Administration

This section provides an overview of administering Custom AI Companion within your company account, including roles, licenses, provisioning, and general feature use.

Role Management

Role Management: Admins must have role-based permissions to access AI Studio

To access AI Studio within a Zoom account’s web portal, users (admins) must have a role with the sufficient AI Studio permissions. By default, only the account Owner will have access to this feature, and the default Admin role must be assigned these permissions.

To update a role with the applicable permissions, perform the following steps:

1. As a Zoom account owner or admin authorized to edit roles, navigate to the Role Management page on the web portal.

2. Click the pencil icon to the right of the role you are editing.

3. Navigate to the AI Studio subsection and define the role access as desired.

4. Repeat for any additional roles.

Example of AI Studio roles

Accessing AI Studio

Admins with sufficient role-based permissions can access AI Studio through the Zoom web portal by clicking the AI Studio menu on the left-hand side of the screen, under the Admin subsection.

Accessing AI Studio in the Web Portal

Licensing

Licensing: Granting Users Access to Custom AI Companion Features

To grant end users access to Custom AI Companion, users must be assigned a Custom AI Companion add-on license.

Account admins can also assign users Custom AI Companion licenses either through manual provisioning or through SSO, via SAML response mapping or SCIM.

Example of assigning a Custom AI Companion License

Refer to Zoom’s support center for more information on manually assigning licenses, provisioning through advanced SAML response mapping, or including the SCIM 2 parameter for Custom AI Companion within their identity provider’s payload.

Knowledge Bases: Things to Know

To configure a Knowledge Base, as an admin, extend the AI Studio menu within the Zoom web portal and click Knowledge.

Knowledge Bases are used to contain relevant Data Sources for each relevant topic

Within the Knowledge tab, admins can create Knowledge Bases, which operate as top-level folders for a given topic, subject, or group.

For example, if creating a Knowledge Base for Human Resources, this Knowledge Base would be filled with Data Sources (i.e., documents, resources) most commonly used by the Human Resources department.

Interface for choosing a Knowledge Base type

Accounts are limited to a maximum of 100 Knowledge Bases

Each account supports the creation of up to 100 distinct Knowledge Bases, providing administrators with extensive flexibility to organize and categorize content sources according to specific business requirements and access governance needs.

Knowledge Bases are customizable with unique icons, names, and descriptions

When creating a Knowledge Base, admins can customize the name, description, and icon for the Knowledge Base from an emoji library.

Example of creating a Knowledge Base

Access to a Knowledge Base can be restricted by Zoom User Groups unless permissions from an integrated cloud-storage provider are being used

When setting up access to a Knowledge Base, administrators can use Zoom User Groups to define and restrict which users within their account are authorized to view or utilize specific content. For instance, a Knowledge Base might be restricted exclusively to a specific department, like Human Resources or Engineering.

However, if the Knowledge Base is connected through a third-party cloud-storage provider—such as Google Drive or OneDrive—and configured to use the provider’s native permissions, access will instead follow those permissions. In such cases, user access defined within the cloud-storage platform itself takes precedence over Zoom User Groups.

Readers with questions about user groups are encouraged to refer to Zoom’s support center for more information.

Each Knowledge Base is limited to one data source type

Each Knowledge Base can only pull data from one source type at a time. This means that data from a Third-Party Cloud Storage provider cannot be mixed with Web Sync, Direct Data Uploads, Third-Party Index Connections, or any combination between them. Each separate data source requires a separate Knowledge Base. For example, after creating a Knowledge Base from a Web Sync, you can't later add documents from Direct Data Uploads or connect it to a third-party index.

Data Sources are the documents and artifacts that fill a Knowledge Base

After a Knowledge Base is created, admins can fill the Knowledge Base “folder” with the corresponding data sources relevant to the Knowledge Base, such as PDFs, .txt, .docx, and .doc filetypes.

Each Knowledge Base supports up to 2,500 sources, while account-wide source capacity is capped at 20,000 total sources

Each individual Knowledge Base supports up to 2,500 data sources, with a combined account-wide maximum of 20,000 sources across all Knowledge Bases. At full capacity (2,500 sources each), an account can have up to eight Knowledge Bases. However, the total 20,000-source limit can also be distributed across a larger number of Knowledge Bases with fewer sources each, depending on your organization’s specific needs (e.g., 100 knowledge bases with 200 data sources each).

