Release Notes

Welcome to the Glean Release Notes. Here, you'll find information on recently launched features and improvements.

Coming Soon

Learn about upcoming features and improvements in Glean

Product and feature retirements

Information about deprecated and sunset features in Glean

Glean Betas

Learn about Glean Betas in the Gleaniverse's Admin Hub (Glean Admins Only)

Oct 29

2026

Release version: 566

This release lets you request higher usage limits when you're blocked, brings spreadsheet editing to Glean Canvas, adds a chat context panel in beta, moves Assistant to GPT-6 Luna by default on Glean Key deployments, and adds GPT-6.1 Terra to the model picker. It also lets admins enable Live Mode for Microsoft Teams messages, strengthens security and permissions, expands connector setup flows, and resolves customer-reported issues.

Major launches

Request higher usage limits

When a usage limit blocks you or your agents, you can now request more usage without leaving your work. Admins review each request and approve or deny it in the Admin console, so teams keep moving while usage stays governed.

How to access: Use the request option that appears when a limit blocks you. Admins review requests in the Admin console.

Learn more

Edit spreadsheets in Glean Canvas

You can now edit spreadsheet cells, formulas, and formatting directly in Glean Canvas. Changes save automatically and are included when you export the spreadsheet, so you can refine generated spreadsheets without leaving Glean.

How to access: Open a spreadsheet in Glean Canvas and edit it directly. No setup is needed.

Learn more

Live Mode for Microsoft Teams messages

Admins can now enable Live Mode for Microsoft Teams messages, which retrieves Teams messages live using each person's own Microsoft permissions instead of waiting for indexing.

How to access: Admins can enable Live Mode for Microsoft Teams messages from the connector setup and settings pages in the Admin console.

Fresher Microsoft meeting events and transcripts

Glean Assistant and agents are gaining live retrieval of Outlook Calendar events and their meeting transcripts, so recent meetings are usable before indexing catches up. This capability is rolling out.

How to access: No action needed. The capability applies automatically as it rolls out to Microsoft 365 deployments.

Unsubscribe from Glean emails without signing in

Recipients can unsubscribe from eligible Glean email categories without signing in, including after losing access to their Glean account. Unsubscribe links require confirmation and apply only to the linked notification category. Required account and security emails are unchanged.

How to access: No action needed. This change is rolling out starting October 29, 2026. Use the unsubscribe link in any eligible Glean email.

Chat context panel in beta

Assistant now has a side panel that shows everything used in the current chat in one place, including the agents that ran, the outputs created, and the sources cited. Use it to quickly review or revisit sources and outputs from the chat session.

How to access: This feature is in beta. Admins can enable it from the Feature rollouts page in the Admin console.

AI disclosure in Assistant is generally available

Assistant can show an AI disclosure label below the chat box. It tells people that they're chatting with AI, which can make mistakes, and that they should double-check responses. The label turns on automatically on November 10, 2026 unless an admin opts out.

How to access: To opt out or turn it on early, go to Assistant > Advanced > Disclaimers in the Admin console.

Learn more

AI & model updates

GPT-6 Luna is the default on Glean Key deployments

Assistant's thinking mode, including Auto, now uses GPT-6 Luna by default on Glean Key deployments. Follow-up suggestions and long-chat summarization also use GPT-6 Luna, and fall back to GPT-5.6 Luna when GPT-6 Luna is disabled.

How to access: No action needed. The change applies automatically on Glean Key deployments.

GPT-6.1 Terra is now available

OpenAI's GPT-6.1 Terra is now available in Glean, replacing GPT-5.6 Terra in the model picker. In our evals, it was preferred over GPT-5.6 Terra 54.5% of the time. The biggest win was in correctness: more precise grounding and fewer overconfident claims.

How to access: Select GPT-6.1 Terra in the Assistant model picker, or choose it as the model for an agent in Agent Builder.

Critical updates

Slack-triggered deliveries check channel visibility

Slack-triggered webhook and agent channel deliveries now check that the subscriber can see the Slack channel before delivering a message.

How to access: No action needed. The check applies automatically.

Artifact sharing can grant organization-wide access

Glean's artifact-sharing tool can now grant organization-wide view access and explicitly make artifacts discoverable in Search and the Library. Organization-wide access remains link-only unless discoverability is requested.

How to access: No action needed. Use the sharing options when sharing an artifact.

Learn more

Reserved internal headers are rejected

Glean now rejects requests to its backend that include reserved internal-service headers unless they originate from the deployment's own egress IP addresses.

How to access: No action needed. The protection applies automatically.

Node Agent runs with least privilege on GCP

GCP deployments with Node Agent enabled now run its pods under the dedicated gke-node-agent service account and Cloud SQL IAM user instead of the query-endpoint identity, completing the least-privilege separation introduced with that service account.

How to access: No action needed. The change applies automatically to GCP deployments with Node Agent enabled.

Workflow-run events indicate uploaded files

Workflow-run customer events now indicate whether the initiating user message included uploaded files. File names and contents aren't logged.

How to access: Available in the customer event logs.

Deprecations & sunsets

Autodesk Fusion MCP server removed from the tools catalog

The Autodesk Fusion MCP server is no longer offered in the tools catalog, because Autodesk doesn't currently support third-party clients connecting to it with their own OAuth apps.

How to access: No action needed.

Enhancements

Clearer tool errors with troubleshooting links

Tool errors in Glean Assistant and agents now explain what went wrong and link to a troubleshooting page for the specific error code.

How to access: No action needed. The clearer errors appear automatically.

Learn more

Agent Library category filter loads independently

The category filter on the Agent Library page now loads independently of the agents on the page, so you can filter sooner.

