> ## Documentation Index > Fetch the complete documentation index at: https://docs.tess.im/llms.txt > Use this file to discover all available pages before exploring further. # Product Updates > The latest Tess releases — new features, improvements, and bug fixes. ## Agent Mode integrations, Skills settings, and chat reliability ### New features * **Import skills from a ZIP file.** You can add custom skills to your workspace by uploading a packaged ZIP from Personal Settings, with a dedicated modal and drag-and-drop flow. * **Custom MCP servers work in Agent Mode.** Workspace-configured custom MCP integrations now load in Agent Mode, and tool chips show a clear server and action name instead of an internal label. * **\[Review] Choose which connected apps Agent Mode can use.** When enabled for your workspace, you can pick which connected apps an Agent Mode session may access instead of exposing every app you have ever linked. ### Improvements * **More accessible Skills settings.** Skill cards now include screen-reader labels on toggles and menus, clearer empty states when search finds nothing, and smoother skeleton loading while the grid loads. ### Fixes * **Reasoning model stays selected after your first message.** Picking a reasoning effort level before you send no longer resets the in-chat model chip to Auto. * **Knowledge-base uploads finish training.** Files added to an agent knowledge base no longer get stuck in an endless loading state or block you from adding the next file. * **Bulk file downloads no longer sign you out.** Downloading all generated chat files handles an expired session gracefully instead of logging you out of the platform. * **Agent steps no longer show false failure notices.** Step-failure alerts appear only when a real step error occurred, so successful runs are not marked as failed. * **Follow-up messages keep context after a cut-off reply.** In economy mode, continuing a conversation after a truncated answer keeps the original request in context. * **Clear guidance when tool calls fail repeatedly.** If tools keep failing in the same turn, you now get an explanation and a chance to confirm before the agent tries again. * **Chat no longer blocks on empty input retries.** Retrying or sending a message with an empty input field no longer shows a blocking error toast. ## Chat history limits and subscription credit refills ### Improvements * **Conversation history fallback now respects your token budget.** If the history indexer cannot summarize older messages, Tess trims what gets sent to the model to match your mode limits and shows a brief "Summarizing conversation history" status so you know the chat is using a fallback path. ### Fixes * **Monthly credit refills on subscription renewals.** Fixed a regression where some plan renewals did not apply your monthly credit top-up after billing. ## Chat, Cowork, connectors, and file handling ### Improvements * **Model detail tooltips open after a short pause.** Hovering through the model list no longer triggers accidental pop-ups; panels open only after you pause on an option for half a second. * **Clearer message when a model times out.** When a model fails to respond in time, you see plain guidance to try another model and send again instead of a technical error. * **Browser extension handles large pages.** Heavy web pages are saved to your workspace so agents can work with the full content instead of truncated snippets. * **Cowork warns when connectors need reconnect.** Mesh nodes and the agent drawer show when a connector lost authentication, with a direct path to reconnect before more runs fail. * **Usage export reflects Cowork activity.** CSV exports now label Cowork-backed runs with the Digital Employee name so usage reports match what you see in the product. ### Fixes * **Chat stays responsive through reconnects and tool runs.** Reconnecting no longer leaves the composer out of sync, and tool chips appear as the response finishes instead of only after a refresh. * **Dense PDFs and large text files ingest reliably.** Knowledge-base PDFs with dense pages and large TXT attachments now process successfully instead of failing silently or showing a generic error. * **Connector setup respects your team policies and app type.** Allowed connectors no longer get blocked by internal integration checks, and OAuth setup shows the right app-specific guidance instead of generic copy. * **Spaces filters show the human owner you selected.** Filtering by user now highlights the person you picked as the primary owner while keeping the creating Autopilot as secondary context. * **Cowork runs stay mapped and readable.** Orphan runs bind to the correct execution, live tables no longer collect protocol residue as data rows, and you see when an agent is waiting for your answer with an honest schedule label. * **Active subscriptions keep platform access.** Paying users are no longer locked out when billing-period data is still syncing. * **Earlier API keys visible in settings again.** API keys created before a recent platform update appear in your settings again for review and revocation. ## Chat reliability and subtasks limits ### Improvements * **Max Subtasks limit you set is what chat enforces.** Your Max Subtasks per Prompt preference now matches runtime behavior in chat. When the limit is reached, you see a clear alert with Continue and Adjust Limit, and Continue sends a follow-up so the agent can resume instead of leaving you stuck. ### Fixes * **Chat history opens after interrupted tool runs.** Viewing past conversations no longer crashes when a tool call was interrupted before it finished, including shared audit views of other members' chats. ## Credit caps, chat completion, and memory privacy ### Fixes * **Accurate monthly credit cap totals.** Per-member monthly credit caps now show consumption that matches the usage report export, including tool, file, and knowledge-base charges attributed to the right team member. * **Complete answers when responses hit length limits.** Chat agents that reach the model output limit now automatically continue generating instead of leaving you with a mid-sentence partial answer. * **Safer memory retrieval in chat.** Memory search now only pulls from memories you own, so unrelated memories cannot appear in your conversations. ## Chat polish, connectors, Agent Studio, and API reliability ### Improvements * **Richer assistant message formatting.** Answers in the main chat now render headings, lists, tables, and code blocks with cleaner typography, better spacing, and a copy button on code snippets. * **Lighter chat startup.** Opening chat loads fewer duplicate scripts, so the page boots faster without changing how credit warnings or session behavior work. * **Extended grace period for API workspace identification.** Integrations that have not yet added the required workspace header to API calls continue to work through August 2026, giving you more time to update without unexpected request failures. ### Fixes * **Gemini 2.5 Pro error recovery.** Recoverable connection or provider issues no longer surface as a blocking technical error or leave the send button stuck when using Gemini 2.5 Pro in chat. * **Agent visibility indicator.** The checkmark in the agent visibility dropdown now matches the option you selected instead of highlighting the wrong sharing level. * **Accurate execution status during long connector runs.** Agent runs that are still processing connector actions are no longer marked as failed in usage history while work is still in progress. * **Generated file preview on existing chats.