Claude 4.5 Sonnet is now available to all users who have access to the AI model selector dropdown.
Export conversation as PDF is now available to end-users:
Users can now choose to export the conversation as PDF, expanding on the previously available CSV export.
Conversations can also be exported directly from the Ask page.
PDF has "Powered by CustomGPT.ai" branding tied with the existing branding/whitelabeling feature.
The feature is available to Premium users, and is immediately active on all agents that have had the Export conversation feature enabled.
Customizable mobile font size is now available for all agents:
Builders can now choose whether to use reduced font size on mobile, or use the same as on desktop.
Reduced size is equivalent to previous behavior, and remains default.
The feature is configured for each deployment separately.
Improvements
GPT-4.1 is now supported when Azure is selected as model provider.
It is now possible to sync the entire SharePoint drive. If this option is selected, and auto-sync is enabled, the content of all present and future files will be added to the agent's knowledge base.
A button has been added to the Create page to build an agent using AI Vision.
Phone numbers have been added to the list of elements that Lead Capture agent is actively looking for, and have been included to CSV and Zapier exports.
Lead Capture event on Zapier now contains the Additional Info field which is JSON of all relevant collected information about the user.
The session_id has been added to the payload of Lead Capture event on Zapier.
Queries that used any Agentic action are now clearly marked on both Ask page and Customer Intelligence page, enabling easier usage tracking.
Summary of used Agentic actions has also been added to Customer Intelligence modal.
Updated the layout of links displayed on Customer Intelligence modal.
Custom MCP Action feedback provided by the agent now takes into consideration the language of the user's prompt.
All new agents now get the default persona, even if a specialized persona is not generated for them.
Auto-generated personas have been updated to better reflect the selected agent’s role.
Translations have been added for new out-of-queries messages for both builder and user, respecting the agent’s selected language.
Translations have been added for Document Analyst statuses, respecting the agent’s selected language.
URLs have been updated on the CustomGPT API Starter Kit card on the Developers page.
Messaging on Google Drive Auto-Sync popup is now clearer.
Bug Fixes
Fixed the issue where the Lead Capture action was not collecting leads after the third message in conversation, if used without Drive Conversions action.
The agent loading spinner is now spinning in the right direction.
Fixed the issue where Live Chat deployment was overlapping Website Copilot deployment, if both are present on the screen at the same time.
Fixed the issue preventing some users from connecting SharePoint libraries with their agents.
Fixed the issue where auto-generated starter questions sometimes focused on information like the content of the cookie consent popup.
Fixed the issue with wrong direction on error messages for right-to-left languages.
Fixed the positioning of the loader on the Ask page.
Fixed the width of pagination elements on the Customer Intelligence page.
Fixed the issue with all links not being correctly displayed on Customer Intelligence modal.
Fixed the issue preventing Customer Intelligence data to load on some agents.
Fixed the wrong Focus dropdown color on the Customer Intelligence page.
Removed the unnecessary code from Website Copilot and Instant Viewer code snippets.
Lead Capture Agentic Acton is now available - perhaps our most anticipated feature yet:
The agent will attempt to collect the user's name, email and company, as well as any other information it deems relevant.
Collected leads can be exported as CSV.
Collected leads can be instantly sent to Zapier.
The action can currently be combined with Drive Conversions and/or Webpage Awareness.
Two modes:
If used with Drive Conversions - Lead Capture agent continuously works in the background collecting every bit of information about the user and their company or use case as possible.
If used without Drive Conversions - Lead Capture agent will attempt to collect the lead in the first couple of messages and after that the usual flow will resume, regardless of success/failure.
The feature is available on our Premium plan.
AI Vision Image processing - it is now possible to use advanced AI vision capabilities when processing uploaded images:
AI will analyze the image intelligently, and provide its description and summary to be added to the agent's knowledge base.
The feature should work particularly well for photos, diagrams and charts.
The feature can be used to extract text from images, but for that particular use case it's not necessarily better than OCR and may especially struggle with non-English texts.
Available on all plans, with images up to 2048x2048 pixels. Enterprise users can request higher resolution and increased limits.
AI vision works best in synergy with existing Image Citations feature, opening a lot of new use cases.
AI Vision can be combined with Document Analyst agentic action to compare images.
Improvements
It is now possible to export Risk Metrics together with the rest of Customer Intelligence data points.
