> For the complete documentation index, see [llms.txt](https://library.zoom.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://library.zoom.com/ai-whitepaper/zoommate.md). # ZoomMate ## **Overview** #### ZoomMate is the central intelligence layer for work across Zoom Workplace and connected business systems ZoomMate is designed to serve as the central place where work context comes together across Zoom Workplace and connected third-party services. Rather than operating as a single-purpose assistant tied to one meeting, message, or document, ZoomMate can draw from the broader flow of work across meetings, team chat, continuous meeting chat, emails, documents, whiteboards, summaries, and other business artifacts. When connected services are enabled, that context can extend to external platforms such as cloud storage systems, customer relationship management tools, ticketing platforms, and other authorized enterprise data sources. This centralized position allows ZoomMate to respond with greater relevance, continuity, and awareness of how work is developing across systems.

The ZoomMate work surface

#### ZoomMate helps turn business conversations into coordinated outcomes ZoomMate supports a conversation-to-completion design in which a single thread of work can move across multiple surfaces without losing context. A discussion may begin with a chat message identifying a problem, continue into a meeting where the issue is explored in more detail, and then be carried forward through the meeting summary and related documents created afterward. From there, ZoomMate can help surface the decisions that were made, identify action items and owners, and use that accumulated context to support the next stage of work. That follow-through can extend into execution-oriented outputs. A meeting summary may become the basis for a working document that organizes the team’s plan, open questions, and next steps. That document can then be refined with AI-generated writing or research support, informed by the earlier chat discussion, meeting transcript (if retained), and any connected internal or external materials relevant to the task. As the work becomes more defined, ZoomMate can help translate that same context into structured deliverables, such as task suggestions, workflow steps, or presentation materials. In this model, ZoomMate does not treat conversations as isolated events; instead, it helps carry a developing body of context from the first discussion through the materials, decisions, and outputs that bring the work to completion. #### ZoomMate’s context layer brings together conversations, content, memory, and connected business data ZoomMate’s broader value comes from the range of context it can work with across Zoom Workplace and connected systems. That context can include Zoom-native content such as meeting summaries, meeting transcripts, continuous meeting chat, chat messages, emails, Zoom Canvas docs, whiteboards, slides, tasks, and related workplace content. It can also extend to authorized third-party sources such as cloud storage platforms, customer relationship management systems, ticketing tools, knowledge repositories, and other connected business systems. What makes this a context layer rather than a collection of disconnected sources is that ZoomMate can work across the relationships between them. A customer issue discussed in chat may connect to a meeting transcript, a follow-up document, related emails, files stored in Google Drive or OneDrive, account details from a connected CRM, and work items in a system such as Jira. This broader layer can also include organizational knowledge such as policies, processes, and product information; historical signals such as prior decisions, actions, and outcomes; and user-specific memory that helps ZoomMate reflect persistent preferences and working patterns when relevant. This connected context helps ZoomMate ground responses in relevant data , prior activity, and authorized business information rather than isolated prompts alone. The result is a more relevant experience that improves retrieval across multiple materials, reduces guesswork, and helps users move work forward across conversations, documents, systems, and stages of execution. #### ZoomMate helps realize the value of Zoom Workplace as a unified platform for work Zoom Workplace provides a unified surface for communication, collaboration, and follow-through across the workday. Instead of treating meetings, chat, phone, email, documents, and related activities as separate destinations, Zoom Workplace brings them into a more integrated experience where work can move more naturally from one stage to the next. This unified experience can also extend to connected third-party systems, allowing relevant external business context to be brought into the broader working environment. ZoomMate strengthens that unified environment by helping users work across the flow of activity already taking place within it. Because the surrounding conversations, artifacts, and tasks remain more easily accessible inside Zoom Workplace, ZoomMate can support users with greater continuity as work develops across products and moments. This makes ZoomMate more than a point solution for isolated tasks. It allows AI assistance to feel more integrated into the broader rhythm of work, helping users carry context forward, reduce fragmentation, and move from communication to action with less interruption. #### ZoomMate requires a specific plan and uses an AI credit-based model for certain AI features, including advanced agentic features and third-party integrations ZoomMate operates through a mixed access model in which some functionality is available without credit consumption, while certain advanced capabilities require a ZoomMate subscription and available AI Credits. In general, functions that operate only within Zoom-native context—such as work based on Zoom meetings, chat, documents, and other Zoom-centric data—may be available without using credits. This allows users to benefit from AI-assisted capabilities within the Zoom environment without a ZoomMate subscription. Advanced AI functionality or agentic actions such as AI agents and integrations with third-party applications require a ZoomMate license and consume AI credits. For example, certain workflows that operate within the Zoom platform and do not include advanced AI nodes can be run without consuming credits. By contrast, workflows that