** Clicking a generated-file chip in a message opens the preview modal when you open an existing conversation from the sidebar, a link, or a refresh. * **Gmail connector tool availability.** Agents using Gmail connectors can reach draft and send actions again when outdated tool names were blocking execution. * **Agent Studio knowledge base file limits.** The upload area now shows the real capacity, warns when you reach the limit, and lets you choose which files to keep when a batch exceeds it. * **Connector resilience after repeated tool errors.** Repeated format errors on one connector action no longer disable the entire integration for the rest of the run. * **Site publishing from agents.** Publishing a site from agent chat works again without authentication errors blocking registration. ## Chat reliability, connectors, cowork, and faster page loads ### Improvements * **Faster initial page load.** The app now loads chat tool panels only when you need them, so every authenticated screen starts quicker without changing how tools behave in chat. ### Fixes * **Clean chat message copy.** Copying a chat answer now pastes with proper formatting and no dark background or raw markdown symbols when you paste into email, documents, or other apps. * **Reliable daily credit refills.** Daily credits now refill according to your subscription owner's time zone, so you receive your allowance on the expected day. * **Reliable Gmail bulk sending.** Bulk email sends through the Gmail connector now report accurate per-message results even when responses are truncated. * **Gmail attachments in chat.** Agents can send emails with workspace file attachments through Gmail without misleading authorization errors. * **Cowork work instructions and status.** Magic Prompt no longer silently truncates work instructions at the character limit, and employee status in the cowork mesh updates correctly when you hover over agents. * **Cowork connector reconnection.** Digital employees stop prompting you to reconnect integrations that are already connected, ending infinite reconnection loops in cowork chat. * **Workspace invite access.** Invited users who log in through an invite link now land in the correct workspace with clearer session choice messaging. * **Audio generation in Portuguese.** The Audio Generator completes voice-over generation for Brazilian Portuguese voices instead of stalling after you submit your script. * **Skills settings layout.** Long skill names no longer block toggles for other skills in your personal settings grid. * **Pipedrive connector setup.** You can connect Pipedrive to your workspace again when API key authentication was failing during credential validation. * **Computer image preview.** Images generated by the agent in chat display inline in the Computer panel instead of showing an unavailable preview. * **\[Review] Preventive media credit caps.** When monthly credit caps are enabled for workspace members, image, video, and voice generations are blocked before they start if the estimated cost would exceed the member's remaining allowance. ## Chat reliability, tool governance, and generator uploads ### New features * **Large chat pastes become attachments.** When you paste a long block of text into the chat composer, it automatically becomes a file attachment instead of flooding the input or hitting character limits. With Smart Search enabled, very large pasted files are indexed so the assistant can retrieve relevant sections instead of loading the entire document every turn. * **\[Review] Agent tool group governance.** Workspace admins can restrict which chat tool groups—such as web search, image generation, and integrations—each team or member may use, matching the existing models and connectors governance controls. * **\[Review] Browser extension for local automation.** Connect a Chrome extension to run browser tasks on your own machine during chat, with layered access controls for teams that enable this capability. ### Improvements * **Clear recovery when providers stall.** If an AI provider accepts a connection but stops sending tokens, chat now times out after 90 seconds with a dedicated message and quick actions to switch models or retry, instead of hanging silently for long stretches. ### Fixes * **Video and image generators with uploaded references.** Playground agents that use reference uploads—such as video upscaling—now send direct file links to external providers instead of internal redirects that caused generation to fail. * **Tool chips and thinking panel in chat.** Connector results no longer leak raw JSON into the Thinking panel during streaming, tool chips stay in the right place after refresh, and the chat header shows Thinking and Responding while the assistant is still writing. * **Image analysis for mislabeled files.** Chat vision analysis now detects the real image format from file bytes instead of trusting the stored file type, preventing provider errors when a file extension does not match its contents. * **Workspace files modal during refreshes.** Selecting files in the workspace files modal no longer closes the panel when the file list refreshes in the background after an agent response. * **Governance access to private workspace files.** Workspace managers with document-read governance permission can open colleagues' private files through document references, consistent with access rules already applied to chat execution results. * **Tool restrictions across all categories.** When admins limit which tool groups a member can use, those restrictions now apply consistently across every category in the chat tool picker instead of mixing fresh and cached results. ## MCP server and API integrations ### Improvements * **Cleaner MCP tool catalog.** The hosted MCP server now lists only documented public API tools, so you no longer see internal or non-working endpoints when connecting agents through MCP. ## Workspace governance, COWORK chart, and digital employee attachments ### New features * **Role-based delete controls.** Workspace admins can grant or revoke, per role, who may permanently delete chats and files, agents and automations, or memories — delete buttons hide when permission is missing, and the API blocks unauthorized attempts. ### Fixes * **Full-org COWORK chart loads reliably.** Switching Mission Control to the whole-organization view no longer hangs on loading layers or fails on large teams; failed loads now show a clear error instead of spinning forever. * **Attached files used in omnibar tasks.** Digital employees now use attached documents even when your instruction is broad (for example, summarize this), reading from the start of each file instead of behaving as if nothing was attached. ## Connectors, workspace files, chat reliability, and voice billing ### New features * **Full connector results saved for agent analysis.** When a connected app returns a very large dataset, Tess saves the complete response to your conversation workspace so agents in Agent Mode can analyze the full payload with code instead of working with a truncated excerpt. ### Improvements * **Faster Connectors modal in chat.** Opening and searching the Connectors catalog loads more quickly, the list stays visible while you type, and revisiting the modal in the same session shows your last results instantly. * **Clear block when credits run out.** If your workspace has no credits left, you cannot start a new voice call and see a clear message explaining why. * **Voice usage in your usage report.