Data Sources are limited to 50MB for each file

As of the date of this document’s publication, each Data Source must be no greater than 50MB in size.

Admins can test new data added to a Knowledge Base

After adding new Data Sources to a Knowledge Base, account admins can test the functionality of it by using the Test feature. Within this testing environment, admins can confirm AI Companion is able to properly reference uploaded Data Sources.

Example of testing a Knowledge Base

Example of testing a Knowledge Base

Knowledge Bases must be formally published after adding new Data Sources to take effect

After adding new Data Sources to a Knowledge Base, account admins must Publish the updated Data Source Knowledge Base to take effect, otherwise, it will remain in a “draft” state. If the changes are not published, AI Companion will be unable to acknowledge and utilize the new Data Sources.

Example of Publishing a Knowledge Base

Knowledge Bases: Direct Data Upload

To upload files to Zoom, a user or admin with sufficient permissions must create or edit a manual Knowledge Base and upload the appropriate files. After testing and publishing the knowledge base, the index information will be available to Custom AI Companion-licensed users via AI Companion queries.

Knowledge Bases: Web Sync

To create or manage a Web Sync Knowledge Base, a user or admin must have sufficient permissions to manage Knowledge Bases.

When creating a new Web Sync Knowledge Base, the user selects the desired indexing method and specifies the relevant URL(s). After, Zoom will begin indexing the designated pages. Once Zoom has indexed all specified pages within the customer-specific RAG instance, the Knowledge Base can be tested and then published for use.

Example of creating a Web Sync Knowledge Base

Web Sync files can be converted to articles

During the Web Sync Knowledge Base creation process, the user may enable settings that will automatically convert content crawled from a webpage into article format to help improve readability, indexing, and accessibility of content.

Knowledge Bases: Third-Party Cloud Storage

The Third-Party Cloud Storage connection allows Zoom AI Companion to access, index, and deliver content from third-party cloud storage platforms (e.g., Google Drive, OneDrive) to authorized users within Zoom. To establish a connection with a third-party cloud storage provider, the Zoom user or admin must have appropriate permissions to manage Knowledge Bases. Additionally, an administrator with sufficient privileges on the third-party platform is required to authorize certain integration steps. This integration involves several key steps, detailed below.

An admin must authorize an OAuth app

When setting up the Cloud Drive integration for the first time, an administrator with appropriate permissions initiates an OAuth authorization flow through the Zoom App Marketplace. This OAuth integration is performed at the admin level, meaning the admin grants permissions on behalf of the entire Zoom account. However, this admin-level authorization grants Zoom access only to files explicitly accessible to the admin within the third-party platform, not all files across the entire platform.

User access to data can be based on Zoom User Groups or the third-party’s native permissions

After connecting to the cloud storage platform, the administrator must choose how access permissions will be managed:

• Zoom User Group Permissions: This method uses permissions based solely on Zoom’s internal User Groups. The admin specifies which Zoom User Groups have access to the selected cloud storage documents, regardless of the third-party’s native permissions.

• Third-Party Native Permissions: This method enables administrators to configure Zoom to respect the third-party platform's native permission structure (e.g., Google Drive or OneDrive user/group access settings). When configured, Zoom indexes permissions from the original source and applies access controls that align with the external platform's authorization model, helping maintain consistent document access policies across platforms.

Third-party permissions require user mapping

If the administrator selects third-party native permissions, and has never previously mapped the users from their third-party platform to Zoom, Zoom will prompt them to complete a user mapping process. This matches Zoom user accounts with corresponding accounts in the third-party cloud platform, typically using unique identifiers such as email addresses. In turn, this step establishes synchronization between Zoom and the third-party system, enabling Zoom to check user access permissions based on the user's existing entitlements in the connected platform.

After integrating, Zoom will index the selected data and its permissions

Once the admin selects the appropriate permission method, Zoom indexes two critical components:

• Document Content: Zoom fetches, segments, and converts document content from the cloud drive into a customer-specific RAG module that can deliver contextually relevant search results based on the user’s access level and query.