integrate with third party applications and include advanced AI search capabilities may require a ZoomMate license and consume AI credits. The practical distinction is therefore based on the scope or type of the task: Zoom-centric functionality may be available without credit usage, while more advanced agentic actions or cross-system tasks that extend beyond Zoom’s native environment are more likely to consume credits. This model helps distinguish between core in-platform assistance and higher-cost actions that depend on broader orchestration, external integrations, or advanced AI processing. Refer to Zoom’s support center for more information on how [Zoom AI credits work, how they’re allocated, and how you can monitor and manage them as a user or admin](https://support.zoom.com/hc/en/article?id=zm_kb\&sysparm_article=KB0085614) and the [pricing page](https://zoom.us/pricing/aic) on storage allotments. #### ZoomMate is available in the Zoom Workplace app and on the web ZoomMate is available both within the Zoom Workplace app and through a web-based experience, giving users flexibility to access AI-assisted search, writing, research, and follow-through across their work. In Zoom Workplace, ZoomMate is integrated directly into the app experience within a dedicated tab, where it can help users engage with the conversations, documents, and tasks that shape the workday. On the web, ZoomMate provides a dedicated workspace for broader interaction with meetings, content, connected data sources, and agentic capabilities. ### Limitations #### ZoomMate is currently only available to customers hosted on Zoom’s United States-based infrastructure ZoomMate is currently available for customers hosted on Zoom’s United States-based infrastructure. Global regions are not currently supported, but are expected to over time. Refer to Zoom’s [Feature Availability Table](/ai-whitepaper/feature-availability.md) for more information on feature availability by region and service model. #### ZoomMate currently only supports the English language ZoomMate currently only supports the English language. This is expected to expand to support additional languages in the future. #### ZoomMate is not available for certain industry verticals ZoomMate is not currently available for certain industry verticals, including government, healthcare, and education. Speak with your Zoom account team for more information. ## **Core Components** ### Operating Modes #### Standard Mode: Generative AI for everyday conversations Standard Mode is the default ZoomMate experience and does not require users to take any action to enable it. It provides a general-purpose conversational AI experience for everyday tasks, allowing users to interact with ZoomMate through natural-language prompts to ask questions, generate or revise content, brainstorm ideas, summarize information, and receive help with a wide range of common work activities. This mode is designed for direct back-and-forth interaction, making it well suited for users who want to engage ZoomMate in a flexible, conversational way similar to other general-purpose AI assistants. #### Advanced Mode: Higher-capabilities for complex tasks Advanced Mode is a higher-capability execution option within ZoomMate for requests that benefit from deeper reasoning, broader retrieval, and more robust task execution than a standard response. Rather than serving as the default mode for every interaction, Advanced Mode is presented to the user as an optional next step when ZoomMate determines that a request may benefit from more extensive processing. For example, after providing an initial response, ZoomMate may present the user with a prompt such as, *“Want a more thorough answer? Try Advanced Mode.”* If the user chooses to continue by selecting the **Advanced Mode** button, ZoomMate can approach the task in a more in-depth and capable way. This may include richer reasoning, more extensive multi-source retrieval, more elaborate synthesis, or more advanced execution across connected tools and workflows.

A conversation prompting Advanced Mode

Because Advanced Mode is intended for higher-complexity operations, its use consumes AI credits and is subject to the user’s available entitlement. In practice, it functions as an opt-in path for users who want ZoomMate to go beyond a lightweight response and perform the task at a more advanced and thorough level. ### Connectors #### Connectors link ZoomMate to external applications and business data sources In ZoomMate, connectors are integrations that link the system to third-party applications and data sources such as Salesforce, Jira, Google Drive, OneDrive, and other external business platforms. These connections allow ZoomMate to work with information that exists outside the Zoom environment, expanding the range of context and actions available to support a user’s request. By connecting ZoomMate to external systems, connectors can allow AI to read relevant business data, retrieve supporting context, and, where supported, take action within those connected platforms. This makes connectors an important part of how ZoomMate extends beyond Zoom-native content and operates across the broader set of tools and systems that organizations use to manage work. ### Web Access #### ZoomMate can use either a local browser extension or a cloud browser to carry out browser-based tasks ZoomMate can use a web browser as an interactive environment for carrying out browser-based tasks as part of a user’s ZoomMate session. This can be done through either a local browser extension or a cloud browser. In both cases, the browser functions as a live execution surface through which ZoomMate can access and interact with web pages, web applications, and other browser-based experiences at the user’s direction. Rather than serving as a stored data source or knowledge repository, the browser provides ZoomMate with a way to perform tasks directly on websites and within active browser workflows. Using the **local browser extension**, ZoomMate can work through the user’s active browser session and use the state already present in that browser environment. This allows ZoomMate to interact with websites and workflows available through the user’s local browser context. Using the **cloud browser**, ZoomMate can perform similar browser-based tasks in a separate sandbox environment hosted for ZoomMate rather than through the user’s local browser itself. This gives users two