Risk Metrics for any given query have been added to the Customer Intelligence modal.
The Customer Support agent role now has numbered references enabled as the default.
API & MCP
The /reports/intelligence endpoint will now return today’s queries if no start and end date are provided.
Added the missing chat_bot_bg_type and chat_bot_bg_color settings to the projects/{agent_id}/settings endpoint.
Bug Fixes
Fixed the issue where spreadsheets were sometimes failing to process.
Fixed the issue where image citations were ignored if the file had an unstandard name (e.g. a lot of spaces, dots and other special characters).
Document Analyst agentic action tooltip now displays the correct amount of spent queries.
Fixed the issue where the agent knowledge processing feedback modal would sometimes get stuck or be completely missing.
Fixed the issue where both Website and Sitemap buttons on the Create page had the same label.
Fixed the issue where Risk Metrics charts on the Analyze page were displaying decimal numbers on Y axis.
Additional Anthropic models have been added to model selection:
Claude 4 Sonnet - equivalent to GPT-4o / 4.1 / 5.
Claude 3.5 Haiku - equivalent to OpenAI mini models.
Models are available for users who have access to models dropdown.
Older Anthropic models have been deprecated for Standard and Premium users, with their agents automatically upgraded to Claude 4 Sonnet.
Google Drive integration update - Recursive Sync
When a folder in Google Drive is selected, the system will now sync that folder and all folders up to 5 additional levels of depth.
Existing connections will automatically sync all sub-folders on their next auto-sync.
Plan Add-Ons - Standard & Premium users can now purchase additional queries, words stored, documents processed, max agents or team seats without upgrading their plan.
Improvements
Status of auto-generated Persona and Starter Questions has been added to the new agent processing feedback modal.
Custom MCP Actions can now handle broader output format coming from MCP server, allowing more versatility.
The selected role of newly created agents now affects the agent's automatically generated persona.
Improved UX of expired token notifications for SharePoint and Confluence agent sources.
An external link icon has been added for all agent’s source types, and clicking on it leads the user to the original source in question.
Loader has been added to the Sources page to ensure that screen isn’t empty before agent’s sources finish loading.
Customer Support agent has been added to the Resources page.
The UI icon for Document Analyst has been updated.
Improved flow for selecting Spotlight Avatar when creating a new agent with Revenue agent role.
Added notice that image citations are not available when GPT-5 is selected as an AI model.
Bug Fixes
Fixed the issue where Google Drive auto-sync folders were being de-selected after auto-sync for some users.
Fixed the issue preventing some users from exporting all conversations from the Ask page.
Fixed the issue where the Export button on the Queries Breakdown chart on the Analyze page wasn’t working.
Fixed the issue where agentic actions used were sometimes included in the numbered references.
Fixed the issue where the UI wasn't refreshing after enabling auto-sync for some source types.
Fixed the issue where the agent's avatar background was black if the uploaded is transparent.
Fixed the issue where PNG files on sitemaps were being processed with a lot of metadata which was confusing to agents.
Fixed issues with re-connecting SharePoint sites with expired tokens.
Risk Analytics suite has been added to Customer Intelligence:
We are now monitoring and reporting on following security events:
Fidelity - is the agent being true to its persona or not - a lot of failures here can indicate that the persona needs improvements.
Jailbreak - tracking of attempted and successful jailbreaks.
Prompt leakage - tracking of attempted and successful system prompt leaks.
Profanity - tracking of nsfw user messages blocked by profanity filters.
New features in the app:
Risk Monitoring section on Analyze Overview tab.
Four new Risk views on Customer Intelligence tab: Fidelity, Jailbreak, Prompt leakage, Profanity.
Filters for all of the above.
Advanced Role-Based Access Control is now available to all users who have access to the CustomGPT for Teams suite:
It is now possible to create a custom user role that applies just to selected agent(s).
This role can have any combination of existing permissions, except account-level permissions, or be limited to chat-only access.
API-key granulation is supported even for users with limited roles, and user can delegate only agents they have access to their own keys
UI has been refreshed for setting up and assigning custom roles, as well as managing team members.
The new layout has been released for the Agents page, with the updated all of the list view, grid view and chat-only user role. Quick buttons are now dynamically chosen based on the agent's state, and take the user where they need to go: Build, Deploy or Analyze.
Improvements
It is now possible to instruct the agent whether or not it should ask for end-user confirmation before using a custom MCP action.