• Permission Metadata: Zoom indexes detailed permission metadata depending on the chosen method:

  • If using Zoom User Groups, Zoom records internal User Group IDs authorized by the admin for each document.

  • If using Third-Party Native Permissions, Zoom indexes the third-party platform’s native user and group IDs associated with document access, as determined through the user mapping process.

Zoom checks user access levels before delivering matching content

When a user performs a query, Zoom AI Companion initially queries indexed documents based on relevance and preliminary permission matching using indexed metadata. Subsequently, Zoom will check the document permissions against the indexed permissions store:

• If permissions are based on Zoom User Groups, the user’s Zoom group membership is validated directly against the document’s indexed Zoom group permissions.

• If permissions are based on Third-Party Native Permissions, Zoom validates the user’s mapped third-party user/group membership against the document’s indexed third-party permissions.

After, only documents that pass the verification are provided to the user.

Zoom performs incremental and periodic permission syncing

To maintain accuracy and security, Zoom incrementally and periodically synchronizes permission data from the third-party platform into Zoom’s internal permissions store. Due to API rate limitations and latency constraints, Zoom does not perform live permission checks against the third-party platform during user queries. Instead, Zoom checks the most recently synchronized permission dataset.

Heads Up

Zoom incrementally and dynamically updates permissions from third-party providers. In most cases, permission changes should sync within minutes due to incremental updates. However, timing can vary based on your organization's size. For larger organizations or during full synchronization cycles, updates may take anywhere from 2 to 24 hours to complete. Overall, permission changes should not require more than 24 hours to fully synchronize across your Knowledge Base.

Knowledge Bases: Third-Party Index

To integrate a Third-Party Index service with Custom AI Companion, a user or admin with sufficient Zoom App Marketplace permissions, as determined by their User Role, must authorize and complete the connection with the chosen provider.

App Marketplace Integrations

Admins can use the following links to access the Third-Party Index integrations for Zoom with Amazon Q and Glean in the Zoom App Marketplace. To use this feature, your account must have the setting enabled that allows AI Companion to access your organization’s data sources.

After adding an integration, it will appear within the Knowledge list.

Knowledge Base with Glean and Amazon Q connections

Custom Avatars for Zoom Clips

Account admins can enable or disable the use of avatars and custom avatars from the Zoom web portal. There are no further admin-level settings to consider for this feature.

User access to Avatars can be configured on the Account, Group, and User levels

Similar to other Zoom features, Custom AI Companion’s Avatars and Custom Avatar features can be disabled on the Account, Group, and User levels.

Admins can access these features from the Zoom Web Portal by editing the appropriate level’s Settings, selecting the AI Companion tab, and locating the Clips subsection.

Settings for Custom Avatar

Custom Dictionary

To configure a Custom Dictionary as a user or admin with sufficient AI Studio permissions, extend the AI Studio menu within the Zoom web portal and click Custom Dictionary.

Custom Dictionary supports both manual editing and CSV-based import/export

Account admins can manage their account’s custom dictionary through two methods: manual management, or via CSV import/export within AI Studio.

Manual Management

To manually add words to a custom dictionary, click the + Add manually button. Within the pop-up, enter one word on each line, separated by a breakline.

Manually adding words to a Custom Dictionary

To manually delete a word from the dictionary, click the trash icon .

CSV Import/Export

To create or update a custom dictionary in bulk, upload a CSV file with all new words in a single column of the document (e.g., Column A).

Before modifying a custom dictionary via CSV, account admins are encouraged to export and save a copy before making any changes. When uploading the file, account admins will be prompted to either Add to or Replace the dictionary with the file. In the event of a mis-click, the previously-saved copy can be used as a backup.

CSV Import Options

After updating the Custom Dictionary, account admins can test to confirm the new words are recognized using a pre-recorded file or an ad-hoc recording

After adding additional words to the account’s Custom Dictionary, account admins can test the performance of Zoom AI Companion’s recognition of the new words either by uploading a pre-recorded audio file, or making an ad-hoc audio recording using the in-browser tool.

Testing a custom dictionary

Custom dictionary words are bolded when detected in testing

Custom Dictionary must be formally published before new words will be recognized by AI Companion

After adding new words to a Custom Dictionary, account admins must Publish the changes to take effect. Failure to Publish may result in AI Companion failing to recognize new terms. Once published, AI Companion will utilize the Custom Dictionary to improve transcription accuracy in all future meetings for users with the Custom AI Companion add-on.