different browser access models depending on the task and the environment they prefer to use. This browser capability allows ZoomMate to support tasks such as navigating websites, gathering information, completing browser-based workflows, and interacting with live web tools that are only available through an active browser session. When a browser session is closed, data used or collected by ZoomMate for that conversation or session, including screenshots, may be saved to the user’s conversation history. Cloud browser sessions may also be non-persistent, which means a user may need to start a new browser session when reopening a saved conversation. Refer to Zoom’s support center for more information on [using web access with ZoomMate](https://support.zoom.com/hc/en/article?id=zm_kb\&sysparm_article=KB0083738#mcetoc_1jpt2vhla8q). ### Web Content #### Web Content can supplement ZoomMate responses with public web content If the **Web Content** setting is enabled, ZoomMate can search the public web to help answer general knowledge questions or requests that benefit from current external information. This allows ZoomMate to supplement available Zoom and connected business context with information drawn from online sources when relevant to the user’s request. When web content is used, ZoomMate can provide citations to the source material so users can see where the information came from. ### Local File Uploads #### Local file uploads can be used as context for ZoomMate conversations If the **Local file uploads** setting is enabled, users can upload files up to 50MB directly to ZoomMate and use those files as context when submitting prompts. This allows users to ask questions about uploaded materials, reference their contents in follow-up requests, and use those files to support drafting, summarization, analysis, or other AI-assisted tasks. When a file is uploaded by the user, it is stored in Zoom File Storage for up to seven days so it can continue to be used as context within the user’s ZoomMate conversation. This temporary storage period allows users to return to the same conversation and continue working with the uploaded material over time. ### Zoom Products #### ZoomMate can use Zoom-native content across the Zoom platform as response context ZoomMate can use Zoom-native content from across the Zoom platform as data sources when responding to user prompts. Depending on the user’s request and available features, this can include information from resources such as Zoom Meetings, Zoom Canvas, Chat, Zoom Calendar, and other Zoom artifacts and experiences that contribute relevant context. This allows ZoomMate to work across the conversations, documents, schedules, and collaboration artifacts already present within Zoom. By using Zoom-native content as part of its available context, ZoomMate can produce responses that are more grounded in the user’s work and more aware of activity already taking place across the platform. ### Third-Party Data Sources #### Connecting third-party data sources allows ZoomMate to draw from a wider body of organizational information Connecting third-party data sources allows ZoomMate to work with a broader range of business context drawn from systems outside Zoom Workplace. These connected sources can include cloud storage platforms, customer relationship management systems, ticketing tools, knowledge repositories, support platforms, human resources systems, and other authorized enterprise services. By extending beyond a single application or repository, ZoomMate can access a wider body of organizational information relevant to a user’s request. This broader context can improve how ZoomMate retrieves information, understands relationships between systems and records, and supports work that depends on multiple external sources. A request may rely on documents stored in a shared drive, customer account details in a CRM, an active support case, a project ticket, or content from an enterprise knowledge base. By connecting these third-party systems, ZoomMate can operate with a more complete view of external business context, helping produce responses that are more grounded, relevant, and useful. #### ZoomMate can connect to third-party services through API or MCP connections ZoomMate can connect to supported third-party services through conventional API-based integrations or through MCP connections, depending on the service and connection model being used. These connection methods allow external systems to provide additional context and capabilities within the broader ZoomMate experience, extending access to information beyond Zoom Workplace itself. This model allows ZoomMate to interact with a wider range of authorized business systems, including platforms for storage, customer records, ticketing, knowledge management, and other enterprise functions. Whether the connection is established through a standard API integration or through MCP, ZoomMate can use the connected service to retrieve relevant context, support user requests, and contribute to workflows that depend on information held outside the Zoom platform. #### Connecting ZoomMate to user-level data sources like email and calendar retrieves information through direct API access User-level connections—such as email, calendar, or third-party cloud storage providers—allow ZoomMate to retrieve relevant information directly from the connected third-party service when needed to fulfill a user’s request. Rather than relying on a pre-indexed copy of that content, ZoomMate accesses the necessary email or calendar data at the time of the request through the associated API connection. #### Connecting ZoomMate to account-level data sources primarily uses indexed retrieval, with real-time MCP access when needed Connecting ZoomMate to account-level data sources uses administrator-managed integrations to make external, account-scoped content available primarily through indexed retrieval by storing approved content in a customer-specific retrieval index for future queries. This indexed retrieval model allows ZoomMate to work efficiently across large bodies of organizational knowledge from connected systems such as Google Drive, Salesforce, ServiceNow, Confluence, Box, Zendesk, Seismic, Workday, and other supported services. For data that changes more dynamically or benefits from fresher request-time access, ZoomMate can also use MCP or similar real-time connection methods to retrieve information directly from the