Release Notes

In Agents, users will be able to query BigQuery data using SQL. These queries will be powered by the BigQuery API.

Agent Builder will autosave drafts while you edit, let you update drafts without changing your published agent, and will keep the last 30 published versions so you can restore them as new drafts.

We are adding the ability for users to edit their chat queries and regenerate the response. This will be accessible via a pencil icon shown below the user's chat bubble.

You can now interrupt an in-progress Glean chat response and immediately start a new message. This lets you seamlessly pivot conversations and issue new instructions without waiting for the previous query to finish. The experience is more fluid, giving you greater control and flexibility during multi-step research and troubleshooting in Glean Chat.

The Assistant is getting major upgrades, including a new all-knowledge mode that merges enterprise and real-time data in one query while giving users the option to choose between faster responses or deeper thinking based on their needs.
The system uses an updated agentic engine with adaptive planning that proactively solicits user clarification, accepts user feedback, and refines its responses to deliver higher-quality results.

Glean in Miro allows users to query Glean search results directly into Miro boards and workspaces. It helps in summarizing information, generating insights, and creating action items, thereby enhancing the productivity and efficiency of collaborative sessions.

We've added "Search" and "Chat" tabs directly inside your Support workspace so support representatives can search on Glean or get help from Assistant without switching tools, speeding up case resolution. This is enabled by default; if you'd prefer not to use these tabs, please reach out to your Glean contact to opt-out or turn it off in your Embedded Integrations setting.

Streamline engineering and related workflows by automatically creating Jira issues, speeding up resolution and increasing efficiency.
Note: This complements the existing Jira Service Management (JSM) capabilities to create JSM tickets from within Slack.

Model choice plays a critical role in agent performance. Previously, changing the model required navigating through multiple screens to access the advanced settings for each step — making it cumbersome for builders and admins to update. We've now moved the model selector directly into the agent step configuration settings, making it more visible and easier to use. This change simplifies model customization for each step, giving customers more control with less friction.

Customers can now use GPT-5 to build agents with Glean.

Note: At this time, GPT-5 is only available through OpenAI directly and not via Azure.

Agents is now a top‑level item in Glean’s left navigation menu, making it easier to find and launch.

Glean Assistant now lets users ask deeper questions about Salesforce data using natural language, including sorting and filtering records. Assistant interprets the user's natural language query into Salesforce SOQL to answer the user’s question, using the Search Salesforce with SOQL Action in the Salesforce Action Pack. You can ask Glean a question that includes hints that the answer lives in Salesforce, e.g. “From Salesforce, what are the top open opportunities by ARR?”

Glean will automatically enable the Search Salesforce with SOQL Action for customers as it will improve Assistant response quality. We recommend that admins leave this action on for better Salesforce querying in Assistant.

When using @ to tag a document in the Glean Chat composer, users will now have an option to search and select any document in Glean.

Support admins can now easily add, organize and update which agents are available to customer support representatives.  Addtionally, Glean Search and Assistant are now directly accessible in Zendesk, ServiceNow and Service Cloud.

Exclude search results with a minus sign. You can now remove unwanted results right from search. Add a minus sign before a filter to exclude matches.

• {{-from:me}} hides results from you
• {{-department:engineering}} removes results from Engineering

As part of the rollout of Glean Agents, we have sunset the Prompt Library. Users will no longer be able to access or create simple prompts. Admins will be able to download a backup of all simple prompts from the Admin Console.

Glean Assistant now lets users ask deeper questions about Jira data using natural language, including sorting and filtering records. Assistant interprets the user's natural language query into Jira JQL to answer the user’s question, using the Search Jira Action in the Jira Action Pack. You can ask Glean a question that includes hints that the answer lives in Jira, e.g. “From Jira, what are the open issues assigned to me?” Please note that first time users of the action will have to OAuth to allow Glean to connect to Jira.

Glean will automatically enable the Search Jira Action for customers as it will improve Assistant response quality. We recommend that admins leave this action on for better Jira querying in Assistant, and encourage them to reach out to their account team if they have questions or concerns.

Web search is enabled for all Glean customers unless explicit instructions provided to not enable it. For customers who have not yet set up web search, Brave and OpenAI will be provided as default search providers, both of which offer zero data retention. For customers who have enabled OpenAI or Gemini, Brave will additionally be enabled.

Due to limited usage of the discover tab in the browser extension sidebar and our intentions to simplify the product, we are deprecating this tab's availability. Users will only see the Chat and Search tabs in the sidebar once deprecated.

We’ve added a “Discover Agents” section to the Glean homepage for all users to help them locate and use agents more effectively.

Agents are now returned in Glean Search results for queries that correspond to relevant agents. Additionally, users can filter Agent results by using the operator app:agents in Search.

Snowflake Actions now feature enhanced SQL query validation, support for querying both semantic views and semantic model files, and more flexible permission management by allowing any custom Snowflake role to be specified in setup.

Agent builders can now use the debug mode when they are building and previewing agents with Glean. The debug mode shows which steps were run and the input and output for each step, allowing agent builders to rapidly build, debug and iterate on agent creation

This update includes quality optimizations that improve response completeness, context handling, and enhance instruction-following within the ‘Plan and Execute’ agent step.

All existing answer boards will be migrated to collections using the same name. Your answers—and their permissions—will be fully preserved in the new collection. We will no longer be supporting answer boards moving forward.

