtering for their business — flag specific message types, allow certain senders, change tone — without code changes.

  ## Improvements

  * Simulator tests now invoke the real lookup if the mock field is left empty.
  * Freshdesk subdomain field now shows `https://` as a visible prefix so admins know it's appended automatically.
  * Inbox date tooltips now clarify whether the timestamp is created, updated, or "waiting for customer reply since".

  ## Bug Fixes

  * AI agent correctly routes pure information requests to the real-agent lookup instead of producing a partial handoff.
  * Specialist knowledge-source creation now succeeds, and failures are reported to Sentry with the real error instead of a generic 400.
  * Knowledge source imports correctly fail for unreachable URLs instead of indexing 4xx/5xx error pages and reporting "ready".
  * MCP/Composio mocked tools are again usable in simulator test runs.
  * Stripe MCP tool is now visible to simulator tests (no more always-handoff).
  * "Snooze until customer replies" option is again visible in the default snooze menu.
  * Diff view shows the action name even when the underlying action has been deleted.
  * "Run all tests" no longer fails with "Failed to create test" on existing test suites.
  * Real-time change notifications are now scoped to the originating workspace.
  * Save button is now disabled when an action or source form has no changes.
  * Entity query cache is now flushed on publish, save, and discard so the UI reflects the latest server state.
  * Diff view filters out net-zero entity changes so undone edits no longer appear as pending changes.
</Update>

<Update
  label="May 14 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Faster knowledge scraping & more",
description: "Push-based knowledge source scraping, CSAT export, and a batch of fixes and improvements."
}}
>
  ## Faster Knowledge Source Scraping

  Knowledge sources now update using a push-based architecture, replacing the previous polling approach. Changes to your connected websites and documents are detected and indexed faster, keeping your AI agent's knowledge up to date with less delay.

  ## Improvements

  * Ticket exports are now downloaded as XLSX files, consistent with the Excel export in Insights Hub
  * Added an Export button to the CSAT table in Insights Hub, including ticket IDs
  * Ignored and spam tickets are no longer synced to connected CRM systems such as Freshdesk
  * The changelog now displays entries in a consistent order; timestamps are shown more clearly to avoid ambiguity
  * Fixed approximately 15 untranslated UI strings across the platform, improving the German locale experience

  ## Bug Fixes

  * Co-Pilot can now correctly access knowledge sources marked as "internal use only"
  * Regenerating a Form API key now correctly invalidates the previous key
  * Taking over a forwarded email now pre-fills the correct sender address in the "To" field
  * Fixed an issue in the Chat SDK where empty suggestions caused an unexpected language input field to appear
  * Fixed a bug in the Test Suite where marking a lookup or source as mandatory caused test runs to fail immediately
  * "Close conversation" now appears correctly as an action in the Simulator
  * The "Transferred to CRM" status message is now shown only after the transfer is confirmed, not when it is first initiated
</Update>

<Update
  label="May 7 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Magento MCP integration and clearer error messages",
description: "Magento lookups from the AI Agent, human-readable MCP errors in the timeline, localized email subjects, and bug fixes."
}}
>
  ## Magento MCP integration

  The AI Agent can now look up orders, products, and customer details directly from Magento while handling tickets, removing the need for manual lookups.

  ## MCP error visibility

  When an MCP tool fails (expired token, rate limit, or other error), the ticket timeline now shows a clear, human-readable message instead of a generic "AI processing failed". The AI Agent also recovers from non-critical failures by skipping the affected tool and continuing the request.

  ## Improvements

  * Webform email replies now use a subject line generated in the end customer's language.
  * Saving an action with incomplete configuration is now blocked, with inline feedback explaining what's missing.
  * Dismissed browser notification permission prompts no longer re-appear.

  ## Bug Fixes

  * Valid incoming emails are no longer flagged as spam by the AI spam filter.
  * CRM sync failure errors no longer show to agents after a ticket is handed off to a Human Agent.
  * Tickets are no longer closed via the Chat SDK when a channel switch occurred.
  * The CRM sync failure notification bar now clears once the issue is resolved.
</Update>

<Update
  label="May 4 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Issue Reporter Widget",
description: "Smarter multi-question email handling, readable AI tool names, and a handful of minor bug fixes."
}}
>
  ## Report-an-Issue widget

  Agents and admins can now report bugs or feature requests directly from the platform. The new chat-based reporter captures the description, generates a clean title automatically, supports attachments, and routes everything into Linear. Admins get a dedicated dashboard to track all issues raised by their teams in one place.