connected system when needed. This combined model allows ZoomMate to work across both relatively stable organizational content and more current system state within the same broader experience. Indexed retrieval serves as the primary foundation for account-level knowledge access, while real-time MCP access complements that foundation for data that is better retrieved at the time of the request. #### Indexed account-level data is kept current through incremental synchronization Indexed account-level data is not reprocessed in full each time a sync occurs. ZoomMate supports incremental synchronization, updating only content that has been added, modified, or otherwise changed since the previous sync rather than re-ingesting the entire connected dataset. This helps keep indexed knowledge current while reducing unnecessary processing across large enterprise repositories. This incremental sync model supports the practical use of indexed retrieval at scale and allows ZoomMate to maintain a current working knowledge base from connected systems without requiring every file, record, or object to be fetched again during each update cycle. #### Administrators define which external content is eligible for indexed retrieval on the account-level Indexed retrieval from connected account-level connectors is limited to the content administrators choose to make available to ZoomMate. For each supported integration, administrators can define the scope of indexed data by selecting or excluding specific repositories, objects, spaces, folders, drives, or similar content groupings supported by the connected system. These administrative selections establish the boundaries of what ZoomMate can index from that source and help ensure that indexed organizational knowledge reflects the account’s intended data scope. Further, a connected system is not treated as an undifferentiated pool of information. Content that falls outside the administrator-defined scope is not included in indexed retrieval, even if the broader service itself is connected. This allows organizations to control which portions of an external system are incorporated into account-level indexed knowledge. #### Access to indexed and real-time third-party data is determined by source-system permissions Access to connected third-party content depends on permission alignment between Zoom and the external platform. If users from the third-party service have not previously been mapped to Zoom users, administrators must complete a user mapping process that associates Zoom accounts with the corresponding accounts in the connected system, typically using identifiers such as email addresses. This mapping enables Zoom to evaluate access based on the user’s existing entitlements in the connected platform rather than treating connected data as broadly available within Zoom. After a connection is established, Zoom can retain the information needed to enforce permission-aware access during retrieval. For indexed content, this includes both the selected content itself and permission metadata from the third-party platform, including the native user and group identifiers associated with access to that content. This allows ZoomMate to return contextually relevant results from indexed sources while continuing to respect the access model of the connected service. The same third-party permission model also applies when ZoomMate uses real-time MCP access. Even when data is retrieved directly from the connected system at request time rather than from indexed storage, the user’s access remains governed by the permissions defined in that external service, including the current state of those permissions at the time of retrieval where applicable. In this way, both indexed retrieval and real-time access remain subject to the authorization model of the connected platform. ZoomMate is also designed to retrieve only the content needed to fulfill the user’s request within the bounds of the connected system’s authorization model, helping align access with least-privilege principles rather than broad or unrestricted data exposure. For example, an administrator may connect Google Drive and designate a shared drive as an approved data source for ZoomMate. During setup, Zoom can map Zoom users to their Google Drive identities and index both the approved content and the associated permission metadata for that shared drive. If a user has permission to access only the “Sales” folder, ZoomMate can retrieve content from files in “Sales” when needed to fulfill that user’s request, but cannot retrieve content from files in “Finance” unless Google Drive permissions also allow access there. The same principle applies when ZoomMate uses real-time access to retrieve current data from Google Drive: retrieval remains limited to the content the user is authorized to access at the time of the request. Refer to Zoom’s support center for more information on [preparing your organization's content for ZoomMate indexing](https://support.zoom.com/hc/en/article?id=zm_kb\&sysparm_article=KB0084618). ### Skills #### Skills are reusable capabilities that gives ZoomMate a structured way to perform specific tasks In practical terms, a skill allows ZoomMate to approach a request with a more targeted method than general conversation alone, using defined logic, steps, or connected capabilities suited to that type of work. This is useful when a task follows a recognizable pattern—such as preparing a briefing, generating a research report, organizing follow-up actions, or working with a particular business system—because the skill helps ZoomMate handle that work more consistently and effectively each time it arises. A user might choose to use a skill when they want ZoomMate to do more than simply respond with general suggestions. Without a skill, ZoomMate may still be able to provide broad assistance, but it may approach each request as a one-off interaction. With a skill, ZoomMate can apply a repeatable method that is better aligned to the task, helping reduce manual prompting, improve consistency, and produce results that are more structured and useful. This is especially valuable for recurring work, where the same general process needs to be repeated across different topics, customers, projects, or meetings. For example, a user preparing for a customer meeting might ask ZoomMate for a briefing. Without a dedicated skill, ZoomMate could still provide a general response based on the immediate