** Completed voice calls appear in the workspace usage report alongside other credit-consuming activity. ### Fixes * **Knowledge base uploads from chat work again.** Adding a chat attachment to your knowledge base completes successfully instead of failing before the upload starts. * **Set as home agent in Agent Studio works again.** Defining an agent as your workspace home agent from the studio actions menu now completes the action instead of failing silently. * **Agent-generated documents preview correctly.** Documents created by agents such as spreadsheets, PDFs, and presentations now open with the correct format instead of downloading as generic binary files. * **Agent-created files appear in chat reliably.** Files your agent produces during a turn now show up as downloadable cards in the conversation even when intermediate tool steps mix success and failure. * **Chat recovers after agent workspace images.** Conversations no longer fail on every reply after an image from agent tools enters the thread—follow-up messages send normally again. * **Voice calls now consume workspace credits.** Digital employee voice calls deduct credits from your workspace based on actual token usage, so usage aligns with the configured voice pricing. * **Each voice call is billed only once.** Ending the same call twice no longer risks a duplicate charge. * **Agent runs and image generation finish promptly.** API executions that wait for completion and image generation confirm screens now return as soon as the work finishes instead of hanging for several minutes. ## Agent Mode, chat reliability, reasoning display, and performance ### Improvements * **Clearer Agent Mode credit requirements.** When your workspace balance or personal monthly cap is too low for Agent Mode, chat explains why you cannot switch modes and what you can do next instead of showing a generic error. * **Polished reasoning trail in chat.** When a reasoning-capable model answers, you see a smooth "Thinking…" pulse with a live reasoning trail that collapses into "Thought for Xs" once reasoning ends—the same design after a page reload. * **Faster files panel in chat.** Opening the files and knowledge base panel loads your workspace files much more quickly, especially in large workspaces. * **Faster chat history loading.** Agent conversation lists load promptly instead of hanging on workspaces with many embedded chats. ### Fixes * **Embedded chat history for your workspace.** Workspace members can find and open embedded anonymous visitor chats in usage history and governance views, with links that open under the correct workspace. * **Removing chat attachments works again.** Detaching a file from the compose box deletes it on the server as expected instead of only hiding the thumbnail. * **Chat history opens reliably on first click.** Switching conversations from the sidebar or search results no longer fails silently right after the page loads. * **Shareable chat clones work again.** Cloning a conversation from a link completes successfully instead of stopping with an error. * **No empty bubble while the model thinks.** The agent name and answer bubble appear only when visible answer text starts streaming—you no longer stare at an empty named bubble during reasoning. ## Chat streaming, usage history, checkout, and agent embedding ### Improvements * **Smoother chat streaming.** Responses arrive in the right order, tool markers land in sensible places, and long conversations keep rendering smoothly without freezing mid-generation. * **Richer usage history filters.** You can filter usage by connector calls, tool runs, file uploads, transcription, knowledge-base indexing, background processing, and more when reviewing workspace consumption. * **Connectors in API agent runs.** Agents triggered through the API or hosted MCP server now activate configured workspace connectors the same way they do in chat. ### Fixes * **Usage history matches deducted credits.** Background charges such as auto-route, search planning, and memory jobs now appear on the correct execution rows, so reported usage aligns with credits actually consumed. * **Reliable checkout activation.** Completing a paid plan checkout consistently delivers your subscription and credits instead of failing silently after payment. * **Clearer agent embed setup.** The embedding configuration modal shows a readable title, copy buttons guide you to enable public access first, and anonymous embed visitors no longer see irrelevant credit warnings. ## Chat streaming, usage history, checkout, and agent embedding ### Improvements * **Smoother chat streaming.** Responses arrive in the right order, tool markers land in sensible places, and long conversations keep rendering smoothly without freezing mid-generation. * **Richer usage history filters.** You can filter usage by connector calls, tool runs, file uploads, transcription, knowledge-base indexing, background processing, and more when reviewing workspace consumption. * **Connectors in API agent runs.** Agents triggered through the API or hosted MCP server now activate configured workspace connectors the same way they do in chat. ### Fixes * **Usage history matches deducted credits.** Background charges such as auto-route, search planning, and memory jobs now appear on the correct execution rows, so reported usage aligns with credits actually consumed. * **Reliable checkout activation.** Completing a paid plan checkout consistently delivers your subscription and credits instead of failing silently after payment. * **Clearer agent embed setup.** The embedding configuration modal shows a readable title, copy buttons guide you to enable public access first, and anonymous embed visitors no longer see irrelevant credit warnings. ## Usage history, agent governance, and connector reliability ### New features * **Unlisted agent governance for managers.** Workspace managers with the right permissions can list, edit, and publish colleagues' unlisted agents in Agent Studio while creators keep full control of their own agents. ### Improvements * **Richer workspace usage API responses.** Token counts, model names, and execution mode (agent or chat) are now included in each item returned by the workspace usage API, so admins and integratioworkspace usage API responses. Token counts, model names, and execution mode (agent or chat) are now included in each item returned by the workspace usage API, so admins and integrations can distinguish agent runs from chat and investigate billing more accurately. * **Clearer anonymous visitors in usage history.** Embedded chat visitors display as anonymous users in usage reports and exports instead of confusing placeholder email addresses. * **Clearer referral messaging.** The referral button now invites you to recommend Tess to a friend, with updated reward copy in all supported languages. ### Fixes * **Monday.com connector setup.** Step-by-step guidance appears when your workspace is missing prerequisites, with clearer recovery when authorization is interrupted. * **Model info See More link.** The See More link in model tooltips only appears when it points to useful documentation, not icon or image URLs. * **Agent form and AI generator pages.** Fixed an error that blocked opening agent create and edit forms and some AI generator pages. ## Billing governance, downloads, chat stability, and agent publishing ### Improvements * **Scheduled agents pause on cancellation.** When a workspace owner's subscription is cancelled, scheduled agents in that workspace are automatically paused so they do not keep running and consuming credits. ### Fixes * **Chat after image attachments.