How to access: No action needed.

AI Gateway guardrails appear with their agent policy

On deployments with AI Gateway enabled, AI Gateway guardrails now appear inside their matching agent guardrail policy instead of in a separate AI Gateway guardrails table.

How to access: No action needed. Applies to deployments with AI Gateway enabled.

Clearer AI security issue details

The AI security issue detail view is easier to scan, with clearer hierarchy and more measured typography for issue headings and summaries.

How to access: Open AI security in the Admin console.

Learn more

Assistant remembers your preference to stop follow-up offers

When you ask Glean Assistant to stop a kind of follow-up offer, it saves that as a memory so later chats respect it too.

How to access: Tell Assistant your preference in chat.

Create button matches the Library page you're on

The primary Create button in the Library now matches the page you're on, so creating a new agent, prompt, or other item takes fewer steps.

How to access: No action needed.

Compact status pills in chat

Chat now displays compact plan and subagent status pills with coordinated composer trays, making long-running work easier to follow.

How to access: No action needed.

One-click microphone access for voice

When your browser has blocked the microphone for Glean, the voice dialog in Chrome now offers a one-click button to allow it again, and voice starts automatically once you do.

How to access: No action needed. The option appears in the voice dialog in Chrome when the microphone is blocked.

Email triggers accept any email address

Email content triggers for Gmail and Outlook now accept any email address in the From, To, CC, BCC, and Recipient filters, including external senders and mailing lists.

How to access: Configure email trigger filters in the agent builder.

Faster suggestions in trigger setup

Filter value suggestions for Slack agent triggers, Slack channel suggestions, and content trigger template facet searches now load faster while typing in the agent builder.

How to access: No action needed.

Delete manually added custom Salesforce objects

Admins can now delete custom Salesforce objects they added manually from the Objects tab of the Salesforce connector setup.

How to access: Open the Salesforce connector's Objects tab in the Admin console.

Unified WordPress setup

Admins can now set up WordPress MCP tools and WordPress indexing together in one guided connector setup flow.

How to access: Admins set up WordPress from the Admin console.

Unified Tableau setup

The Tableau connector now uses unified setup by default, with Connected App authentication and permission-aware indexing through the Tableau permissions API. Tableau tools for Assistant are available through the Tableau MCP integration.

How to access: Admins set up Tableau from the Admin console. For the Tableau MCP integration, reach out to your account team.

Clearer tool enablement status on the Tools page

The admin Tools page now shows whether each tool pack, MCP server, and tool is Enabled for all, Enabled for some, or Off, with a per-surface breakdown on hover, instead of showing "All enabled" or "Enabled" as soon as setup is complete.

How to access: Open the Tools page in the Admin console.

Safer tool deletion with clear impact visibility

Actions and action packs can't be deleted while an agent or Glean MCP server uses them. The deletion modal groups references by Agents, Glean MCP servers, and Chat, and deletion stays blocked until agent and Glean MCP server references are removed. Chat-only use doesn't prevent deletion.

How to access: No action needed. The checks appear when you delete an action or action pack.

Control write-action approvals for MCP dynamic tools

Admins can now control how write actions are handled through MCP dynamic tools, choosing the approval behavior that fits their governance needs.

How to access: Configure the approval mode in the MCP server's settings in the Admin console.

Box setup includes MCP tool instructions

The Box setup Connect step now includes instructions for enabling Box MCP Server tools in the Box Admin Console, so Glean can use Box tools after you connect.

How to access: Admins see the instructions during Box setup.

Unified Gong setup

Gong can now be set up in a single unified flow that connects indexing and the Gong MCP tools together, for customers setting up Gong for the first time.

How to access: Admins setting up Gong for the first time see the unified flow in the Admin console.

Bedrock model ID overrides

Customer-key Bedrock administrators can now set an optional model ID override for each configured Bedrock model in the admin LLM setup flow. This supports tagged AWS application inference profiles, such as for AWS MAP discounts, while leaving the default Bedrock model ID unchanged when the field is empty.

How to access: Admins set overrides in the LLM setup flow in the Admin console.

Slack tools setup lists only instances that can send messages

Slack tools setup now lists only the Slack datasource instances that can send messages. Instances set up without a Slack bot are hidden; use a Slack Real Time Search instance for Slack tools instead.

How to access: Admins see the updated list during Slack tools setup.

Simpler scheduled refresh settings

Scheduled refresh settings for artifacts now offer On for everyone, On for some teammates, and Off. Organizations using the previous managed rollout setting now default to On for everyone; explicit Off and teammate-only choices are preserved.

How to access: Admins manage scheduled refresh settings in the Admin console.

Learn more

Faster crawl scheduling for large deployments

Crawl scheduling overhead is reduced for deployments with many crawl scopes, so crawls start more promptly.

How to access: No action needed.

Fixes

Agents always respect admin write-action confirmation settings

Agents now always respect the admin "Run without user confirmation" setting for write actions. Actions that require confirmation now pause for approval, as the agent builder's warning indicates. No changes to agents or admin settings are needed.

How to access: No action needed.

AI Answer stays visible on the Search page

Fixed an issue where the AI Answer on the Search page disappeared after it finished loading, leaving only the sources count.

How to access: No action needed.

Long responses keep their ending after a dropped connection

Fixed an issue where, if the connection dropped during a long chat or agent response, the conversation showed the response as finished but left out the end of the answer, such as download links, until the page was reloaded.

How to access: No action needed.

Google Drive file properties no longer block indexing

Fixed an issue where Google Drive file properties could cause search indexing failures for other documents.

How to access: No action needed.