Glean Assistant design changes improve the user experience with Chat, intermediate steps, and citations.
Chat interface improvements include a lightweight chat-bubble to make the experience more conversational.
Intermediate steps will now appear in a collapsed UI, without the previously prominent steps with borders. This allows users to see more of the actual chat response and remove unnecessary visual clutter.
Citations user experience has been improved to focus on clarity including:

• Clustering citations when there are more than three cited sources.
• Scrollable citation previews for easier navigation.
• “View sources” option is hidden when no sources are explicitly cited.
• Removed unused sources from citation drawer
• Additional polish and refinements to improve overall citation presentation.

The S3 connector (GCP platform) enables customers to index text-based documents from their S3 buckets into their Glean instance.

The S3 connector for AWS lets customers index text-based documents from their S3 buckets into Glean.

Use the new extension sidebar on unindexed pages, including summarization, Doc Q&A, and any other chat features. The extension will automatically contextualize the page's contents even when the document is not within your company index.

Users can now capture screenshots directly from the extension sidebar to include visual content in their chat sessions. This enables analysis of images, PDFs, presentations, and specific page sections that weren't previously accessible through text-only capture.

Agent creators can now use the Use AI to generate a draft option to generate conversational agents from natural language instructions.

Agent creators can reorder steps in the agent builder by dragging and dropping them. Steps can also be moved between branches in the same manner.

Since we made Glean Agents generally available on May 20, we have provided a transition period for admins to preview and decide when to launch Glean Agents for their deployment. The transition period will end on July 15, and we will automatically launch Glean Agents in all remaining deployments.
When Glean Agents are launched, all existing advanced prompts and apps will be automatically migrated to become Agents. The Prompt Library will then only contain simple prompts and we will offer users the ability to convert their simple prompt into an agent. We will begin showing a warning banner that the Prompt Library will be deprecated by Aug 5.

Agents with chat triggers that use conversation starters containing placeholders (e.g., I would like to file PTO for [[start date]] to [[end date]]) now prompt users to fill in the required placeholders. This update restores the previous experience for conversation starters in apps.

Glean has improved OpenAI web search performance by upgrading the action pack to use the OpenAI Responses API. This change will only impact -Glean Key customers- customers on their own key.

Glean Assistant responses use a more structured format, with dividers, headers, and tables to organize the information.

Ask Glean Assistant questions like “What did I work on last week?” or “Help me prepare for my upcoming performance review,” and you’ll get an AI-generated summary of your key accomplishments—drawing from the Glean Personal Knowledge Graph, which connects your daily activities to higher-level tasks and initiatives.

Users can now set their default experience on the homepage to *Chat* or *Search* or *Sticky* (last used mode) based on how they use Glean. This new setting provides flexibility and lets them choose their preferred starting mode.

The “Send email from Glean” action lets Agents automatically email their output—text, tables, links, etc.—directly to the user who ran or subscribed to the agent (ie: send me a daily email summary of action items).

Glean has added the ability to create Go Links on any page a user is working in, even if the page does not contain an indexed document.

Agent creators can now see metrics on active users, agent runs, and upvote/downvotes for their agents in Insights.

We have sunset the ability to create new answer boards for legacy customers, as we transition to collections, a single tool to store answers as well as other enterprise artifacts. To ensure a seamless transition, legacy customer answer boards will be automatically migrated to collections in mid-July.

Glean is upgrading its underlying architecture from a RAG-based system to an agentic reasoning engine, providing greater access to tools and multi-step workflows across both the Assistant and Agents.

Glean supports Claude Sonnet 4, hosted on both Amazon Bedrock and Google Vertex AI, as part of its model hub. Customers can select Claude Sonnet 4 for each agent and for individual agent steps.

The Glean companion is a floating widget that contextualizes any text the user highlights on their screen, enabling them to instantaneously access company knowledge and insights wherever they work. Users can ask the companion to explain highlighted text, find relevant documents, summarize writing, and more without navigating away from the document or application they are working in.

New action for Glean Agents that allows for the retrieval of all personal activities. This action can be used for self-assessments or to show strategic impact over a period of time.

With Agent API, you can programmatically list, retrieve details, and run agents your team creates in Glean.

Notion arranges information in 'Blocks' and allows users to create 'Toggles' which are sub-blocks within these blocks. Notion users can create many toggles inside a block which Glean was previously unable to retrieve. This feature recursively retrieves information from {{toggleable}}child blocks in Notion and deepens Glean's integration with Notion.

The Insights page will now include a new "Overview" tab replacing the Teammates tab; it features key metrics including usage across search and chat as well as data source usage. A new tab for "departments" will provide a table view of each department's usage, enabling you to understand the departments driving adoption and those at risk. This will only be available to GCP customers as part of this release.

Glean is introducing a new approach to citation generation using LLM-based citations. With this method, the same LLM that generates the response also provides citations for the claims it makes—ensuring that references are directly tied to the information presented. This update is designed to improve citation accuracy, clarity, and trustworthiness by aligning responses more closely with their original sources.

Glean supports the analysis of multiple-file uploads of CSV, XLS/XLSX, and JSON formats in Assistant, 5 files up to 64MB each.

Glean establishes an efficient execution pathway for straightforward information-seeking queries, while also providing an advanced planning mechanism to handle more complex queries effectively. With this architectural change, we expect information-seeking queries to experience lower latency.

Glean supports the analysis of multiple-file uploads of CSV, XLS/XLSX, and JSON formats in Assistant, 5 files up to 64MB each.

Go hands-free with voice-to-text support for Android devices. The new speech-to-text feature lets you dictate questions and messages—no typing is required.