All-time Customer Intelligence & Analytics data access - Enterprise customers can now request for access to all-time data for all metrics on agent’s Analyze page.
Turn off markdown - It is now possible to disable markdown formatting in agents responses, for agents that are using GPT-4o or GPT-4.1 models.
GPT-4.1 support on Azure - Latest OpenAI model is now available to Enterprise customers using Azure as AI provider.
🛠 Improvements
Copy has been improved for Document Analyst feedback messages.
🐞 Bug Fixes
Fixed the issue with some sitemap pages being constantly updated on auto-sync even if set to “Update on Change”.
Fixed the issue where clicking on numbered citations while chatting with agent on one of settings pages resulted in two browser tabs being opened.
Fixed the issue where the system was sometimes marking Github links as YouTube links.
Spotlight Avatar - Attract user attention and get more conversations:
It is now possible to set up Spotlight Avatar - different from the agent's primary avatar.
Spotlight Avatar welcomes users when there is no conversation in chat.
Spotlight Avatar replaces the bottom-right corner icon for Live Chat deployment.
Spotlight Avatar can be round or rectangular
In-chat avatar is not affected.
Available on Personalize > Advanced page.
Available on Premium plan.
Document Analyst (Beta) - Add full business intelligence to every single document.
Allow end-users to upload any file - pitch deck, cookie policy, contract proposal, resume - and analyze it in the context of the agent's entire knowledge base.
Uploaded files are available only inside the conversation where they were uploaded.
Can be enabled on the new Agentic page for every agent.
When enabled for the agent, the end-user will see the attachment icon in the user prompt field.
Stats about how many times it was used and how many documents were analyzed in the current billing period are available
During beta, the feature is available only on demand - contact our sales team!
CustomCustomGPT.ai for Teams - Secure and granulated large-scale access to advanced AI agents:
Chat-only user role:
Team member role that can chat with agents, but can't modify agents or create new ones.
Users have access to their personal chat history.
Private agent deployment
Deployment code snippets can be made available even for private agents.
Team members can log in to interact with agents.
The feature is available on our Enterprise plans.
🛠 Improvements
On the Ask page, now it is possible to filter for personal conversations only.
New login screens for Share Link and deployed agents.
Improved signup experience for users who are creating the account via invitation.
Improved AI flows related to responding to meta knowledge about agent’s database.
Updated OpenAI moderation model to the latest version.
🐞 Bug Fixes
Added missing labels for the Sent From section on the Analyze page.
Fixed the issue with Starter Questions-related labels not being updated live on the Agent Preview panel.
Improved knowledge base awareness - Agents can now respond to meta questions about their knowledge base, such as providing users with information about what topics they are knowledgeable about.
MCP support for Claude Web - MCP Server page now contains URL tailored for Claude Web, which can seamlessly be added as a tool in their web app.
🛠 Improvements
Users creating the account via the Invite link now have a special version of the Register page, and are not required to verify their email address.
Handling for uploaded spreadsheets with multiple sheets has been improved, to enable agents to be aware that they are providing responses from a sheet of a larger file - previously they were treating each sheet as an individual file.
Users can now fully customize copy for the following Starter Questions strings:
Try asking these questions…
View more
View less
Updated styling and layout of Billing page for subscribed users.
When creating an agent via YouTube, it is now possible to simply input @username.
When configuring Whitelisted Domain list, users are now warned that Share Link deployment is not compatible with this setting.
Updated the Windsurf MCP configuration snippet.
Copy on Analyze > Reports page has been changed for clarity.
🐞 Bug Fixes
Fixed the issue with missing content of citations column on Ask page exports.
Fixed the issue that was causing the “No conversations found” alert to trigger on agent’s settings pages.
Fixed the issue with markdown list items from being incorrectly rendered as code blocks.
Fixed the issue with text color on the Copy Code button.
Fixed the issue with Confluence documents not having the full URL stored in metadata.
Fixed the issue where the Ask page sidebar would open itself in a weird way after the agent's first message.
Fixed minor styling issues on the Customer Intelligence page.
Conversation retention security feature - Users can now set the conversation retention time for their agents, and choose between 12 months (Standard users) and Unlimited or custom time window (Premium & above).
Private Agent Deployment mode - Enterprise customers who have access to Chat-only user roles can now also deploy private agents on their platforms, only accessible to users who are logged in into the CustomGPT.ai system.
Support for processing websites with slow JavaScript - When creating a new agent, or adding a new agent source, it is now possible to select ‘slow mode’ which will allow longer timeouts. This enables correct syncing of websites with slow JavaScript, which previously was not possible.
🛠 Improvements
The Customer Intelligence page now displays a card with a visual summary of presented data.
The user experience and the general styling of the Customer Intelligence page have been updated and improved.
Updated the layout and styling of the Ask page for agents with no documents in their knowledge base.
On the Create Agent page, users can now quickly access links to CustomGPT.ai Free Tools that enable them to build agents from Web Search, URL List, RSS Feed or Podcast.
Added a link to Postman Configuration on the Developers page.
If a user who is not logged in attempts to access the agent via the private link, the app will now correctly redirect them to the agent’s page after logging in.
The All Agents page can now display up to 12 agents per page.
🐞 Bug Fixes
Fixed the issue with filters on the Customer Intelligence page being reloaded before the user provides confirmation.
Fixed the issue where agent’s UI strings were not being automatically updated when changing the language.
Fixed the issue with broken styling when sitemap is not found on connectors like WordPress, Wix, Framer, etc.