  ## Improvements

  * **Auto-reload after entity changes**: when you modify a specialist or other entity, related pages now refresh automatically without a full reload.
  * **Readable names for AI tools**: tool calls in the timeline now show human-readable names instead of internal identifiers.
  * **Super-admin Client Management page**: a new page in the Flow Platform consolidates client/workspace administration that previously required two separate platforms.
  * **Multi-question emails**: the AI supervisor now keeps working through every question in the same incoming message instead of stopping after the first one.
  * **Webform handoff to Salesforce**: when a conversation comes from a webform, Salesforce handoff sets the case origin to "Webform" and forwards webform variables.
  * **PlentyMarkets (PlentyOne) integration**: added support for PlentyMarkets as a back-office integration, covering order status lookups, shipment tracking, product catalog search, and a cancellation/refund flow that requires explicit human approval.

  ## Bug Fixes

  * Chat SDK variables are read again on new chats (regression fix).
  * Resolved an issue where messages would briefly disappear and only reappear after a page reload.
  * Remaining hardcoded UI strings in the AI Operator are now properly localized.
  * Spam filter no longer incorrectly blocks emails the AI has already processed; you can also un-ignore an email.
  * Insights Hub layout now stays responsive when the AI Operator panel is widened.
  * Cleaned up Salesforce knowledge base articles that referenced Zendesk-style tags Salesforce doesn't support natively.
  * Images attached when reporting an issue are now visible to admins.
  * AI processing failures now hand off to a human agent instead of replying with a generic error.
  * Fixed an edge case where, after email verification, an older untrusted email could still reach downstream tools.
  * Multi-specialist email handling no longer always hands off to a human when not needed.
  * Image upload now gracefully falls back to the original when optimization fails.
</Update>

<Update
  label="April 23 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Diff view and more",
description: "Review & edit changes before publishing, filter tickets by specialists and more."
}}
>
  ## Diff View: Review and edit changes before publishing

  A new diff view lets admins see exactly what changed before publishing — with word-level highlights for added and removed text across all field types (instructions, toggles, checkboxes, tables). Changes can be edited inline directly in the diff, so there is no need to navigate back to the editor. The view also shows who made each change.

  ## Improvements

  * Supervisors can now filter the ticket list by specialist, including a link to the filtered view in Insights Hub. The fallback specialist is also shown when enabled.
  * The CSAT Evaluation table now shows whether each ticket was handled by AI or a human agent, using the same color coding as the CSAT graphs for easy cross-referencing.
  * CSAT exports now include customer comments alongside ratings, giving teams a complete picture of satisfaction feedback.
  * Updated the integration documentation to recommend direct script injection over tag managers for more reliable widget setup.
  * Fixed an issue where the AI agent asked for confirmation of a manual action that had already been approved, causing duplicate approval prompts.
  * The Web Search lookup tool now displays a clean, generic label instead of the full URL, preventing the label from being cut off in the interface.

  ## Bug Fixes

  * Selecting additional attachments now adds to selection instead of replacing previous ones.
  * Changing specialists in Typewise Internal workspace no longer triggers incorrect publish prompt.
  * CRM handoff notifications now follow correct behavior on email channels, matching internal handoffs.
  * Spam and ignore detection improved; replies to AI-created tickets are no longer misclassified.
  * Layout stability improved in input tables; validation errors no longer shift adjacent fields.
  * CSAT survey timing corrected to wait until conversation ends rather than triggering after AI question.
  * Newly created users no longer see unpublished changes on first login.
  * Tickets are no longer flagged for agent takeover before customer reply.
  * Inbox display corrected for tickets with switched channel types; no longer pushed to top of unassigned queue.
  * Lookup file preview no longer throws "Not Found" in new workspaces.
  * Spam filtering improved to reduce unwanted emails reaching inbox.
  * Customer details panel now shows readable workspace names instead of internal IDs.
  * Duplicate tickets from single customer interaction prevented.
  * CRM handoff emails now respect "Require Human Approval" setting.
  * Email subjects now render as readable text instead of MIME-encoded strings.
  * Serialization issue on agent response endpoint resolved, preventing production crashes during approval interrupts.
</Update>

<Update
  label="April 16 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Ticket API & Notifications",
description: "Plus multi-question email handling, verified variable copy updates, and several bug fixes."
}}
>
  ## Ticket API

  Clients can now retrieve tickets programmatically via a new REST API. This enables integrations with external systems — for example, triggering Trustpilot review invitations for customers with high satisfaction scores.

  ## Notification for Live Chat & Urgent Tickets

  Agents now receive real-time notifications when a live chat request arrives or an urgent ticket needs immediate human attention. This ensures faster response times for time-sensitive conversations.