Airtable crawls recover from expired pagination cursors

Fixed an issue where Airtable crawls with per-table indexing could retry the same table indefinitely when Airtable's pagination cursor expired.

How to access: No action needed.

Oct 27

2026

Release version: 565

This release brings the Azure Files connector to beta, lets agent creators import and export agents, speeds up complex agent workflows with an improved execution engine, adds an AI-generated response notice in Assistant, and expands the Model Context Protocol (MCP) server library with 57 new templates. It also adds Gemini 4 Argon, strengthens permissions and API privacy, and delivers a set of reliability fixes.

Major launches

Rename chat sessions through the Platform API

Developers can now rename chat sessions through Glean's Platform API, so custom applications and integrations can keep chat titles meaningful without opening the Glean interface.

How to access: Available to authorized API clients through the Platform Chat API.

Azure Files connector now in beta

The Azure Files connector is now available in beta across all cloud deployments. Connect Azure Files to make its content searchable in Glean alongside the rest of your company knowledge, with existing file permissions respected.

How to access: Admins can add Azure Files as a data source from the Admin console. The connector is in beta.

Learn more

Import and export agents

Agent creators can now import and export auto agents and workflow agents — this is no longer limited to admins. Import is available for agents you can edit, and export is available for agents you can view, with edit access required to export never-published auto agents. This makes it easier to move agents between deployments and share them across teams.

How to access: Open an agent and use the import and export options. Import requires edit access; export requires view access.

Track indexing progress for new Google Drive and Box connections

Admins can now follow priority-content indexing progress on the Overview page for new Google Drive and Box connections. After discovery finishes, a progress bar shows a representative-sample estimate of high-priority content crawled and indexed, so you can gauge readiness while the initial sync continues. The percentage is an estimate, not an exact count of all documents.

How to access: Open the connection's Overview page in the Admin console during initial sync. No setup is needed.

Learn more

Agents complete complex workflows faster

Agents that run autonomously now use an improved execution engine by default, completing complex workflows that involve multiple tools faster and more efficiently.

How to access: No action needed. The improvement applies automatically.

AI-generated response notice in Assistant

Assistant now shows a short notice below the chat box that responses are AI-generated and can contain mistakes, with a link to answer quality guidance. This supports AI transparency expectations, including the EU AI Act. The notice turns on automatically on November 10, 2026.

How to access: Admins can turn the notice on early, or opt out, in Assistant advanced setup.

Learn more

Action items from meeting notes

Meeting notes now generate action items that are ready for completion when summaries are generated, so you can move from meeting to follow-through without extra steps.

How to access: No action needed. Action items appear when meeting summaries are generated.

57 new MCP server templates

The MCP server library keeps growing. This release adds 57 new remote server templates: Pitch, FinzData, Checklist for Jira, Airwallex AgentOS, Omega Point, Advisor Copilot, Casepoint Gov, IANS, Payhawk, DDAI, LSEG AI, Epiq AI, Openlayer, 15Five, Allstacks Product Studio, Google Ads, 6sense, Fluxmail, Lukka-AML, Ridian, Lukka Pricing, Lucen Track, Gloom, C8 FX Hedge, AgenticPM Deal Scout, LinearB, Mitzu, Lukka Reference Data, Deltek Dela, anvendor, Kong Konnect, TickerInside, Cashfree, IPRally, Wexlyn, TheirStack, BambooHR, Vendasta, InfiniteWatch, WASK, eResource Scheduler, FinDash, Emma Legal, ZIP Codes, Pipefy, SESAMm, Meridian for Xero, Adsby, Hypersonix ProfitGPT, CommonQuant, LinkedIn Hiring Pro, NetSuite QuickStart, FundFacts API, GovIntel.ai, D&B Digital Audiences, CompShop, and Dataddo Data Access. Most templates authenticate with per-user OAuth, so review each template's requirements before enabling it for your organization.

How to access: Admins enable the templates they want from the MCP server library.

Learn more

AI & model updates

Gemini 4 Argon now available

Gemini 4 Argon is the latest model release from Google, targeted at handling complex tasks. In our internal evals, Argon excels in task completion, being preferred in that dimension 61.5% of the time over GPT-Sol 6.1 and almost keeping up with Claude Opus 5.5. Generally, it uses more tokens and has higher latency than the comparative models, but it's an exciting new player in Glean's Model Hub.

How to access: Admins manage model availability in the Admin console.

Critical updates

Service-credential MCP tools require an explicit Platform audience

Existing service-credential MCP tools now require an explicit Platform audience before users can access them through Platform or Glean MCP gateway surfaces. This keeps service-credential actions scoped to the audiences admins intend.

How to access: Admins can review each tool's audience in its settings.

MCP server list API returns only basic details to non-admins

The Glean MCP server list API now returns only basic details — ID, name, path, status, and timestamps — of enabled servers to users who can't manage MCP servers. Admins and MCP server moderators are unaffected.

How to access: No action needed. The change applies automatically.

Usage APIs no longer expose organization budget details

The personal usage evaluation API and the Evaluate My Usage Limits MCP tool no longer return organization-wide budget amounts, usage, or headroom to any caller. They still report when an organization limit blocks usage, and admins can view organization budgets through the admin usage evaluation API.

How to access: No action needed. Admins can use the admin usage evaluation API for organization budgets.

Group permissions preserved for user-created content

Search permissions now preserve access granted through synced identity provider (IdP) groups for supported user-created content, while retaining private-content, publication, and link-sharing restrictions. Shared chats support the same group permissions when shared-chat indexing is enabled.

How to access: No action needed. The change applies automatically.