** Fixed a recurring provider error in conversations that included certain image attachments, so follow-up messages send reliably again. * **Skill and file downloads.** Restored working download links for skills, workspace files, and files attached in chat. * **Site publishing from agents.** You can publish sites from agent chat again without configuration errors blocking the action. * **Prepaid credit purchases.** Fixed a failure that could occur when updating prepaid credits after a subscription payment. ## AI Studio, digital employees, usage history, and chat reliability ### New features * **Talent Pool digital employees.** Browse pre-built digital employees with app connections already configured, so you can deploy useful agents faster without wiring integrations from scratch. * **Updated image and video models in AI Studio.** Generator models in AI Studio now stay in sync with the latest available options, keeping creative workflows current. * **Agent Studio access for lite users via Team Board.** Lite plan users can now open Agent Studio when entering through Team Board, expanding self-serve agent discovery without a full upgrade. * **Richer usage history for large workspaces.** Usage history loads more reliably at scale and supports dynamic filters by execution type, making audit and cost review easier for admins. ### Improvements * **Clearer chat attachments in messages.** Attached files now always render in the user message area, so shared context stays visible throughout the conversation. ### Fixes * **Image upscale stability.** Fixed a crash when using 2x upscale in the image generator. * **Phantom tool calls with auto tools off.** Resolved cases where tool actions appeared even when automatic tool selection was disabled. * **Playground gallery after generation.** The gallery view now returns correctly once an image or video generation completes. * **Duplicate workspace owners.** Prevented edge cases that could assign more than one owner during workspace setup. * **Connector tools with empty settings.** Connectors with blank configuration schemas now load correctly in chat instead of failing silently. ## Memory import and chat stream reliability ### New features * **Memory import from other AI tools.** Import memories from other AI providers during onboarding or from preferences, with progress tracking and a follow-up in chat. ### Fixes * **Chat stream resumption.** Improved reliability when resuming a chat stream after missing elements in the conversation view. ## Chat thinking state, attachments, and UI polish ### New features * **Paste long text as attachment.** The composer turns large pasted text into a clean attachment instead of flooding the input box. ### Improvements * **Chat thinking state and agent reveal.** While you wait for a reply, you see the Tess logo with rotating status messages and a smooth transition to the agent identity when streaming starts. * **Onboarding invite share on dark theme.** Invite share dropdown is readable on dark UI. * **Support reply indicator.** A Gleap bubble appears when support replies are unread. ### Fixes * **Role dropdown stays open while scrolling.** Role menu no longer closes on internal scroll. * **Navbar layout routing.** Correct routing for legacy and v2 navbar layouts. * **Subscription shortcut permissions.** Settings hide the subscription shortcut without billing write permission. ## Workspace roles, chat files, SSO, and connectors ### New features * **Custom workspace roles.** Workspace owners can create roles from templates, with permissions that respect your plan and stay consistent when the feature catalog changes. * **Knowledge Base in the files modal.** Knowledge Base files and links now live alongside chat files in one unified modal. ### Fixes * **Enterprise SSO login.** Fixed failures that blocked SSO login completion for some enterprise tenants. * **Connector reconnect in chat.** When a connector session expires, the chat shows a reconnect chip instead of failing silently. * **SemRush connector errors.** Connection errors from the provider are surfaced clearly when linking SemRush. ## Background LLM, uploads, and file downloads ### Improvements * **Background LLM fallback.** Background LLM work can fall back across services when the primary path fails. ### Fixes * **Clear errors for bad audio uploads.** Invalid transcription uploads return structured errors instead of opaque failures. * **Downloads for agent-generated Office files.** PPTX, DOC, and XLS produced by agents download correctly from chat. * **Legacy Space file downloads.** Legacy knowledge-base files in Spaces expose a working download URL. * **\[Review] Memory Collections identifier validation.** Invalid collection IDs should fail gracefully instead of a generic server error—confirm customer-facing API impact. ## Chat reliability and connector tools ### Fixes * **First-message chat step errors.** The first message in a chat no longer fails with a broken step error. * **Connector tools in Digital Employee editor.** Integration and connector tools load fully in the digital employee editor and voice discovery. ## Chat attachments and wallet clarity ### Improvements * **Compact markdown images in chat.** Images in assistant markdown render as small link chips so long threads stay readable. ### Fixes * **Magic Prompt and Background LLM reliability.** Magic Prompt no longer fails with server errors, and model customization works in background LLM runs. * **Wallet state for uncapped members.** Members without a personal usage cap no longer see stale open-usage wallet state. * **Smarter chat file context.** The latest edited file and current-turn smart-search attachments are passed to the model so answers match what you attached. * **\[Review] Agent version baseline for legacy agents.** Backfill adds a v1 baseline for agents missing version history—confirm whether workspace-visible behavior changes. * **\[Review] Chat credit calculation guard.** Guards credit calculation when chat context is incomplete—confirm user-visible billing impact. ## Usage history, agent governance, and connector reliability ### New features * **Unlisted agent governance for managers.** Workspace managers with the right permissions can list, edit, and publish colleagues' unlisted agents in Agent Studio while creators keep full control of their own agents. ### Improvements * **Richer usage history details.** Token counts, model names, and execution mode now appear in workspace usage history so admins can distinguish agent runs from chat and investigate billing more accurately. * **Clearer anonymous visitors in usage history.** Embedded chat visitors display as anonymous users in usage reports and exports instead of confusing placeholder email addresses. * **Clearer referral messaging.** The referral button now invites you to recommend Tess to a friend, with updated reward copy in all supported languages. ### Fixes * **Monday.com connector setup.** Step-by-step guidance appears when your workspace is missing prerequisites, with clearer recovery when authorization is interrupted. * **Model info See More link.** The See More link in model tooltips only appears when it points to useful documentation, not icon or image URLs. * **Agent form and AI generator pages.** Fixed an error that blocked opening agent create and edit forms and some AI generator pages. ## Billing governance, downloads, chat stability, and agent publishing ### Improvements * **Scheduled agents pause on cancellation.