Agents in Operator Mode can now send a custom notification to a team member or a group. Each person receives it through their own Guru notification channel: Slack, MS Teams, or email.

This is for teams that build automations, such as verification digests and content gap reports, and need those results to reach the people who have to act on them. It removes the need to wire up an email MCP.

Guru now picks the best-fit agent for your question when the search bar is sitting on a stale, last-used agent. It switches only when it is confident, and otherwise stays on the agent you already have selected.

This is for anyone who searches in Guru, and it matters most to admins who hear about people asking the wrong agent and hitting dead ends.

Try it first by opening search with an older agent selected and asking a question that belongs to a different agent.

Knowledge Agent Automations can now run when something happens, not only on a schedule. This is generally available.

It is for anyone who builds or manages Automations. Agents can react to changes as they occur, such as a new call recording arriving from a connected source or a card being verified.

When you set up an Automation, choose “On an event”. You can pick a card event (created, updated, verified, or unverified, in the collections you select) or a source event (new or updated records from a connected source). The agent receives the triggering card or record as context. One thing to watch: one Automation’s output can trigger another, and not everyone can see every Automation on the team, so a chain of triggers is possible. It should be rare, but review related Automations before turning on event triggers.

Admins can now pull a log of every call made to Guru’s MCP server. Each entry shows who called which tool, from which client app, when, whether it succeeded, and what was requested.

This is for Admins and security teams who need to audit how MCP clients are using Guru, without relying on logs from the client side.

Retrieve the log with GET /api/v1/mcp/toollogs. Access is controlled by a new permission, View MCP Tool Call Logs. Admins have it by default, and it can be granted through a custom role so a security reviewer can audit without full Admin access. This is API only for now, with no screen in the app.

Try it first by assigning the new permission to a custom role and testing the endpoint with a read-only key.

The card editor now supports superscript and subscript. Use it for things like footnote markers, chemical formulas, and French ordinals such as 1er and 2e.

This is for anyone who writes or edits cards, and it removes the need to hand-write <sup> tags in markdown blocks or work around the gap by copying text from another tool.

Apply the formatting from the toolbar, with Cmd/Ctrl + . for superscript and Cmd/Ctrl + , for subscript, or paste text that already has it from Word or Google Docs. There is no feature flag, no new permission, and no credit cost. Slack shows the text as plain text for now.

Try it first by selecting a character in any card and using the new toolbar buttons.

This recap video gives a walkthrough of numerous October 2026 updates to Knowledge Agents including the Microsoft Teams/Copilot release, as well as updates to security, and admin options.

Every source can now be renamed, and the new name appears in the Knowledge Agent source picker and anywhere else agents use sources. Admins can label each folder or drive, for example “People Ops” or “General Claims,” instead of seeing many identical “Google Drive” entries.

Knowledge Agents already answer analytical questions with real numbers. They can now also return the matching records, so a question like “How many cards need verification?” can be followed by “Which ones?” and get the actual list in chat.

This is for admins, owners and authors who use Knowledge Agents to check on the health of their knowledge base. The Data Retrieval Tool covers Cards, connected Sources, Answers, Card Comments and a new Card Comment Replies dataset.

SCIM provisioning now populates Guru profiles. Job title, department, sub-department, entity, manager, location, time zone, work phone and custom fields flow in from Okta or any SCIM identity provider. Knowledge Agents can use all of these fields to tailor answers by role and team.

This is for admins who manage Guru users through an identity provider. It is on for every team. Only job title, work location and manager appear on the profile page. The other attributes are stored and available to agents.

Guru now renders Mermaid diagrams in cards and in Chat. Insert a diagram from the slash menu or the block menu and you get a rendered flowchart in a markdown block. When a Knowledge Agent’s response contains Mermaid, it shows up as a diagram inline instead of raw syntax.

If you write or maintain cards, this is for you, especially if you have wanted diagrams without uploading and updating images every time something changes. Anyone asking an agent to explain a process benefits too, and it works for content added through MCP as well.

Diagrams are text-based, so they stay editable and AI can write them for you. Cards that already contain Mermaid code blocks render automatically, with no migration. There is no feature flag, so it is on for everyone. Slack, email, and exports still show the code instead of the rendered diagram for now.

Open a card, insert a diagram from the slash menu, and edit the text to see it update. Then ask a Knowledge Agent to draw a process you know well and check how it renders in Chat.