It is now possible to completely turn off the avatar displayed when there's no active conversation.
When using Agent Initiative for Live Chat deployment, it is now possible to choose whether or not to automatically start the conversation when a user clicks on the widget, or simply open the chat window.
When using Agent Initiative for Live Chat deployment, it is now possible to hide the starter question used for the initiative prompt from the starter questions list inside the chat window.
Auto-generated personas now include an instruction to insert a call to action or a hook for the end-user in every response.
The layout has been improved for the Demo agent welcome screen.
A modal with the best practices and limitations of Numeric Search feature has been added to the app.
“Sitemap” has been renamed to “Website” on the Sources page.
Bug Fixes
Fixed the issue where agents were sometimes returning “I don’t know” responses for queries like “Hello”.
Fixed the issue where the context-rich starter question was never displayed inside the chat window if there wasn’t at least one other starter question.
Fixed the issue where the context-rich starter question had slightly different styling from the rest of starter questions.
Add Custom Action dialogue now correctly displays errors for incorrectly filled input fields, and prevents users from proceeding in such cases.
Manage API keys modal will now not automatically close upon deleting an API key.
Fixed the issue with long website URLs breaking the UI on Sources page.
Fixed the issue where the agent would sometimes mention the custom context as one of numbered references.
Fixed the issue with numbered references sometimes being provided in the wrong format.
Numbered references in agents’ responses have been disabled for the GPT-5 model due to sub-par performance.
Fixed the CORS error that prevented some users logged in via IdP to log out of the system.
The default AI model has been upgraded to GPT-4.1, with the following changes and benefits:
Latency is improved.
GPT-4.1 is better at following instructions than GPT-4o.
GPT-4.1 responses will generally be somewhat shorter than GPT-4o responses.
Agents belonging to all users who don’t have access to AI model selector dropdown have been automatically changed to GPT-4.1.
Improvements
The Agent Initiative widget is now displayed even if there is active conversation with the agent.
Bug Fixes
Agent’s prompt input field now correctly displays on mobile devices, including tablets.
Fixed the problem where, if numbered references are enabled, the agent was displaying only one citation, even if multiple were used to build the response.
Fixed the issue where, when the user's account reached processing limit, and limit was removed later,SharePoint auto-sync was ignoring previously limited files if they were not in the meanwhile changed on the SharePoint end.
Fixed the issue where external_id was not always correctly displayed on the Ask page.
Fixed the issue where the error message was missing if domain validation service times out when building an agent using sitemap or website.
Fixed the issue where, sometimes when updating the default payment method, the subscription's payment method didn’t use the new default
MCP server support has been added as the Custom Actions feature:
You can connect any MCP server, opening countless possibilities for both data retrieval and taking outside actions.
Supports all MCP servers that don’t require auth or support token auth.
Up to 3 custom actions are allowed per agent.
The agent decides when the action is called.
Up to 3 action calls are allowed per query.
Each action call costs 1 additional query.
Available on Premium.
Improvements
The new custom SharePoint file picker is now available to users on demand, and is recommended for users who have issues syncing their SharePoint files.
Auto-sync settings modal has been overhauled in order to improve readability and overall user experience.
Agent will now, if agent initiative is enabled and the conversation window closed, prompt the user with the context-rich starter question even if there is an active conversation.
Improved loading performance on Sources page.
Improved the clarity of the error message when the document is not processed due to account limits being hit.
Slight improvements to the Drive Conversions system instructions.
API & MCP
Document Analyst queries and end-user file upload are now supported via API.
Added the new endpoint /user/search-team-member, which enables team owners and admins to search team members by their email or user ID.
Bug Fixes
Fixed Instant Viewer highlighting issues.
Fixed the issue where sometimes old agents were wrongly showing the “Syncing” status.
Fixed the issue on the Ask page, where citations would sometimes overflow along the horizontal axis of the screen.
Fixed the issue where it was impossible to add new websites or YouTube channels to demo agents.
Fixed the issue where clicking to make notification read still redirected the user to notification’s call to action.
Fixed the rare issue where if the system failed to generate the demo agent, the user was not able to proceed to the app.
Fixed the issue where the profile icon wasn’t being displayed properly on the Limits page.
The Numeric Search option is now available to all users:
This feature greatly improves responses when the agent's knowledge base has a lot of items that are alphanumeric codes, which regular RAG greatly struggles with, especially if such data is in spreadsheets.