Azure Files recognizes cloud identities

Azure Files now recognizes cloud user and group security identifiers (SIDs) when evaluating file access and displaying document owners, and Microsoft 365 identity crawls now fetch cloud user SIDs to support this. Access decisions and ownership stay accurate for cloud-managed identities.

How to access: No action needed. The improvement applies automatically to Azure Files connections.

Prepared for DNS-based GKE control plane endpoints

Starting in December 2026, new Google Kubernetes Engine (GKE) clusters running Kubernetes version 1.38 switch to DNS-based control plane endpoints by default, disabling direct IP-based access. This release grants the container.clusters.connect permission to the relevant Glean service accounts so kubectl, CI/CD pipelines, and API clients stay connected.

How to access: No action needed. The permission is granted automatically for GCP deployments.

Enhancements

Unsubscribe from Glean emails without signing in

Email unsubscribe links now open a confirmation page that doesn't require sign-in. Signed-in users keep their existing email preferences experience.

How to access: No action needed. Use the unsubscribe link in any Glean email.

Agents keep running when you edit their triggers

Editing the triggers of an agent that runs on content updates no longer pauses it. Everyone using the agent keeps it running on the updated triggers without activating it again, and they're notified that the agent changed. The agent still pauses for anyone who can't access the updated triggers, and for everyone if the edit adds or changes a required input.

How to access: No action needed.

Suggestions to get started in voice mode

Voice mode now offers readable suggestions alongside its greeting and rotates through a broader set each session. Start with a question that needs no extra setup, or open More to browse grouped ideas for planning, finding answers, writing, and learning. Suggestions disappear when you start talking or typing.

How to access: Open voice mode. Suggestions appear with the greeting.

Memory settings for deployments with Glean memory

Deployments with Glean's memory experience enabled now get the memory settings experience, including the chat preferences prompt, so people can control what Glean remembers.

How to access: Available automatically for deployments where Glean memory is enabled.

Learn more

Clearer errors for removed API features

When an API request is rejected because it uses a removed field, parameter, enum value, or endpoint, the error message now names each removed item and its removal date, and links to migration documentation. Error codes and HTTP status codes are unchanged.

How to access: No action needed. The clearer errors appear automatically.

Sort service identities

Admins can sort service identities by email, name, description, and creator.

How to access: Use the column headers on the service identities page in the Admin console.

See why a usage limit warning or block happened

Usage limit evaluation responses now include decision causes that say whether a warning or block comes from your own allowance or from a shared budget, such as your organization's, without revealing shared budget amounts.

How to access: No action needed. The detail appears in usage limit evaluation responses.

New independent agents are no longer shared with everyone

New independent agents are no longer automatically shared with everyone as viewers. Owners control access with the same sharing options as any other agent, including restricted access and individual viewers.

How to access: Use the agent's sharing options to grant access.

Assistant respects your preferences to stop follow-up suggestions

Glean Assistant now honors requests to stop follow-up suggestions, whether saved as a memory or stated in a chat. A saved preference such as "no follow-up questions" turns off that kind of suggestion in every chat, and saying it in a chat turns it off for the rest of that chat.

How to access: Tell Assistant your preference in chat, or save it as a memory.

Voice mode greets you sooner

Voice mode now begins its opening greeting sooner after the session connects.

How to access: No action needed.

Improved GitHub pull request indexing

GitHub pull request crawling now uses the discovery crawl framework by default, improving pull request reconciliation and deletion handling.

How to access: No action needed.

Personalized tool lists for MCP servers

MCP servers with dynamic skills and tools now include personalized tool lists by default in supported hosts, making recently used tools available for direct calls. The MCP server setup page explains the personalized tool list and links to documentation.

How to access: No action needed for servers with dynamic skills and tools enabled.

Learn more

Invocation IDs in Glean MCP tool responses

Glean MCP tool responses now include an invocation ID in response text and metadata, making it easier to identify a specific tool call when troubleshooting.

How to access: No action needed.

Unified Zoom setup

Zoom setup now combines MCP tools and connector configuration in one flow. Optional indexing isn't currently supported, so complete setup for the MCP tools to be enabled.

How to access: Admins set up Zoom from the Admin console.

Clearer GitLab token permissions during setup

The GitLab setup instructions now list the exact permissions the connector requires when you authenticate with a GitLab fine-grained access token, and setup validation accepts fine-grained tokens that have the Webhook: Read and Webhook: Create permissions when write privileges are enabled.

How to access: See the GitLab connector setup instructions in the Admin console.

Fresher Azure Files indexing

Azure Files discovers new content every two days by default, improving indexing freshness.

How to access: No action needed.

Guided Zendesk setup by default

New Zendesk connector setups now use the guided unified setup flow by default, combining connection and indexing configuration in one experience.

How to access: Admins see the guided flow when adding a new Zendesk connection.

Zendesk OAuth setup by default

Zendesk setup now defaults to the OAuth migration path when no instance-specific override is set. Existing crawl authentication isn't switched until OAuth reauthorization is completed.

How to access: Admins see the OAuth path during Zendesk setup.

Limit assistant features to specific people or groups

Admin controls for assistant features now support lists that enable a feature for specific people or groups, giving you finer-grained rollout control.

How to access: Admins configure feature audiences in the Admin console.

Connector suggestions in the Admin Dashboard

The Admin Dashboard Suggestions section now has a Connectors tab. It lists apps your employees use — based on Glean browser extension activity — that aren't yet connected to Glean, and links to set each one up.

How to access: Admin Dashboard → Suggestions → Connectors.

Central OAuth app support for GitLab Cloud

GitLab Cloud setup now supports a Glean-managed central OAuth app, reducing setup steps. This capability is in private beta.

How to access: Private beta — reach out to your Glean account team to participate.