  ## Improvements

  * The AI agent now handles multiple questions in a single email, answering all of them instead of asking the customer to prioritize
  * Email templates now include a "Conversation History" heading for clearer structure
  * Improved copy text for verified variables to clarify how "Verified" and "Any value" sensitivity settings work

  ## Bug Fixes

  * Fixed image attachment processing that caused timeouts and error states
  * Selecting a URL and clicking the link button now auto-fills the link field
  * Fixed broken "View ticket" links in the Insights Hub flagged content list
  * Deleted tickets no longer appear in the inbox
</Update>

<Update
  label="April 10 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Chat verification, PDF export, and MCP integrations",
description: "New email-based customer verification for chat, PDF export for tickets, custom MCP integrations, and multiple platform improvements."
}}
>
  ## Customer Verification in Chat

  Users can now verify their identity directly in the chat widget through a secure email-based verification flow — no login required. This enables businesses to authenticate customers before accessing sensitive data, while keeping the conversation seamless.

  ## PDF Export for Tickets

  Individual tickets can now be exported as PDF documents directly from the platform. This makes it easy to share ticket details offline for review, compliance, or record-keeping.

  ## Custom MCP Integrations

  New custom MCP integrations enable AI agents to access external product catalogs and data sources in real time during conversations, making responses more accurate and helpful.

  ## Improvements

  * Knowledge source creation now uses a cleaner two-step flow: select the source type, then configure it in a dedicated drawer
  * Knowledge Hub displays detailed error messages when file ingestion fails, so admins can quickly identify and resolve issues
  * Knowledge source configuration shows inline validation errors for invalid input
  * Redesigned navigation bar with improved layout and visual hierarchy
  * Added the fallback specialist to the Insights Hub filter for more complete analytics

  ## Bug Fixes

  * Fixed an issue where multiple tickets were created for a single email conversation thread
  * Fixed cut-off text suggestions on mobile viewports
  * Resolved a configuration issue with Notion MCP sources
  * Fixed special character validation in tool and specialist names
  * Fixed incorrect translations in ticket description fields
  * Fixed the 'Add' button not working when adding multiple include URLs to a knowledge source
</Update>

<Update
  label="April 1 2026"
  tags={["Features", "Improvements", "Bug Fixes"]}
  rss={{
title: "Workspaces, Autonomous Ticket Closing & More",
description: "Introducing Workspaces support and autonomous ticket closing, plus Knowledge Hub web fetch, WhatsApp channel, and a wave of improvements and bug fixes."
}}
>
  ## Autonomous Ticket Closing

  AI agents can now close tickets autonomously once a customer request is resolved — no handoff to a human agent required. This action is configurable per Specialist and enables fully autonomous, end-to-end customer service workflows.

  ## Workspaces Support

  Organisations can now create multiple workspaces within a single Typewise account, each with its own AI configuration, knowledge sources, and branding. This makes it straightforward for multi-brand companies to manage distinct customer experiences from one central platform.

  ## Improvements

  * **Knowledge Hub web fetch**: The Knowledge Hub AI can now retrieve content from external URLs in real time, enabling more accurate responses when a website link is referenced in a conversation.
  * **Web fetch visibility**: Fetch tool calls are now shown in the conversation sidebar, so you can see exactly which URLs the AI accessed and whether the fetch succeeded.
  * **URL validation in message reviewer**: The AI message reviewer now checks that URLs included in responses actually resolve, preventing broken links from reaching customers.
  * **Variables & metadata across all AI endpoints**: Variables and conversation metadata passed via the API are now available to all AI endpoints — including metadata generators, inbox filters, and team assignment — enabling richer, context-aware AI behaviour.
  * **WhatsApp channel**: WhatsApp is now available as a channel on the Typewise Platform. Access is currently guided by Customer Success; self-service onboarding is coming soon.
  * **More languages for Chat Widget CSAT**: French and Portuguese are now supported in the Chat Widget for CSAT surveys and chat interactions.
  * **Channel-optimised AI models**: The AI now uses a faster model for chat and a higher-quality model for email, improving both response speed and email quality.
  * **Greeting messages passed to AI**: Pre-defined greeting messages configured in the Chat Widget are now included in the AI context, so the AI can respond in continuity with the initial messages shown to users.
  * **OpenAI European data residency**: AI processing is now routed through OpenAI's EU endpoint, improving data sovereignty compliance for European clients.
  * **Simulator: bulk delete of test tickets**: Simulator-generated tickets can now be deleted in bulk while keeping all real tickets intact.
  * **Chat Widget quick start guide update**: A new section documents the auto-open API, making it easier to configure the widget to open automatically on page load.