Examples of these codes:
Product codes (SHOE-1234)
Error codes (ERR1234)
Version numbers (GPT-4.1)
Order numbers (TRACK12345)
Improvements
Build > Sources page has been overhauled in order to improve the overview of agent's sources:
All sources now have their own distinct card.
The design of cards has been standardized.
Only sources added to the project are visible.
New Add Sources dropdown menu with the same flow as on Create page-
The modal on the Explore tab now displays all links (both in-chat and citations) that were provided with the specific message.
Links & clicks analytics charts are now available on the Analyze page Overview tab.
New layout for the Developers page with many newly added resources.
New agent statuses have been added to Agents, in order to provide better feedback if the agent is still processing its documents.
Improved quality of automatically generated Starter Questions.
Improved styling of API key creation modal.
Removed character limit for the in-chat agent title length.
API & MCP
CustomGPT.ai RAG API Starter Kit - A fully functional open source repository that includes multi-instance embedded widgets, floating buttons, and standalone application with one click deployment.
CustomGPT.ai Integrations - A fully functional open source repository that includes starter kits for integrating CustomGPT.ai API with the following platforms: Slack, Telegram, Discord, WhatsApp, Rocket Chat, MS Teams, Instagram, Google Chat and Messenger.
It is now possible to send the custom_context parameter when making a message request via API.
Bug Fixes
Fixed the issue where YouTube auto-sync wasn’t picking up videos that were scheduled before publishing.
Fixed the issue where pages saved with Webpage Awareness feature stopped working after 30 days.
Fixed the issue where PDF was sometimes not being correctly highlighted with the Instant Viewer deployment option.
Fixed the issue where auto-sync settings “Refresh existing pages”, “Remove deleted pages” and “Add new pages” sometimes couldn’t be saved.
Available to all users who have access to model selector dropdown.
Available with Agent Capabilities: Optimal Choice, Highest Relevance, Complex Reasoning
Available with Agentic Actions: Drive Conversions, Website Awareness, Doc Analyst
This is the basic GPT-5 and does not have reasoning capabilities - these options will be added later.
GPT-4o remains the default model for the time being, as we continue GPT-5 testing.
Document Analyst Agentic Action is now available for our Premium users. Additionally, we have expanded support for additional filetypes: PDF, TXT, DOC, DOCX, ODT, JPG, JPEG, PNG and WEBP.
We have added support for API key per-agent granulation, and users can now:
Create an API key that pertains just to a selected agent(s).
Such API keys can't create new agents or do any account-level actions, like view user profile.
Unlike global API keys, these granulated keys can only work within one team!
In-chat avatar controls - Our premium users can now enable or disable both agent’s and user’s avatar within the chat interface.
Improvements
The in-chat text input field now gets automatically in focus on any click inside the agent’s window.
The in-chat text input field now gets automatically in focus as soon as the agent's response finishes streaming.
The in-chat text input field now gets automatically in focus as soon as the agent window is opened for Live Chat, Website Copilot and Instant Viewer deployments.
Links & Sources Analytics page now supports URL parameters for selected filters, enabling easier sharing.
A border has been added to the in-chat agent avatar, when there is no active conversation.
It is now possible to track added query cost on Actions page in more detail.
When activation agentic action, the user is now shown a warning on exactly how much queries using it cost.
Agentic actions on Actions page have been re-ordered.
When context-rich Starter Questions are used outside Live Chat Agent Initiative, the first-person version is displayed to the user now, instead of the agent's direct prompt.
Live Chat Agent Initiative styling has been updated to ensure improved readability on dark backgrounds.
Live Chat icon size on mobile is now always small, to improve readability.
Document Analyst can now correctly read PDFs with user-fillable forms.
Agent role selector copy has been improved for clarity.
The styling of the notification system has been improved.
Bug Fixes
Fixed the issue where clicking on numbered citation wasn’t correctly opening Instant PDF Viewer.
Fixed the issue where user feedback export on the Analyze page was not working.
Fix the issue where CTA to add knowledge to the agent on the Ask page for new users was not always correctly dismissed.
Fixed the issue with the Agent role selector sometimes showing no available options.
Fixed the broken “Learn more” link for Custom Context on the Deploy page.
Fixed the broken “Learn more” link for Agent roles on the Personalize page.