Learn more

GitHub skill imports use your connected credentials

GitHub skill imports now prefer a user's connected GitHub credentials by default, reducing reliance on mismatched or stale action-pack tokens.

How to access: No action needed.

Easier left navigation color management

Admins can reset a saved primary brand color so each person's chosen left navigation background applies, and undo an unsubmitted color change in the picker.

How to access: Manage brand colors in the Admin console.

Deleted OAuth clients are kept for 30 days

Deleted OAuth clients are now soft deleted and kept for 30 days to maintain client history for billing.

How to access: No action needed.

Connect confidential MCP clients with signed assertions

MCP clients that publish an OAuth Client ID Metadata Document declaring private_key_jwt authentication — such as Claude's and ChatGPT's confidential connectors — can now authenticate to Glean's OAuth server using signed client assertions, without requiring a static client registration.

How to access: No action needed. Compatible clients authenticate automatically.

Metrics export through Google's Telemetry API

The Glean OpenTelemetry collector service account is granted the Cloud Telemetry Writer and Service Usage Consumer roles, so it can export metrics through Google's Telemetry API.

How to access: No action needed. Applies to GCP deployments.

Voice chat events name the serving model

The VOICE_CHAT event in the Glean customer event log now includes a Model field naming the realtime voice model that served the session.

How to access: Available in the customer event logs.

Fixes

SharePoint subsite documents stay visible

Fixed a rare race condition where documents in SharePoint subsites could be hidden from users who have access.

How to access: No action needed.

Deleted SharePoint site pages are removed reliably

Deleted SharePoint site pages are now processed correctly, so they no longer linger in search results.

How to access: No action needed.

GitHub Enterprise webhooks respect organization settings

Real-time webhook updates for the GitHub Enterprise connector now respect the allowed and blocked organization settings, so content from excluded organizations is no longer indexed.

How to access: No action needed.

Google Sheets Append rows action accepts Markdown tables again

Restored Markdown table input for the Google Sheets Append rows action across chat and agent workflows.

How to access: No action needed.

Updated foundation for Glean-managed service containers

Glean-managed service containers now run on Ubuntu 24.04 while retaining the supported Python runtime, keeping the underlying platform current and secure.

How to access: No action needed.

Oct 22

2026

Release version: 564

This release makes the ClickUp connector generally available, adds Auto mode policies for balancing model cost and new model setup guidance in the Admin console, and introduces Gemini 4 Argon on Vertex AI. It also makes earlier chats and artifacts easier to rediscover from autocomplete, strengthens permissions handling, expands connector and admin controls, and resolves customer-reported issues.

Major Launches

ClickUp connector is now generally available

The ClickUp live-mode connector is out of beta and offered by default in the Admin console. Live search, fetch, and per-user authorization work the same as before, so teams can search their ClickUp content with their existing permissions.

How to access: In the Admin console, open connector setup and select ClickUp. Existing ClickUp configurations continue to work without changes.

Learn more

Balance model intelligence and cost with Auto mode policies

Admins on Glean-key deployments can assign departments to Auto mode policies: Economy, Balanced, or Frontier. Each policy balances model intelligence and cost for its assigned group. Assistant responses show which models were used, and intermediate steps indicate when Assistant switches models. Auto mode adjusts its reasoning effort automatically, so the Fast, Adaptive, and Thinking modes are removed. Users who select a specific model can now choose its reasoning effort, from Low to Max, where supported.

How to access: Auto mode policies are generally available on Glean-key deployments as of October 15, 2026. Managed-rollout deployments can enable them from the Feature rollouts page and reach general availability on October 29, 2026. Open Admin console → Platform → Models.

Learn more

AI & Model Updates

Gemini 4 Argon is available on Vertex AI

Gemini 4 Argon from Google is now available on Vertex AI. Select it in the Complex section of the Assistant model picker for an individual conversation, or choose it as the model for an agent in Agent Builder. It works with both Glean-managed keys and customer-managed Vertex AI keys.

How to access: Open the model picker in Assistant and choose Gemini 4 Argon under Complex, or select it as an agent's model in Agent Builder.

Model setup guidance in the Admin console

The Admin console now recommends models by task category, helping admins configure a mix of cost-efficient and highly capable models for finding information, complex reasoning, coding, and artifact generation. A model coverage indicator highlights gaps when a category has no recommended model enabled, including the share of users affected by department-level access settings. Admins can also see which models are required for features such as image generation, real-time voice, and transcription. This guidance is available on both Glean-key and customer-key deployments.

How to access: Open Admin console → Platform → Models.

Learn more

Critical Updates

Files uploaded in Agent Builder stay private

Files uploaded through the builder assistant now remain private to the person who uploaded them and are no longer attached to the agent being built. This prevents a permission bypass through a public-upload fallback.

How to access: No action is needed.

Stronger Content Security Policy for extension pages

The Content Security Policy for Glean browser extension pages is tightened in line with best practices. Behavior is unchanged.

How to access: No action is needed.

Azure Files recognizes Microsoft Entra cloud identities

Azure Files now recognizes cloud user and group security identifiers when evaluating file access and displaying document owners, and Microsoft 365 identity crawls now fetch cloud user security identifiers to support this. Search access for Azure Files content stays aligned with how your files are permissioned at the source.

How to access: No action is needed.

Enhancements

Richer visual answers in Chat

When interactive content is available, Chat now applies stronger visual-formatting guidance by default, so comparisons and relationships render as clear visuals while simple lookups and source-focused answers stay in text. Existing interactive-content settings and capability restrictions still apply.

How to access: No action is needed.