prompt. With a meeting-preparation skill, however, ZoomMate could follow a more specialized process: retrieving related meeting history, pulling relevant account details, identifying open action items, organizing recent documents or chat context, and assembling those materials into a structured briefing. In that case, the user is better off with the skill because ZoomMate is not starting from scratch each time. Instead, it is using a reusable capability designed for that kind of work, which helps produce a more complete, relevant, and dependable result. #### ZoomMate includes default skills and can help identify new skills from recurring patterns of work ZoomMate includes default skills that provide built-in support for common tasks and workflows. These default skills give users ready-made capabilities for types of work ZoomMate is already designed to handle in a structured way. In addition, ZoomMate can help identify when a user’s request reflects a repeatable pattern of work that may be useful beyond a single conversation. When that pattern is clear, ZoomMate can help define a new skill that captures the recurring method behind the task. This allows skills to emerge not only from predefined product design, but also from real usage. A user may begin with a one-time request, but if the interaction reveals a repeatable process, ZoomMate can help express that process as a reusable skill. In this way, ZoomMate supports both out-of-the-box capabilities and the creation of new skills that are shaped by the user’s own workflows, priorities, and recurring tasks. #### Skills can be shared so reusable capabilities benefit other users and teams Skills do not need to remain limited to the person who created them. When a user creates a skill that is useful for a particular kind of work, that skill can be shared with others so the same structured capability can be reused more broadly. This allows helpful methods for recurring tasks to extend beyond an individual user and support a wider team, group, or organization. For example, a user may create a skill for generating a standardized customer briefing, a recurring research report, or a structured follow-up process after meetings. By sharing that skill, other users can benefit from the same reusable capability without needing to recreate the logic themselves. This helps organizations spread useful patterns of work more efficiently and allows ZoomMate skills to function as shared operational knowledge rather than only personal tools. #### Shared skills remain linked to the creator’s current version When a user creates a skill in ZoomMate and shares it with colleagues, that shared skill remains tied to the creator’s maintained version rather than becoming a separate static copy for each recipient. This means that if the creator later updates the skill, those changes can flow through to the shared version used by others. In practice, this allows teams to benefit from ongoing improvements to a shared skill without requiring each user to recreate or manually replace it, helping keep shared capabilities aligned as the skill evolves over time. ### Projects #### Projects create scoped AI assistants tailored to a specific team, function, or workflow Projects allow users to create scoped AI assistants designed for a particular job function, team, or recurring workflow. Rather than starting every interaction with a general-purpose assistant and re-establishing the same context repeatedly, a Project gives ZoomMate a predefined operating context. This allows the assistant to begin from a clearer understanding of the work it is intended to support. A Project is shaped by the configuration the user provides. This includes a name, description, and a set of custom instructions that define the assistant’s purpose, behavior, tone, and boundaries. In this way, a Project functions as a structured container for intent. The Project provides the framework, while the user determines how the assistant should behave within that framework. #### Projects can combine instructions, tools, and knowledge into a single reusable configuration Projects are valuable because they bring together the key elements an assistant needs to operate effectively in a particular context. In addition to instructions, a Project can include access to connected tools, integrated services, and selected knowledge sources. This allows the assistant to do more than respond conversationally by giving it the systems, reference materials, and capabilities needed to support the work it was created to handle. For example, a Project may be configured with third-party integrations such as Salesforce, Jira, Google Drive, Confluence, Slack, HubSpot, or ServiceNow, along with Zoom-native capabilities such as Chat, Meetings, Recordings, Canvas, or Clips. Users may also attach skills that extend what the assistant can do, such as spreadsheet generation, presentation creation, or workflow-specific task support. In this way, a Project can serve as a unified configuration that brings together instructions, tools, and capabilities for a defined kind of work. #### Projects use bound knowledge sources to ground the assistant in relevant context A Project can also be paired with selected knowledge sources that provide the assistant with informational grounding. These may include Zoom Docs, meeting recordings, chat channels, Google Drive files, OneDrive files, Gmail threads, Outlook messages, or other supported materials. Rather than granting blanket access to an entire repository, the Project is grounded in the specific resources the user chooses to bind to it. This selective approach helps keep the assistant focused on the materials most relevant to the intended use case. Once those sources are attached, the assistant can retrieve from and reason over them when responding to requests within the boundaries of what has been explicitly included. This makes Projects useful when users want an assistant to operate with a defined body of knowledge rather than a broad, undefined information space. #### Projects provide a sandboxed environment for more capable multi-step work Projects run in a sandboxed execution environment, known as Sandbox, which means the assistant can do more than generate conversational responses. Depending on its configuration, a Project can support code execution, file handling, document generation, data processing, and other multi-step operations. This makes Projects especially useful for work