Publish Custom Dictionary

Accounts are currently limited to one custom dictionary

As of the date of this document’s publication, accounts are currently limited to one custom dictionary at a time. Speak with your Zoom account team for more information.

Custom Meeting Summary Templates

The Custom Meeting Summary Templates section is where account users or admins with access to AI Studio can create custom meeting summary templates.

Admins create custom meeting summaries by entering a brief prompt, which generative AI expands into a tailored summary template based on the topic

When creating a custom meeting summary, admins enter a brief prompt describing the type of meeting—such as “Quarterly Business Review”—and generative AI uses that input to build a tailored meeting summary template. For example, the resulting structure might include sections for financial performance, goal progress, key challenges, and next steps, making it easier to capture the most relevant insights for that meeting type.

Example of generating a new Custom Meeting Summary Template from a prompt

Example of Custom Meeting Summary Template Sections

Admins can customize the list of meeting summary topics, by including topic descriptions that AI Companion will use to match content to the summary

During the custom summary creation process, admins can customize the list of summary topics by defining specific sections and adding descriptions that guide how AI Companion organizes meeting content. These descriptions help AI Companion understand what type of information to look for and include in each section. For example, if a summary topic is labeled “Project Risks” with a description like “Identify any potential risks or blockers mentioned during the meeting,” AI Companion will analyze the transcript and populate that section with relevant content. This flexibility gives organizations greater control over how meeting insights are captured and structured.

Example of defining the purpose of a Custom Meeting Template section

Once the topics are created, admins can test the summary templates in AI Studio

After confirming the list of summary topics, admins can test the customized meeting summary template using simulated conversational data, which is consequently processed through the template to check for quality and consistency.

Admins can also test the custom summary template using a prior Zoom meeting transcript

If desired, admins can test the template using a past Zoom meeting transcript to evaluate how it performs with real meeting content and assess the quality of the custom meeting summary template. This is especially useful for refining templates based on the types of meetings their teams regularly conduct.

After testing the template, it can be published to the entire account or a subset of User Groups

Once the template’s test is complete, it can be published to either the entire account, or a specific subset of Zoom User Groups.

Example of selecting User Groups permissions

Note

Users must be assigned a Custom AI Companion add-on license to use a customized meeting summary template.

Third-Party Meetings with AI Companion

This section explains how to adjust Custom AI Companion’s settings for joining third-party meetings.

Using AI Companion in Third-Party Meetings can be configured on the Account, Group, and User Levels

Similar to other Zoom features, Custom AI Companion’s ability to join Third-Party Meetings can be disabled on the Account, Group, and User levels.

Admins can access these features from the Zoom Web Portal by editing the appropriate level’s Settings, selecting the AI Companion tab, and locating the Custom AI Companion subsection.

Third-Party Apps

To integrate Third-Party Apps with Custom AI Companion, a user or admin with sufficient Zoom App Marketplace permissions, as determined by their User Role, must authorize and complete the connection with the chosen provider.

App Marketplace Integrations

Admins can search the Zoom App Marketplace for supported AI Companion integrations, such as Jira. To use this feature, your account must have the setting enabled that allows AI Companion to perform tasks in third-party apps.

Custom Agents: How to Build

Building a Custom Agent

Account admins or authorized users can create a Custom Agent under the AI Studio sub-section by clicking the Custom Agents option in the Zoom Web Portal.

Note

If your account has purchased Custom AI Companion and the Custom Agents option is not available, an account admin must submit a Zoom support ticket requesting to enable the feature.

Custom Agent Templates

When creating a Custom Agent, admins can select from a pre-defined list of templates or start from a blank configuration. Templates offer a fast way to launch a ready-to-use setup that’s aligned with common roles or workflows, helping organizations quickly scale AI capabilities without starting from scratch.

Pre-built templates include:

• Engineering Task Tracker: Helps developers manage assigned tickets and technical tasks across sprints and projects. It searches and updates issues, tracks progress, and flags bugs to keep engineers focused on priorities and deadlines.