** When a workspace owner's subscription is cancelled, scheduled agents in that workspace are automatically paused so they do not keep running and consuming credits. ### Fixes * **Chat after image attachments.** Fixed a recurring provider error in conversations that included certain image attachments, so follow-up messages send reliably again. * **Skill and file downloads.** Restored working download links for skills, workspace files, and files attached in chat. * **Site publishing from agents.** You can publish sites from agent chat again without configuration errors blocking the action. * **Prepaid credit purchases.** Fixed a failure that could occur when updating prepaid credits after a subscription payment. ## AI Studio, digital employees, usage history, and chat reliability ### New features * **Talent Pool digital employees.** Browse pre-built digital employees with app connections already configured, so you can deploy useful agents faster without wiring integrations from scratch. * **Updated image and video models in AI Studio.** Generator models in AI Studio now stay in sync with the latest available options, keeping creative workflows current. * **Agent Studio access for lite users via Team Board.** Lite plan users can now open Agent Studio when entering through Team Board, expanding self-serve agent discovery without a full upgrade. * **Richer usage history for large workspaces.** Usage history loads more reliably at scale and supports dynamic filters by execution type, making audit and cost review easier for admins. ### Improvements * **Clearer chat attachments in messages.** Attached files now always render in the user message area, so shared context stays visible throughout the conversation. ### Fixes * **Image upscale stability.** Fixed a crash when using 2x upscale in the image generator. * **Phantom tool calls with auto tools off.** Resolved cases where tool actions appeared even when automatic tool selection was disabled. * **Playground gallery after generation.** The gallery view now returns correctly once an image or video generation completes. * **Duplicate workspace owners.** Prevented edge cases that could assign more than one owner during workspace setup. * **Connector tools with empty settings.** Connectors with blank configuration schemas now load correctly in chat instead of failing silently. ## Memory import and chat stream reliability ### New features * **Memory import from other AI tools.** Import memories from other AI providers during onboarding or from preferences, with progress tracking and a follow-up in chat. ### Fixes * **Chat stream resumption.** Improved reliability when resuming a chat stream after missing elements in the conversation view. ## Chat thinking state, attachments, and UI polish ### New features * **Paste long text as attachment.** The composer turns large pasted text into a clean attachment instead of flooding the input box. ### Improvements * **Chat thinking state and agent reveal.** While you wait for a reply, you see the Tess logo with rotating status messages and a smooth transition to the agent identity when streaming starts. * **Onboarding invite share on dark theme.** Invite share dropdown is readable on dark UI. * **Support reply indicator.** A Gleap bubble appears when support replies are unread. ### Fixes * **Role dropdown stays open while scrolling.** Role menu no longer closes on internal scroll. * **Navbar layout routing.** Correct routing for legacy and v2 navbar layouts. * **Subscription shortcut permissions.** Settings hide the subscription shortcut without billing write permission. ## Workspace roles, chat files, SSO, and connectors ### New features * **Custom workspace roles.** Workspace owners can create roles from templates, with permissions that respect your plan and stay consistent when the feature catalog changes. * **Knowledge Base in the files modal.** Knowledge Base files and links now live alongside chat files in one unified modal. ### Fixes * **Enterprise SSO login.** Fixed failures that blocked SSO login completion for some enterprise tenants. * **Connector reconnect in chat.** When a connector session expires, the chat shows a reconnect chip instead of failing silently. * **SemRush connector errors.** Connection errors from the provider are surfaced clearly when linking SemRush. ## Background LLM, uploads, and file downloads ### Improvements * **Background LLM fallback.** Background LLM work can fall back across services when the primary path fails. ### Fixes * **Clear errors for bad audio uploads.** Invalid transcription uploads return structured errors instead of opaque failures. * **Downloads for agent-generated Office files.** PPTX, DOC, and XLS produced by agents download correctly from chat. * **Legacy Space file downloads.** Legacy knowledge-base files in Spaces expose a working download URL. * **\[Review] Memory Collections identifier validation.** Invalid collection IDs should fail gracefully instead of a generic server error—confirm customer-facing API impact. ## Chat reliability and connector tools ### Fixes * **First-message chat step errors.** The first message in a chat no longer fails with a broken step error. * **Connector tools in Digital Employee editor.** Integration and connector tools load fully in the digital employee editor and voice discovery. ## Chat attachments and wallet clarity ### Improvements * **Compact markdown images in chat.** Images in assistant markdown render as small link chips so long threads stay readable. ### Fixes * **Magic Prompt and Background LLM reliability.** Magic Prompt no longer fails with server errors, and model customization works in background LLM runs. * **Wallet state for uncapped members.** Members without a personal usage cap no longer see stale open-usage wallet state. * **Smarter chat file context.** The latest edited file and current-turn smart-search attachments are passed to the model so answers match what you attached. * **\[Review] Agent version baseline for legacy agents.** Backfill adds a v1 baseline for agents missing version history—confirm whether workspace-visible behavior changes. * **\[Review] Chat credit calculation guard.** Guards credit calculation when chat context is incomplete—confirm user-visible billing impact. ## Chat file references and image reliability ### Fixes * **Image chip in agent mode.** Fixed cases where an image attachment chip disappeared after being selected. * **Inline document and image links.** Document and image references in chat messages resolve and display correctly, including in agent mode. * **Image compare mode.** Restored compare mode for image inputs that use durable file references. * **\[Review] Model selector search highlight.** Fixed misaligned row highlight when filtering models in the grouped chat selector. * **\[Review] File details without storage URL.** Opening file details no longer crashes when optional storage URLs are missing. ## Autopilot approvals and connector actions ### New features * **Batch decision queue.** When an autopilot run needs your approval on connector actions, review them in a card deck—approve, edit, reject, or skip each item, then submit the whole batch at once. ### Improvements * **Clearer autopilot work modes.** Work modes are labeled Autonomous and Supervised, with Supervised as the default. You can also choose a manual-only schedule that runs only when you trigger it. * **Ghost plan timeline.** Pending approval steps appear on an interactive timeline so you can see where each decision sits in the run. ### Fixes * **LinkedIn publishing detection.** Safer detection when an autopilot action targets LinkedIn publishing. * **User input in run results.