that depends on coordinated actions across several steps or systems. In this sense, a Project is best understood as a container that defines what the assistant can access, how it should behave, and what kind of work it is equipped to perform. The value of a Project comes from how well that container has been configured to match the needs of the user, team, or workflow it is intended to support. ### Sandbox #### Sandbox lets ZoomMate perform advanced tasks in a separate environment Sandbox is the isolated execution environment ZoomMate can use for tasks that require code execution, file generation, automation, or other advanced forms of processing. Rather than running those tasks on a user’s local machine or directly within Zoom’s primary production environment, Sandbox gives ZoomMate a separate place to carry out that work in a more controlled manner. This allows ZoomMate to support more sophisticated processing while maintaining separation from the user’s device and from core production services. Sandbox is intended to support tasks that go beyond ordinary conversational assistance. For example, it may be used when ZoomMate needs to execute code, generate files, render outputs, or support other tool-driven operations that require a dedicated execution layer. In some cases, Sandbox may also support a browser-based preview or interactive workspace for viewing generated outputs or carrying out related steps within the execution flow. #### Sandbox runs on Zoom’s AWS infrastructure Sandbox runs on Zoom’s AWS infrastructure rather than on the user’s local device. This means tasks that depend on Sandbox are processed in Zoom’s cloud environment using a separate execution layer designed for those advanced operations. #### Sandbox is designed to be short-lived and isolated Sandbox is not intended to function as a permanent user workspace. In general, a sandbox instance exists for a limited period of time and is then automatically removed, helping manage resources while keeping execution temporary by default. This means Sandbox is best understood as a short-lived environment used to complete a task rather than as a persistent computing session; however outputs (such as generated Slides) from the Sandbox are retained as part of the session history. Sandbox is also designed to run in isolation. Sessions are containerized and restricted, with controls intended to limit access, enforce allowed execution paths, and prevent direct interaction with host systems. In this way, Sandbox helps provide a bounded environment for advanced ZoomMate tasks rather than an unrestricted computing layer. ## **Features** ### Actions #### Actions are the execution layer that helps ZoomMate turn requests into completed work In ZoomMate, Actions are the execution-oriented tasks the system can perform to help move work from discussion into outcome. Rather than stopping at a response, recommendation, or summary, an Action allows ZoomMate to carry out the next step of the work itself. This can include creating a document, building a slide deck, sending a follow-up, conducting research, organizing tasks, or performing another result-oriented task on the user’s behalf. This execution layer is what helps distinguish ZoomMate from a system that only answers questions. Actions allow ZoomMate to help complete work, not just describe it, making them a central part of how the platform supports users from conversation to completion. #### Actions can be triggered through conversation, context, or follow-up needs Actions can begin in several different ways. A user may invoke an Action directly through a natural-language prompt, such as asking ZoomMate to prepare a summary, create a plan, or send an update. In other cases, an Action may be suggested based on the surrounding context, such as a meeting that produced action items, a document that needs to be turned into a presentation, or a conversation that clearly requires follow-up. This means Actions are not limited to one type of entry point. They can emerge from direct user intent, the content of an ongoing interaction, or the recognized need for a next step. In each case, ZoomMate uses the request and available context to determine what work should be performed. #### Actions can create outputs, carry out tasks, and update connected systems Actions can take many forms depending on the task at hand. Some Actions create outputs, such as documents, research briefs, slide decks, or summaries. Others carry out tasks, such as organizing follow-up work, preparing a meeting briefing, or assembling materials from multiple sources. Where supported, Actions can also update connected systems by interacting with external tools and services as part of completing a request. This makes Actions broader than simple content generation. They can combine retrieval, reasoning, creation, and system interaction into a single execution path. As a result, ZoomMate can help users not only understand what should happen next, but also produce the deliverables or changes needed to move the work forward. #### Actions may run automatically or involve user review and approval Depending on the task, Actions may be completed automatically or may include a human in the loop through review, confirmation, or approval before execution is finalized. This helps preserve user control, especially when an Action affects communications, deliverables, or connected external systems. In many cases, ZoomMate can prepare the work, present it for approval, and then complete the final step once the user confirms. This balance allows Actions to support meaningful automation without removing oversight. ZoomMate can therefore help reduce manual effort and speed up execution while still allowing users and organizations to maintain appropriate control over how and when actions are carried out. ### Agentic Search #### Agentic search connects Zoom, web, and integrated data sources to surface more relevant context Agentic search is an intelligent search capability that allows ZoomMate to retrieve and connect relevant information across Zoom-native content, uploaded files, public web sources, and authorized third-party systems. Because Zoom operates as a unified environment for conversations, documents, schedules, and other work artifacts, ZoomMate can search across those connected materials in a more integrated way rather than treating each source as a separate destination. This search model can use Zoom artifacts, temporary