• Interview Support Assistant: Helps interviewers prepare by providing quick access to HR hiring policies, interview tips, and job-specific details. It retrieves job descriptions, reviews candidate profiles, and lists open roles from ERP platforms.

• Incident Management Assistant: Helps developers streamline incident response workflows across teams. It helps organizations efficiently manage, document, and respond to incidents of all severity levels throughout the incident lifecycle.

• Marketing Campaign Assistant: Supports marketers in planning, executing, and optimizing campaigns. It streamlines campaign creation, generates content ideas, and tracks key deadlines and performance metrics.

• Search Support Ticket: Helps agents quickly find previously resolved tickets with similar issues, reducing resolution time and improving consistency by searching ticket history and providing context from past cases.

Templates are also customizable and can be built from a blank template, allowing admins to refine their purpose, update knowledge sources, or extend capabilities with tools and prompts as needed.

Custom Agent templates

Knowledge Bases provide Custom Agents with specialized context

Each Custom Agent can be associated with one or more Knowledge Bases that define its domain expertise. These Knowledge Bases may include internal policy documents, product guides, onboarding materials, or team-specific process outlines.

For example, Support Agent could be configured to access a combination of sources from the Support Center, internal IT articles, and product documentation. By referencing both public-facing support content and internal technical materials—such as setup guides, known issue logs, or configuration best practices—the Custom Agent can deliver precise, up-to-date answers aligned with the most current product information. This combined approach helps ensure that responses are consistent, reliable, and grounded in the same resources used by official support teams.

Knowledge Bases assigned to a Custom Agent

Tools empower Custom Agents to perform actions with third-party systems

Tools define what a Custom Agent is capable of doing beyond conversation — they are the permissions and integrations that allow the agent to act on your behalf.

Some tools are built-in, such as the ability to read, create, or interact with a Zoom Doc. Others can be connected through Zoom App Marketplace integrations using Model Context Protocol (MCP), allowing Custom Agents to perform actions across external platforms. For example:

• Create new issues or update existing ones in Jira

• Retrieve and modify records in Salesforce

• View or update tasks in Asana

• Access structured data from internal systems via API

Tools assigned to a Custom Agent

Admins can also design custom tools through API calls or JavaScript functions, or an MCP server to enable specific functionality unique to their organization.

Important: When creating a custom tool, the tool description becomes especially important. It tells the Custom Agent when and how to use that tool. This description acts as the instruction set the model relies on to decide whether invoking the tool is relevant to a user’s request, helping ensure that every automated action remains precise, intentional, and contextually sound.

Custom Tool creation interface

A Custom Agent’s description defines its behavior and how AI Companion uses it

When creating a Custom Agent, the description becomes especially important—it tells AI Companion what the agent does, when to use its tools, and how it should behave within a given context. For example, including a line like “Post results in the #Engineering Slack channel” helps the agent understand not only what to do, but where and how to do it. These details guide AI Companion’s reasoning and directly shape how it interprets and applies the agent.

A strong, well-written description helps ensure every automated action is accurate, purposeful, and aligned with the intent of the conversation. The clearer and more specific the description, the more effectively AI Companion can understand context, follow direction, and deliver results that match your expectations.

Hero Prompts make it easy to perform common tasks

Admins can define Hero Prompts — pre-populated prompts that act as curated shortcuts for common or high-value tasks users often perform with a Custom Agent. These prompts appear directly in the AI Companion panel, allowing users to initiate workflows with a single click.

For instance, a Hero Prompt could:

• Summarize the shift’s ticket activity: Gather metrics on new, resolved, and open cases from your CRM, along with trends or recurring issues.

• Provide today’s Jira summary: Retrieve all newly created tickets or bug reports from the last day, categorize them by project, component, or priority, and generate a concise summary highlighting trends or recurring issues. Automatically publish the summary to a designated message channel for team visibility.

• Show renewal opportunities closing this month: Query CRM data, calculate renewal timelines, and surface accounts needing attention.

• Provide a weekly onboarding summary: List new hires, completed tasks in Workday, and outstanding onboarding steps.

By predefining these interactions, admins help users bypass redundant setup steps, enabling faster, more consistent outcomes with minimal manual input.

Example of Hero Prompts with a Custom Agent

Note

The ability for a Custom Agent to perform these example functions may be limited by its Tools and Description.