** Answered user-input requests now show every field in the run output, not just the first one. ## Advances in documents, chat, and stability ### New features * **Smart document linking.** You can now connect and organize documents in a more integrated way, making it easier to continue and retrieve information. * **Content recovery commands.** New commands allow you to restore or adjust content automatically, improving information management. ### Improvements * **More consistent display of actions in chat.** Adjustments to how interactions and tools are presented, making usage clearer and more predictable. * **Content listing performance.** Optimizations that make navigating and loading lists faster. ### Fixes * **Server error in specific chats.** Fixed an issue that prevented usage in some cases, ensuring greater stability during conversations. ## Onboarding fixes and improvements in model selection ### Improvements * **Enhanced model selection (conditional access).** A new way to choose models with reasoning level options, allowing greater control over response behavior. ### Fixes * **Onboarding flow.** Fixed an issue that allowed redoing onboarding and generating duplicate agents. ## Improvements in credits, chat, and stability ### New features * **Enhancements in on-screen element detection.** Improvements in how content is dynamically loaded and displayed, providing smoother navigation. ### Improvements * **More reliable credit top-ups.** Adjustments to the payment system to ensure credits are added correctly. * **Clearer messages and feedback on login.** Better communication when redirecting users during account access. * **Interface stability and consistency.** General improvements that make the experience more predictable and free of visual inconsistencies. ### Fixes * **Sending messages in agents.** Fixed an issue that prevented sending after filling required fields. * **Magic Prompt restored.** Functionality is now working correctly again in chat and in the image generator. * **Display of long messages in chat.** Fix to automatically collapse long content without needing to refresh the page. * **Use of images and tools in models.** Adjustments to ensure correct operation in multimodal interactions. * **Duplicated interface elements.** Fixed duplicated icons in settings menus. * **Interface and dependency updates.** Fixed inconsistencies that caused failures or unexpected errors. * **Credit processing.** Fixed conflicts that could cause inconsistencies in the user balance. ## Improvements in chat, permissions, and stability ### New features * **Access control by permissions and credits.** The use of features like content generation now correctly respects permissions and limits available in the workspace. ### Improvements * **Drag-and-drop upload in chat.** Updated interface to make file uploads clearer and more consistent. * **Enhanced chat experience.** New quick actions and better visual organization of long messages. * **Sending large files.** Greater reliability when working with heavy files in chat. * **Consistency in workspace selection.** Adjustments to ensure the displayed environment matches the selected one. ### Fixes * **Chat message field.** Fixed visual cuts and limitations when typing long content. * **Tool selection.** Resolved an issue that prevented unselecting tools after the first interaction. * **Scrollable forms.** Fix to allow proper navigation on screens with many fields. * **File generation.** Resolved a freeze that caused indefinite loading. * **Insufficient credits.** Fixed inconsistent behaviors and loops when reaching the limit. * **Price display on mobile.** Adjusted incorrect values when navigating between plans. ## Searchable connectors, chat cloning with attachments, and fixes ### New features * **Searchable connectors list.** The connectors panel now has a live search with incremental loading so you can find the right connector faster, even when you have many integrations connected. * **Clone chat with attachments.** You can now optionally include attached documents when cloning a shared chat, so the full context is preserved in the copy. ### Bug fixes * **Copy button on code blocks.** The copy button was missing on monospaced code blocks in chat. It is now visible and working again. * **Login and sign-up pages in light mode.** These pages were incorrectly inheriting dark mode styles. They now always render in light mode. ## AI employees, credit cycle alignment, and chat fixes ### New features * **Cancel in-progress agent runs.** You can now stop a digital employee run that is still in progress directly from the run history, without waiting for it to finish. * **Rehire terminated AI employees.** Browse and restore AI employees you previously terminated via the new "Open for Work" option in the Deploy menu. ### Bug fixes * **Credit cycle aligned to your billing date.** Credit cap windows now follow your subscription's actual billing dates instead of calendar months, ensuring caps and renewals match your real billing period. * **Wallet loads instantly.** The wallet icon was spinning indefinitely for many users. The credit usage panel now loads quickly on every account. * **Files stay when you edit a message.** Attached files and tool selection were being cleared when editing a chat message. Both are now preserved when you edit and resend. * **Usage history privacy.** Links to private chats in usage history were accessible to unauthorized viewers. Private executions are now correctly hidden from users who do not have access. ## Cowork canvas in React, digital employee actions, and reliability fixes ### New features * **Cowork canvas rebuilt in React.** The LLM canvas in Cowork has been migrated to React Flow, improving rendering performance and responsiveness during complex multi-agent interactions. * **Digital employees interact with external tools.** Agents can now validate and process external action requests, enabling tighter integration with third-party services during autonomous runs. * **Memory role-gated access.** Memory collection grants now support a minimum membership role, giving workspace managers finer control over who can access shared memory. ### Improvements * **Agent Studio tag filtering.** Tag-based filtering in the Agent Studio catalog now works correctly, making it easier to browse and find agents by category. * **Wallet usage label.** The wallet now shows the correct calendar month label for monthly usage, removing a confusing mismatch. ### Bug fixes * **Scheduler next-run timing.** Scheduled agent runs were drifting from their expected times. Next-run is now stored and computed correctly. * **Account deletion error.** Deleting an account was failing with a generic error for all users. This has been fixed. ## Secret invite links, no-tools mode, connectors view-only, and Spaces fixes ### New features * **Secret invite links.** Workspace members can now be invited via a private link. Accepting a secret link automatically removes any duplicate pending email invitations for the same user. * **No-tools mode.** Admins can now lock the tool selection for an agent to "no tools," and the chat UI clearly communicates when tool selection is restricted by the workspace configuration. * **Connectors view-only mode.** Users without permission to manage connections now see the connectors panel in read-only mode rather than encountering an error. * **Loading indicator on workspace switch.** A visible loading state now appears when switching workspaces, preventing accidental duplicate clicks. ### Bug fixes * **Spaces cross-workspace file leak.