Learn more

Find earlier chats and artifacts as you type

Autocomplete now makes Glean Assistant chats and artifacts easier to rediscover as you type, so you can return to earlier conversations and work without scrolling through your history. This applies to autocomplete suggestions in both Search and Chat.

How to access: No action is needed. Start typing the title of an earlier chat or artifact in the search box or chat composer.

Export presentations and spreadsheets to SharePoint

Presentation and spreadsheet artifacts can now be exported to SharePoint when a Microsoft 365 upload action is configured.

How to access: Configure a Microsoft 365 upload action, then use the export option on a presentation or spreadsheet artifact.

Agent Builder applies proven design patterns

Agent Builder now uses built-in design patterns when it creates or updates Auto mode agents. When an agent's task fits a known pattern, such as running the same process with several steps for many items in parallel, processing items in batches, or isolating broad research tasks, the builder structures the agent to match. This produces more reliable and efficient agents without extra setup.

How to access: No action is needed.

Share projects with identity provider groups

You can now share projects with identity provider groups and select multiple people or groups before saving their access permissions.

How to access: Open a project's sharing options and add people or groups.

Memory settings are on for customers with Assistant memory

For organizations that have received Glean Assistant memory, the memory settings experience, including the chat preferences modal, is now turned on even where the modal was previously turned off. Everyone in those organizations gets the full memory experience and can manage it from settings.

How to access: No action is needed. Manage memory from the Assistant settings.

Learn more

Use Meeting Lookup from the Glean plugin in coding tools

The Glean plugin for Claude Code, Cursor, and Codex can now call Meeting Lookup directly when Meeting Lookup is enabled on the Default Tools Model Context Protocol server.

How to access: Enable Meeting Lookup on the Default Tools Model Context Protocol server, then use the Glean plugin in your coding tool.

Spend-limit emails for Glean Gateway usage

Users of non-Glean clients through the Glean Gateway now receive a dedicated email when they reach their monthly spend limit, with a link to view usage and request more.

How to access: No action is needed.

Gemini tool specification names your connected sources

The Gemini tool specification copied from the Glean Admin console now names the connectors connected to Glean, so Gemini routes questions about that content to Glean first.

How to access: Copy the updated tool specification from the Admin console and use it in your Gemini configuration.

Set up Miro indexing and actions together

Miro connector setup now configures content indexing and Miro actions together in a single unified flow.

How to access: In the Admin console, open the Miro connector setup.

Control scheduled refresh for interactive artifacts

Admins can now see scheduled refresh for interactive artifacts in Assistant settings and choose who may use it. Scheduled refresh will be enabled by default for all customers no earlier than October 22, 2026. To keep it off or limit who can use it, choose Off or On only for admins. Glean respects your choice, and you can change it anytime.

How to access: Open the Assistant settings in the Admin console to review the scheduled refresh setting.

Learn more

Authenticate third-party agents with bearer tokens

Admins can now authenticate with a bearer token when fetching the agent card of a third-party Agent2Agent protocol agent.

How to access: Choose bearer-token authentication when configuring the third-party agent connection.

Faster Slack membership syncs for large workspaces

Slack connector channel-membership updates now apply only the changes for each user instead of rewriting every mapping, which reduces long-running database operations on large Slack workspaces.

How to access: No action is needed.

Glean plugin stays enabled across deployments

Resolved an issue where Glean plugin requests could fail with "The Glean Plugin is disabled" on some deployments.

How to access: No action is needed.

Fixes

Teams meeting transcripts are found for Outlook Calendar meetings

Glean Assistant once again finds Microsoft Teams meeting transcripts for Outlook Calendar meetings when the Teams connector indexes those transcripts.

Live SharePoint search respects the subsite denylist

Fixed an issue where live SharePoint search could return results from SharePoint subsites on the denylist.

Salesforce permission validation uses complete metadata

Fixed a Salesforce connector authorization issue where cached object metadata could cause permission validation to omit profile and permission-set details.

Meeting summaries no longer show raw XML

Fixed a formatting issue that could display raw XML in generated meeting summaries.

Box permission updates are no longer delayed

Fixed an issue where Box file downloads could concentrate API usage on a single account, delaying permission updates for files owned by that account.

Coming soon

Ask Glean in Microsoft Teams channels

Currently in beta, you can mention Glean in a Microsoft Teams channel to ask questions and get AI-powered answers, making it easier to support common HR, IT, and sales workflows where work already happens. General availability is expected on October 29, 2026.

How to access: No action is needed yet. This capability arrives with general availability.

Oct 15

2026

Release version: 563

This release introduces Skill reflection to keep your Skills current, Auto mode model policies and a redesigned Models page, a cleaner Assistant experience with the collapsed Show work summary, new MCP and connector integrations including Klue and Power BI, and a broad set of security, administration, and reliability improvements. It also deprecates OpenAI Web Search in favor of Brave Web Search.

Major Launches

Keep Skills current with Skill reflection

Skills go stale as tools, processes, and team preferences change, and the corrections people make while using a Skill usually stay in the conversation. Skill reflection identifies feedback you've shared in chat that's worth keeping for future runs and proposes it as an update to your Skill. For example, if you keep asking a Skill to put the decision summary first, reflection can propose adding that instruction.

After about 10 learnings, Glean asks clarifying questions and suggests an update. You can also type /reflect at any time, the same way you invoke a Skill. You review the feedback behind each proposal and approve the change before it applies, so one-off corrections don't become permanent rules. Editors of a shared Skill can improve it the same way, so the Skill gets better as each editor contributes what works.

How to access: Available to anyone who can use Skills. Wait for Glean to suggest an update, or type /reflect in chat.