Fixed the issue with the Customer Intelligence page having errors on agents without prompts.
Fixed the visual glitch in MCP code snippets.
Fixed the issue with the export conversations feature not working for users with chat-only access.
Fixed the issue with PDFs sometimes being detected as images when the image citations feature is enabled.
Fixed the issue with Complex Query Breakdown agent capability sometimes not having the correct I’m sorry messages.
Use images as citations in agent responses - Premium users can now use the new setting that replaces numbered references with images. This will happen only if the source used to build the response was an image, and it will be added right after the first paragraph it was used for.
Chat-only user role - Enterprise customers can now request the new chat-only user role to be enabled on their teams. Users assigned with this role can login to CustomGPT.ai app and see all agents, but they can’t access any settings or configurations, and can only use agents for chat. All users have access to their personal conversations history.
GPT-4.1-mini support - GPT-4.1-mini is now available for all users who have access to model selector dropdown on following agent capabilities: Optimal Choice, Fastest Responses, Highest Relevance. GPT-4.1-mini capabilities are roughly between GPT-4.1 and GPT-4o-mini. Additional benefit on Fastest Responses capability is that the new model supports numbered references as citations.
Analytics reports sent on email - Users can now get email reports with activity stats and actionable insights on their email, once a week or once a month, depending on the setting for a particular agent.
Custom prompt length - Enterprise customers can now request custom prompt length for their end-users.
🛠 Improvements
When no conversation is active in agent UI, avatar is now centered.
All numbers on the All Agents page now use the same formatting.
🐞 Bug Fixes
Fixed the issue with tables in responses breaking agent UI.
Fixed the issue with inconsistent styling if numbered references are missing the title.
Fixed the issue with Starter Questions being wider than the agent's chat area.
Fixed the issue with mathematical formulas not being escaped in CSV exports.
GPT-4.1 support - GPT-4.1 is now available for all users who have access to model selector dropdown on following agent capabilities: Optimal Choice, Highest Relevance, Complex Reasoning. On average, GPT-4.1 is capable of providing a bit faster and more conversational responses than GPT-4o.
GPT-o4 mini support - GPT-o4 mini has replaced the previous GPT-o3 mini model for Complex Reasoning agent capability, in all three modes: Fast, Balanced and Smart. On top of increased reasoning capabilities, this comes with the additional benefit of supporting numbered citations when selected.
Unlimited conversation time window - Premium users can now allow agents to access conversation history older than 24 hours, enabling chats of potentially unlimited duration. Agents are still bound by their history token window.
🛠 Improvements
Updated styling for all in-chat error messages.
Copy on the new Starter Questions card is now automatically adjusted for the agent's selected language.
Agents newer than 24 hours now have Today as the selected range on Analyze and Dashboard pages.
Added two new states for agent processing feedback: Account limit reached and Success state variant with not all pages successfully processed.
Improved UX of agent processing feedback widget and model when on Ask page.
Improved general flow and page redirects for agent processing feedback.
🐞 Bug Fixes
Fixed the issue where the View more button on the Starter Questions card was not visible on white background.
Fixed the issue where Customer Intelligence custom range selector was always offset by one day if the user's timezone is behind UTC.
Fixed the positioning issue of Developers button on mobile screens.
Fixed the issue with the Microsoft Account button missing on one of the login screens.
Fixed the issue where agent title color was sometimes missing.
Fixed the issue with numbered references format on some models.
Fixed the issue with the free sitemap finder tool that was causing it to fail.
Customer Intelligence data export - All data available in the Customer Intelligence feature can now be exported in full detail.
New Starter Questions layout - Starter Questions layout has been updated to increase visibility and user engagement, and to make agents with a large number of questions, or with particularly long questions, more enjoyable to use.
Automatically generated Personas - All new agents created using sitemap or URL will now have their Personas automatically generated.
🛠 Improvements
Introduced a number of performance improvements when exporting a large amount of data, including breaking data into chunks and adding them in a single ZIP file.
Revamped styling of Share and Export Conversation modals.
Improved styling of indexing feedback widget and modal.
It is now possible to access an agent's page by clicking on its name and avatar on the All Agents page.
🔌 API & MCP
A new Developers page has been added to the app, intended for API keys management and to serve as a resource center.
Numbered citations are now supported when making requests via API.
When retrieving messages via API, it is now possible to choose whether to include numbered references (if present).
API keys per account limit has been increased from 5 to 20.
Revamped tool names and descriptions for MCP server, improving general usability and understanding by third party AIs.
MCP access tokens are now automatically generated and assigned to agents.
A number of smaller improvements and tweaks for MCP servers.
🐞 Bug Fixes
Fixed the issue with the agent’s shareable slug not being correctly generated when duplicating the project.
Fixed the issue where agent preview on Personalize page was not being automatically reset when Persona change is made.
Fixed the issue with the agent loader not being visible if the background is white.
URLs for exported files are now shorter and easier on the eye.
Agent indexing feedback will now gradually display found pages when creating the agent using sitemap.
A number of improvements were released for our MCP server including improving the authentication method, improving connection stability and adding configuration for Trae IDE.
A new layout has been applied to all tabs in the app.
🐞 Bug Fixes
Fixed the issue where very large sitemaps were not fully processed when re-syncing.
Fixed the issue with agent prompt field height on mobile devices, when the user prompt is very long.
Fixed the issue with document metadata sometimes not loading properly.