file uploads, and enabled web content alongside indexed third-party business data to surface a broader and more useful body of context. When necessary, it can also attempt real-time retrieval from connected external systems to obtain fresher information at request time. The result is a more dynamic search experience that helps ZoomMate connect insights across sources and return more relevant, context-aware results. #### A user’s quantity of indexed data varies by their assigned ZoomMate license The amount of data that can be indexed and made available for use within ZoomMate may depend on the user’s applicable license, service tier, or other entitlement. This means the indexed memory or retrieval capacity available to one user or account may differ from that available to another, depending on the ZoomMate offering that has been assigned or purchased. In practice, this can affect how much indexed content is available to support ZoomMate experiences over time. Users with broader entitlements may have access to a larger indexed data allocation, while users with more limited entitlements may have a smaller amount of indexed capacity available. As a result, the amount of indexed data available for retrieval and AI-assisted use is not necessarily uniform across all ZoomMate users or deployments. Refer to Zoom’s support center for more information on [understanding and managing Zoom AI Credits](https://support.zoom.com/hc/en/article?id=zm_kb\&sysparm_article=KB0085614). ### Agentic Workflows #### Workflows: Custom automations that help simplify your work Workflows give users a simple, powerful way to automate recurring tasks using natural-language instructions. Whether starting from an out-of-the-box template or building a workflow from scratch, users can describe the task they want to automate, add any needed inputs or context, and set the schedule. ZoomMate then designs the workflow accordingly, reducing the need to manually repeat the same follow-up steps each day or week. For example, a user can create a workflow that reviews all of their meeting summaries from the week and automatically delivers a consolidated report every Friday at a time of their choosing. The report can highlight key decisions, action items, and responsibilities, and it can also include a “looking forward” section that outlines upcoming priorities. Once set up, the workflow can run on its own—no manual intervention required.

Example of building a Workflow using natural language prompting

This example is just the beginning—a well-designed workflow can go beyond summaries and reporting, including the ability to execute follow-up actions on the user’s behalf, such as creating a Daily Reflection Report or sending updates and structured data to third-party tools like Slack. This gives users a flexible way to streamline routine work, reduce manual effort, and stay consistently aligned with their own priorities. Refer to Zoom’s support center for more information on [using Workflows](https://support.zoom.com/hc/en/article?id=zm_kb\&sysparm_article=KB0083738#mcetoc_1jc9o690i2m). #### Agentic Workflows: Custom automations that can incorporate Zoom and other third-party services Agentic Workflows extend workflow automation beyond Zoom-native actions by allowing automations to interact with connected third-party services as part of completing a task. Like standard workflows, they can be created from natural-language instructions and used to automate recurring work. The difference is that Agentic Workflows can go beyond internal Zoom context and invoke actions or retrieval across supported external systems when the workflow requires it. This allows users to automate processes that span both Zoom and other business platforms. For example, an Agentic Workflow might review a meeting summary in Zoom, identify follow-up work, create a related item in Jira, update a record in Salesforce, or send structured information to another connected service. In this way, Agentic Workflows help turn ZoomMate from an in-platform automation tool into a broader cross-system automation layer that can coordinate work across the tools an organization already uses. Agentic Workflows are especially useful when the desired outcome depends on action outside Zoom itself. While a standard workflow may be sufficient for summarizing meetings, generating reports, or organizing Zoom-native outputs, an Agentic Workflow is better suited for processes that require third-party retrieval, external writes, or interaction with connected services as part of the workflow’s execution. | Aspect | “Standard” Workflows | Agentic Workflows | | ---------------- | ----------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------- | | Primary scope | Automate recurring tasks within Zoom | Automate recurring tasks across Zoom and connected third-party services | | External systems | Do not invoke third-party services (except the Send email node for Microsoft Outlook and Gmail) | Can invoke supported third-party services and functions | | Best suited for | Zoom-native summaries, reports, reminders, and follow-up tasks | Cross-system processes involving external retrieval, updates, or actions | | Data and actions | Operate on Zoom context and outputs | Can combine Zoom context with actions in tools such as CRM, ticketing, or collaboration systems | | Example | Generate a weekly report from meeting summaries | Generate a weekly report and send structured follow-up data to Jira, Salesforce, or Slack | #### Workflows are best suited for repetitive tasks that require consistent, repeatable outcomes Workflows are most effective when the task follows a known pattern and benefits from being performed the same way each time. This makes them well suited for recurring activities such as compiling updates, generating reports, distributing information, or carrying out routine follow-up actions after a meeting, event, or status change. In these cases, the value of a workflow comes from reducing manual repetition while producing dependable results. #### Workflows automate structured processes through controlled and predictable execution A workflow follows a defined sequence of steps to complete a task in a controlled manner. Rather than deciding dynamically what to do next in the way an agent might, a workflow carries out the logic, instructions, and actions that have been established for that process. This makes workflows especially useful in situations where