  ## Bug Fixes

  * Fixed product carousel responses displaying raw HTML instead of formatted product brackets in chat.
  * Fixed URLs in handoff conditions not being evaluated by the message reviewer.
  * Fixed Insights Hub channel filter returning N/A data.
  * Fixed Insights Hub data display regression introduced in the previous release.
  * Fixed file attachments missing when tickets were created in Salesforce via handoff.
  * Fixed Chat Widget failing to load on websites using RequireJS.
  * Fixed Salesforce userId not populated correctly during handoff.
  * Fixed handoff routing to draft Specialist configurations instead of the published version.
  * Fixed sentiment endpoint error on inputs exceeding the token limit.
  * Fixed duplicate parameters being allowed in tool configuration.
  * Fixed diacritic characters rendered incorrectly in email subjects.
  * Fixed Chat Widget preview mode: widget button no longer jumps unexpectedly.
  * Fixed Knowledge Hub old source versions not being cleaned up after publishing.
  * Fixed tool agent outputs not following the sharable/not\_sharable pattern.
  * Fixed mid-message diacritic characters being dropped in AI responses.
  * Fixed 500 error when fetching tickets the user was not authorised to access.
</Update>

<Update
  label="January 19 2026"
  tags={["Features", "Improvements"]}
  rss={{
title: "Inbox sorting, multi-language support, large file handling",
description:
"New inbox filtering and sorting options, platform available in English, German, and Italian, and improved handling of large attachments."
}}
>
  ## Features

  * Filter inbox by unread tickets and sort by creation date, last message, or
    smart sorting.
  * Platform interface now available in English, German, and Italian.
  * Attachments too large for processing now display with download links instead
    of failing.

  ## Improvements

  * Tickets marked as urgent are now pinned to the top of the inbox.
  * Improved messaging when handing off conversations to human agents.
  * AI responses now properly match the conversation language.

  ## Bug Fixes

  * Fixed inbox sorting by creation date showing incorrect order.
  * Fixed ticket status not updating immediately after approving a message.
  * Fixed inbox view disappearing after page refresh.
  * Fixed handoff failures when tickets contain large attachments.
</Update>

<Update
  label="January 8 2026"
  tags={["Features", "Improvements"]}
  rss={{
title: "Error alerts, real-time collaboration, quick assign",
description:
"Proactive error notifications, see who's viewing tickets, keyboard shortcuts for assignment, and numerous bug fixes."
}}
>
  ## Features

  * Show an alert and send emails when critical errors occur (handoff failures,
    email issues, message drops).
  * See in real-time if another agent is viewing the same ticket.
  * Press A to quickly assign tickets to yourself or others.
  * Assignment now means ownership without taking over from AI.

  ## Improvements

  * Chat widget attachments limited to 15 MB for reliable processing.
  * Signature translations popup now supports variables and highlights.
  * Unsupported variables in signatures highlighted with a warning.

  ## Inbox & Tickets

  * Fixed tickets reappearing without updated messages after AI processing.
  * Fixed team briefly appearing as "Deleted team".

  ## Email & Messaging

  * Fixed cursor jump when editing AI messages.
  * Prevented draft flicker when sending edited messages.
  * Chat widget now scrolls correctly when new messages arrive.
  * Fixed handling when email addresses are empty.
  * Fixed double handoff messages appearing sequentially.

  ## Editors

  * Fixed glitchy behavior at start of bullet points in instructions editor.
  * Fixed non-functional link button in editor.

  ## Integrations

  * Fixed MCP connections showing as "unknown" or failing silently.
  * MCPs no longer send tools while in a broken state.
  * Fixed Salesforce field truncation for multi-byte characters.
</Update>

<Update
  label="December 30 2025"
  tags={["Features", "Improvements"]}
  rss={{
title: "CSAT insights, signature translations, chat widget auto-open",
description:
"New CSAT ratings table, automatic signature translations with variable support, and chat widget timing controls."
}}
>
  ## Features

  * Added option to auto-open chat widget after a configurable delay.
  * Added CSAT ratings and comments table to Insights Hub.
  * Enabled automatic translations with override support for signatures.
  * Added variable highlighting and dropdowns to signatures.

  ## Improvements

  * Relative time now shows "Just now" for events under 1 minute.
  * Added "Powered by" logo to outgoing emails.
  * Incoming emails are ignored when the platform address is not in the To field.
  * Renamed Typewise AI to General Specialist.
  * Added ability to sort tickets by ID and created by.

  ## Bug Fixes

  * Fixed endless channel switching and double handoff issues.
  * Resolved an issue where emails would not be sent if company name had a comma.
  * Fixed chat widget color inconsistencies (title and actions now aligned).
  * Fixed multiple Insights Hub bugs and layout overlaps.
  * Fixed conversation history occasionally including reply messages.
  * Fixed title generation requests firing without conversation history.
  * Fixed LinkedIn casing inconsistency across integration views.

  ## Integrations

  * Fixed MCP tool agent automatic field insertion with parameter mappings.
  * Improved MCP timeout handling and retry logic.
</Update>