Learn more

Focus on answers with the collapsed Show work summary

Assistant now collapses its intermediate reasoning and tool calls into a single Show work summary once a response is ready. Instead of scrolling past step-by-step planning and tool activity, you see the finished answer first, with the full trail one click away. Expand the summary anytime to review exactly which sources and tools Assistant used and how it reached its conclusion. This keeps conversations easier to scan while preserving full transparency into how answers are produced. The feature is now available by default for everyone.

How to access: No action is needed. Select Show work on a completed Assistant response to expand or collapse the details.

Connect Informatica catalog tools to Glean with a new MCP template

Glean's MCP template library now includes Informatica Catalog Discovery, so admins can connect governed catalog search and metadata tools to Glean Assistant and agents. Users can then look up governed data assets and metadata from Informatica without leaving Glean, using their own Informatica permissions through OAuth. The template is published in Glean's unverified template tier, so review it carefully before enabling it broadly: an Informatica tenant admin must register Glean's OAuth callback URL and provide client credentials to complete the connection.

How to access: In the admin console, open the MCP server catalog, select Informatica Catalog Discovery, and complete the OAuth setup with your Informatica tenant details.

AI & Model Updates

See recommended models and balance cost with Auto mode policies

The Models page in the admin console now shows the recommended models for each task and how well your enabled models cover them, making it easier to confirm your setup matches Glean's guidance. On Glean-key deployments, admins can also assign departments to Auto mode policies — Economy, Balanced, or Frontier — that automatically balance model intelligence and cost, and can turn on automatic downgrades as users approach their monthly usage limit. Auto mode policies are available through Feature rollouts starting October 15, 2026 and become generally available on October 29, 2026.

How to access: Open Admin console → Platform → Models to review recommended models and coverage. On Glean-key deployments, admins can enable Auto mode policies from the Feature rollouts page during the rollout period.

Critical Updates

Tableau permissions now follow your configured authentication method

Tableau permission discovery now consistently follows the authentication method configured for the connector. Connections that use a personal access token preserve per-user access, while Connected App connections use the Permissions API. This removes unsupported combinations and keeps search access aligned with how your Tableau connection authenticates.

How to access: No action is required. The behavior applies automatically based on your existing Tableau connector configuration.

Gong folder-based call access is preserved during unrelated crawls

Fixed an issue where an unrelated Gong crawl could remove valid folder-based access to calls. Users who have access to calls through folder permissions now keep that access consistently.

How to access: No action is required.

Azure Files search access stays aligned with New Technology File System permissions

For organizations in the Azure Files connector beta, Glean now uses document-specific New Technology File System (NTFS) permission data to keep search access aligned with source permissions, stores document metadata and permissions consistently during indexing, and refreshes both whenever files change at the source. Together, these changes keep Azure Files results permission-correct as files and access change.

How to access: No action is required for existing Azure Files beta configurations.

Block specific users from ActAs impersonation

ActAs tokens let approved integrations act on behalf of a user. Admins can now block specific users — such as executives or other high-sensitivity identities — from being impersonated this way, covering both Global Token ActAs and user-scoped ActAs tokens. The control is opt-in: list the protected users' emails or user IDs in security.actAsProtectedUsers and enable the ActAs protected-identity redlist.

How to access: An admin must list protected users in security.actAsProtectedUsers and enable the ActAs protected-identity redlist. The control is off by default.

Update integrations that call the renamed usage-limit MCP tools

The usage-limit MCP tools set_usage_limit and evaluate_usage_limit are renamed to set_usage_limits and evaluate_usage_limits. The old names no longer resolve, so refresh your client's tool list and update any hard-coded calls to the previous names. Tool arguments and behavior are unchanged, so no other integration changes are needed.

How to access: Refresh the MCP tool list in affected clients and update references to the old tool names.

Trace actions to the agent run that triggered them

Action events in customer event exports now include a workflow run ID that matches the run ID on workflow-run events. This makes it easier to see which Assistant or agent run triggered each action when auditing activity. The ID is available in both raw and scrubbed event exports and is empty for actions executed outside a workflow.

How to access: No action is required. The field appears automatically in exported action events.

Deprecations & Sunsets

OpenAI Web Search is deprecated in favor of Brave Web Search

OpenAI Web Search is no longer available for new customer setups, and web search now defaults to Brave Web Search. Starting October 19, 2026, we will begin migrating all existing OpenAI Web Search instances to Brave Web Search. Existing instances continue to work until they are migrated and show a deprecated badge pointing to the replacement. Glean handles the migration, so there is nothing you need to change ahead of time.

How to access: No action is needed. Use Brave Web Search for new web-search configurations; Glean migrates existing OpenAI Web Search instances starting October 19, 2026.

Enhancements

Build topic guardrail rules with less friction

Topic guardrail rule editing now lets admins add multiple activity rows before entering text, making it easier to lay out a rule's structure first and fill in the details afterward. Required-field validation still runs when you save, so incomplete rules can't be published by mistake.

How to access: Edit a topic guardrail rule in Glean Protect and add activity rows as needed.

Add conversation starters to agents built outside the UI

Agents created outside the UI — through headless Agent Builder — now support chat conversation starters through trigger.prompts in the agent's spec.yaml. Teams that manage agents as code can offer the same starter prompts as agents built in the UI, so users get a consistent way to begin a conversation no matter how the agent was created.

How to access: Add trigger.prompts entries to the agent's spec.yaml and publish the agent.

Scope the Web SDK to a single agent

The Web SDK adds a restrictToAgent option that replaces the deprecated restrictToApplication, letting you limit an embedded Glean experience to a specific agent.

How to access: Set restrictToAgent in your Web SDK configuration and remove uses of restrictToApplication.