** Files from other workspaces were incorrectly appearing in the Spaces personal files list. This is now fixed. * **Provider quota errors in chat.** When an AI provider returns a quota failure mid-stream, the error is now handled gracefully instead of causing a blank or stuck response. * **Connector OAuth sync after popup close.** Closing the OAuth popup after selecting an account was not completing the connection. The sync now finishes correctly. ## Billing and renewal reliability ### Bug fixes * **More reliable credit renewals.** Duplicate renewal prevention and late-cycle backfilling now work correctly, ensuring credits are granted on time without blocking subsequent cycles. * **Coupons applied with correct duration.** Promotional coupons were being applied indefinitely in some cases. They now expire as intended and are correctly resolved across accounts. ## Workspace Artifacts, API Playground billing awareness, and more ### New features * **Workspace Artifacts.** Admins can now manage published artifacts directly from workspace settings — filter, paginate, and toggle publish/unpublish with full permission controls. * **API Playground billing awareness.** The playground now warns you before a generation that may count against your billing limits, so you can decide before any charges apply. * **Advanced settings multi-select.** Fields in the Advanced Settings panel now support selecting a subset of options, giving you finer control over agent configuration. * **Pending workspace invitations.** Admins can now see pending invitations alongside active members and the add-member panel prevents duplicate invites. ### Bug fixes * **More reliable subscription renewals.** Some subscriptions were failing to renew correctly in specific billing-cycle timing scenarios. Your plan now renews as expected. * **Signup on iOS Safari.** The email step of the signup modal was broken on iOS Safari. This is now fixed. ## Workspace teams, digital employees, MCP plugins, and connectors ### New features * **Workspace Teams.** Group workspace members into teams, assign team-level memory settings, and control which agents and tools each team can access. * **Digital Employees.** Agents can now run on a schedule, wake on events, wait for human approval on specific steps, and leave a visible activity trail — all within spending limits you define. * **MCP Plugins.** Workspaces can register trusted external tool servers. Agents in chat and autonomous runs can invoke those tools directly. * **Connectors dropdown in React.** The connectors selector in the chat input area is now built in React for a faster and more consistent experience across the platform. ### Bug fixes * **Chat history context budgeting.** Isolated message tokens are now used for history budgeting, preventing context overflow in long conversations. * **Mobile chat and menu.** Several mobile layout issues in the chat interface — bottom sheets, tool tooltips, and menu rendering — have been resolved. * **Subscription renewal timing.** An edge case where the billing cycle anchor caused renewal failures due to a Stripe API change has been fixed. ## Custom workspace roles and private agent governance ### New features * **Custom workspace roles.** Workspace managers can now create, edit, and delete custom roles with specific permission sets tailored to their team's needs — such as a content reviewer with restricted access. * **Private agents and results.** Agents and their outputs can now be set to private. Only authorized members can view private agents or access their results, with a clear experience when access is denied. ### Bug fixes * **Skills slash command restricted to agent mode.** The skills slash command (/) was accessible in plain chat mode. It is now correctly restricted to agent mode only. * **Agent type preserved on templates.** The agent type was being reset when saving new templates. It is now correctly preserved. ## Economy Mode, Memory Boost, Skill Library, and chat history filters ### New features * **Economy Mode.** A new chat setting that reduces the token window to lower credit consumption on longer conversations. The interface shows estimated savings as you adjust the setting. * **Memory Boost.** Search your full conversation history and select which past context to include in the current chat, with a model picker for the search step. * **Skill Library.** Skills are now organized by categories with browsable libraries. You can add official skills from the library, preview and edit skill files in the browser, and use slash-command autocomplete to trigger skills in chat. * **Chat history filter.** The chat sidebar now lets you filter your conversation history by execution origin — Platform or API — so you can find the right conversation faster. ### Bug fixes * **Embedded chat infinite errors.** Embedded chat streams were failing with repeated errors due to a cross-origin URL issue. This is now resolved. * **Agent mode context model.** The context model used in agent mode was not resolving correctly in some cases. This has been fixed. ## Multi-agent support, custom credits, and email change ### New features * **Multi-agent as a native model option.** You can now use multi-agent orchestration directly as a model option in chat, with Smart Components and real-time streaming output. * **Custom credit purchases.** Pay As You Go users can now enter a custom credit amount, see a live price estimate, and complete the purchase via Stripe checkout. * **Email change with confirmation.** You can now change your account email address from account settings. A confirmation flow validates the new address before the change takes effect. ### Bug fixes * **Billing recovery window.** Credits and subscription state are now correctly preserved when a renewal happens shortly after a plan cancellation. * **Autofill crash on Instagram and Safari.** A browser autofill conflict that caused crashes on Instagram in-app browser and Safari has been resolved. ## Improvements in billing and currency consistency ### Improvements * **Country and currency settings.** Fixes to ensure users have the correct currency set, avoiding inconsistencies in charges. ## Improvements in agent mode ### Improvements * **Icebreakers in agent mode.** Fix to ensure the selected model is correctly reflected in the URL. ## Settings redesign and consumption history ### Improvements * **Model sync in URL.** The selected model is now correctly reflected when navigating between chat modes. * **Redesigned settings and history.** New interface for personal settings and credit usage view, with better clarity and usability. ## Visibility logic for custom templates ### New features * **Custom template visibility.** Update to display logic to ensure greater control over how and to whom custom templates are shown. ## Improvements in paywall, chat, and member management ### New features * **Coupons in paywall.** Adjustments to ensure price changes with coupons are correctly reflected in the interface. ### Improvements * **Chat streaming stability.** Better error handling and consistency in model display. * **Member role management.** Adjustments to allow more consistent permission changes in the members table. ## Improvements in API, integrations, and ticket management ### Improvements * **Enhanced ticket management.** Improvements in querying and organizing tickets for greater operational efficiency. * **Removal of image editing integration.** Platform simplification by removing unused components. * **More consistent model configuration.