Export sensitive findings as a CSV file

Glean Protect admins can now export sensitive findings as a CSV file, with one row per finding, by creating a findings export with export type FINDINGS_CSV in the Governance API. The CSV opens directly in Excel or Google Sheets, making it easier to review, filter, and share findings with stakeholders who work in spreadsheets. Existing FINDINGS, DOCUMENTS, and ISSUES exports are unchanged and remain JSON Lines files.

How to access: Create a findings export with exportType set to FINDINGS_CSV using the Governance API.

Confirm staged agent changes before publishing

For deployments that use staged agent versions, publishing now always asks for confirmation. When change review is enabled, the confirmation shows the staged changes compared with the published version, so reviewers can verify exactly what will go live before it reaches users.

How to access: Publish a staged agent version in Agent Builder. The confirmation appears automatically where staged publishing is enabled.

Clearer condition labels for multi-value trigger fields

Agent content-trigger conditions on multi-value fields now read "has any of" and "has any value not in" instead of "is one of" and "is not one of." The matching behavior is unchanged — the new labels make it clear how a rule matches when a document has several values.

How to access: Edit a content trigger's conditions in Agent Builder.

Learn more

Share agents that are working well for you

The agent landing page now highlights sharing after you have used an agent successfully several times, making it easier to pass useful agents along to teammates at the moment they have proven their value.

How to access: No action is needed. The sharing highlight appears automatically on the agent landing page.

Approve access requests directly from Slack

Access request notifications in Slack now include Approve and Deny buttons, so owners can respond to requests for protected chats and artifacts without leaving Slack. This shortens the wait for requesters while keeping the owner in control of access.

How to access: No action is needed. The buttons appear in Slack access-request notifications when Slack notifications are set up.

Meeting briefing cards handle long custom sources

Long custom source names added to meeting briefing cards now stay within the card instead of spilling outside it.

How to access: No action is needed.

Give feedback on agent responses in Slack

Agent responses delivered in Slack now include feedback buttons, so you can rate responses and help improve the agents your team relies on.

How to access: No action is needed. Select the feedback buttons on an agent's Slack response.

More reliable Google Sheets and Gmail tools

Google Sheets tools now accept structured cell values alongside Markdown tables and can create blank Sheets without a content field. Gmail draft rate limits provide clearer retry guidance, and Sheet updates can recover from a transient timeout without creating a duplicate spreadsheet.

How to access: No action is needed. The improvements apply automatically to Google Sheets and Gmail tools.

Search Klue competitive intelligence from Glean

Klue MCP is now generally available. Admins can connect the Klue MCP server so users can search Klue competitive intelligence from Glean Assistant and agents, with results governed by each user's own Klue permissions. Teams that track competitors in Klue can bring that intelligence into the flow of their work in Glean.

How to access: In the admin console, open Connectors → Klue and set up the Klue MCP server. Users then connect with their own Klue credentials.

Learn more

Discover and query Power BI data in Glean

The Power BI connector is generally available. Admins can connect Power BI through Microsoft 365 so users can discover governed reports, dashboards, and semantic models in Glean and query live data in Assistant through Microsoft Fabric IQ MCP, all with their own Power BI permissions.

How to access: Connect Power BI from the admin console connectors page through your Microsoft 365 setup, then enable the Fabric IQ MCP integration for live queries.

Learn more

Add icons to custom People profile links

Admins can upload an icon for a custom People profile link configured on a connector, and People profiles show that icon next to the link. This makes custom links easier to recognize at a glance.

How to access: Configure the custom profile link on the connector in the admin console and upload an icon.

Azure Files crawls use Glean's standard frameworks by default

For organizations in the Azure Files connector beta, each configured connector instance now uses discovery, generic, and chunked crawls by default, improving crawl consistency and reliability.

How to access: No action is required for Azure Files beta configurations.

Cleaner Box setup experience

Box's unified setup now uses the standard display-name field, clearer unnumbered instructions, and help links that stay within their tooltip.

How to access: Open the Box connector setup in the admin console.

Smarter tool suggestions based on app usage

The admin console now ranks tool-setup suggestions by how many employees actually use each app, so the most valuable integrations surface first.

How to access: No action is needed. Review the updated suggestions in the admin console.

Use a separate Microsoft app for SharePoint and OneDrive

SharePoint and OneDrive setup now supports configuring a separate Microsoft app for each source, with permissions tailored to its connection mode. This lets you scope each app to only the permissions its source needs.

How to access: Configure the apps during SharePoint or OneDrive connector setup in the admin console.

Zendesk setup now uses OAuth

New Zendesk connector instances use OAuth authorization during setup instead of requiring an API token. Existing API-token instances keep their current setup flow, so no migration is forced.

How to access: Create a new Zendesk connector instance in the admin console and complete the OAuth authorization flow.

Learn more

Improved platform availability during transient stalls

When a server that runs Assistant and agents slows down briefly, Glean now routes new requests away from it sooner and gives it more time to recover before restarting it. This keeps responses flowing during momentary disruptions.

How to access: No action is required.

Fixes

Outlook search results stay scoped to the signed-in user

Fixed an issue where loading more Outlook search results did not verify that the next page belonged to the signed-in user. Results now always stay within the requesting user's own mailbox.

Faster chat loads for admins in large deployments

Fixed slow chat page loads for admins in deployments with many identity provider groups. Chat no longer downloads the full group list on every load.

Canceled Assistant responses now stop cleanly

When an Assistant response is canceled, loses its connection, or is interrupted by a server restart, the tool calls and background tasks it started now stop as well, instead of continuing to run, re-running in the next response, or appearing stuck. Interrupted runs end immediately with a final status.