** Adjustments in the API and platform to align required properties and avoid inconsistencies. ## Improvements in plan upgrade ### New features * **Improved plan upgrade modal.** Adjustments to display rules and behavior for a clearer experience when changing plans. ## Branding and localization update ### New features * **Branding and localization.** Visual updates and improvements in multi-language support for a more consistent experience. ## Improvements in modals, files, and chat ### New features * **Fixed scroll in modals.** Better navigation and usability in platform modal windows. * **Chat history update.** New chats now appear correctly in history without needing to reload the page. ### Improvements * **Enhanced file classification.** Better identification of file types and more robust error handling. * **General dependency updates.** Small adjustments to improve stability and performance. ## Improvements in onboarding, API, and agent mode ### Improvements * **API Token management.** Fix in token deletion for greater reliability. * **Smoother onboarding.** Adjustments in timing and transitions for a more pleasant experience. * **Localization links.** URL updates for more consistent navigation in settings. * **File handling in agent mode.** Improvements in file upload and processing. * **Subscription management.** Adjustments in subscription resumption flow. * **Workspace name.** Fix in saving the workspace name. * **Agent execution with images.** Fix in image usage to ensure correct association during executions. ## Improvements in chat and fixes with integrations ### New features * **Embedded chat fixes.** Adjustments to improve embed functionality. ### Improvements * **Stability in chats with files.** Fixes to prevent failures when files are not correctly associated in certain flows. ## Improvements in onboarding, interface, and dark mode support ### New features * **Experimental features modal in dark mode.** Visual and usability improvements with full dark mode support. * **Improved workspace dropdown.** Smoother and more consistent navigation when selecting workspaces. * **Compatibility updates with external domains.** Adjustments to support integrations and marketing experiences. ### Improvements * **Onboarding for agent creation.** New more guided and intuitive experience during the initial process. ## Improvements in payments and general adjustments ### New features * **Improved payment event processing.** Enhancements in reliability and traceability of payment-related events. ### Improvements * **General experience adjustments.** Fixes and refinements in specific flows for greater stability. ## Improvements in agent mode and chat ### Improvements * **Agent mode in dark mode.** Visual fixes in modals for better consistency in dark mode. * **Website publishing in agent mode.** Adjustments to ensure publishing works correctly within the flow. ## Affiliate tracking support ### New features * **Affiliate tracking integration.** Added support to enable proper tracking in external integrations. ## Improvements in agent execution and Agent Studio ### Improvements * **Agent execution and playground.** Fixes to prevent improper failures during executions. * **Cleaner Agent Studio.** Non-relevant fields are no longer displayed, making configuration simpler and more intuitive. ## More robust history export ### New features * **CSV history export.** Improvements to prevent timeout failures when exporting large volumes of data. ## Improvements in showcase chat and paywall conversion ### New features * **Attachment indicators in showcase chat.** The chat now displays visual indicators for attached files, with a better experience on mobile devices. * **Paywall conversion experience.** Adjustments to make the journey for non-subscribed users clearer and more efficient. ## Adjustments in plans and overall experience ### New features * **Improved language settings.** Enhancements in language definition and consistency in external integrations. ## New agent mode and reliability improvements ### New features * **New agent mode component.** Introduction of a new dedicated experience for interactions with agents. ### Fixes * **Error logging improvements.** Adjustments in error tracking for greater visibility and reliability. ## Improvements in billing and subscription management ### New features * **Adjusted access to billing modal.** Better control of display for users with different permissions. * **Scheduled cancellation.** Ability to schedule cancellation for a future date with automatic charge suspension. ### Improvements * **Redirect links in history.** Adjustments to make redirects more consistent and reliable. ## Adjustments in collection retrieval ### Fixes * **Collection loading limit.** Fix in collection query limit to ensure stability and proper performance. ## Improvements in payments and memory ### New features * **Subscription price range.** Support for displaying price ranges on the plans screen, providing more clarity to the user. * **Expanded memory collection loading.** Increased limit of retrieved collections to improve access to stored data. ## Improvements in onboarding and voice selection ### New features * **Improved voice selection in chat.** Enhancements in choosing and configuring voices within chat. ## Adjustments in tool selector and chat improvements ### New features * **Tool selector adjustment.** The selector has been simplified to avoid redundant display when multiple tools are available. * **Translation improvements.** Enhanced handling to ignore keys composed only of numbers. * **Unified model resolution.** Update to model selection logic for greater consistency between chat and video components. ### Improvements * **Share button in chat.** Fix in displaying the share button when starting a new conversation. ## Improvements in plans experience and retrospective ### New features * **Plan display in paywall.** Unconfigured plans no longer appear in the carousel, making the experience clearer. * **Adjustments in 2025 retrospective.** Improvements in texts and messages for greater clarity and consistency. ## Improvements in chat and member invites ### Improvements * **Long session alert with dismiss.** You can now dismiss long session alerts directly in chat. ### Fixes * **Member invites on mobile.** Adjustments in modal width and improvements in the experience on mobile devices during invite sending. ## Adjustments in vision model billing ### Improvements * **Vision model billing in chat.** Adjustment in consumption calculation for lite users, ensuring greater accuracy in usage accounting. ## Improvements in chat, subscription page, and general adjustments ### New features * **Improvements in WhatsApp interaction.** Adjustments in button behavior in promotional experiences to make navigation more consistent. ### Improvements * **Voiceover flow in charges and refunds.** Adjustments to make the process clearer, especially in scenarios with prepaid credits. * **Long conversation warning in chat.** Fix for improper persistence of the warning when starting a new chat. * **Clearer subscription page.** Improvements in communication and presentation of plan information. * **Plan naming updates.** References to previous campaigns have been standardized to current plan names. ### Fixes * **Credit purchase modal.** Fix in image display and error handling during the purchase process. ## 2025 retrospective and preparation for Chat V2 ### Improvements * **2025 retrospective.** New interactive experience that presents statistics, achievements, and usage milestones of users and teams throughout 2025.