consistency, predictability, reliability, and traceability are important. #### Workflows can be triggered manually, by events, or on a defined schedule Workflows can begin in different ways depending on the task they are designed to support. A user may start a workflow directly, a workflow may run automatically after a defined event such as the end of a meeting or the creation of a new record, or it may run on a schedule such as every morning, every Friday afternoon, or at another recurring time. This flexibility allows workflows to support both user-initiated tasks and background automation that operates with little ongoing effort. #### Agentic Workflows involve third-party systems and consume credits Agentic workflows are workflows that use third-party systems, data sources, or connectors to retrieve information, update records, or take action outside the Zoom environment. For example, a workflow that connects to Jira to update a ticket, retrieve issue details, or trigger an action in another external application would be considered agentic. Because agentic workflows consume AI credits, they are designed to be used by users with an assigned ZoomMate license. Workflows that remain entirely within Zoom—using only Zoom data, Zoom services, or Zoom-native functions—are not considered agentic for this purpose and do not require a ZoomMate license. #### Workflows can operate independently or be invoked by agents as part of larger tasks A workflow can run on its own to automate a specific task, but it can also serve as a building block within a broader ZoomMate experience. In more complex scenarios, an agent may invoke one or more workflows as part of fulfilling a larger request. This allows workflows to provide structured execution for well-defined tasks while supporting more adaptive, multi-step interactions when used alongside agents. ### Scheduled Tasks #### Scheduled tasks are recurring automations that help reduce friction across repeat work Scheduled tasks are recurring automations that run on a defined schedule to help users reduce the manual effort associated with tasks that repeat over time. Rather than requiring a user to remember, initiate, or reconstruct the same process each day or week, a scheduled task allows ZoomMate to perform that work automatically at the appropriate time. This makes scheduled tasks especially useful for recurring activities such as daily briefings, weekly roll-ups, meeting preparation, action item tracking, and other routine summaries or follow-up processes. Scheduled tasks can help ZoomMate identify repeated patterns of work and turn them into ongoing automation opportunities. For example, ZoomMate may analyze a user’s recent meetings, chats, and documents to identify recurring tasks that could be automated, then propose a set of scheduled tasks for the user to review and confirm. Those tasks might include drafting a daily standup update, producing a recurring digest from high-volume channels, checking related documents for discrepancies, generating a weekly meeting summary roll-up, or extracting action items after a set period of meetings. In this way, scheduled tasks help convert repeated work into a persistent automation that runs at regular intervals with less ongoing user effort. #### Scheduled Tasks run recurring work on a set cadence, while Workflows define repeatable processes Because both features automate work and help reduce repeated effort, **Scheduled Tasks** and **Workflows** can be easy to confuse. They often overlap in practice, and from the user’s perspective they may sometimes feel similar. The key difference is that Scheduled Tasks are centered on recurring, time-based execution, while Workflows describe the broader automation process that defines how the work is performed. **Scheduled Tasks** are a more specific form of automation focused on recurring work that happens at regular intervals. They are typically used for tasks such as daily updates, weekly summaries, recurring meeting preparation, or other routine activities that benefit from running automatically on a defined schedule. In this sense, Scheduled Tasks emphasize **when** work runs. **Workflows** are structured automations that define how a task should be carried out. They can include a sequence of steps, logic, inputs, and outputs, and they may be triggered manually, by an event, or on a schedule. Their main purpose is to automate a known process in a consistent and repeatable way. In this sense, Workflows more broadly define **how** the work is performed. ### AI Agents #### AI agents: AI assistants built in Zoom Chat that can help you be more productive AI agents allow users to interact with AI in a more dynamic and adaptive way within Zoom Chat, either through a direct interaction with the agent or by adding an agent to a chat channel. Rather than following a fixed sequence of predefined steps—like one might do with a Workflow—agents can interpret a user’s natural language request, reason about the task, determine what information or tools are needed, and take action across connected or authorized systems and workflows. This makes them especially useful for requests that are broader, less structured, or require coordination across multiple sources of context. For example, a user may create a custom sales agent to prepare them for an upcoming customer meeting. In response, the agent can gather relevant meeting history, review related documents, pull in connected business context from a CRM, identify open action items, and help produce a briefing document or follow-up materials. Instead of requiring the user to manually assemble each piece of information, the agent can coordinate those steps as part of a larger conversational request. Agents can also help bridge the gap between information and execution. A user might ask an agent to summarize recent project activity, identify unresolved issues, draft a status update, and prepare the next set of follow-up actions. In these cases, the agent can perform multiple tasks. It can coordinate context, invoke tools or workflows when needed, and help move work forward through a more unified experience. #### AI agents are best suited for complex tasks that require reasoning, coordination, and adaptation AI agents are most effective when a task cannot be reduced to a single repeated procedure in the way a Workflow often can. Instead, they are well suited for requests that involve multiple steps, changing context, ambiguous inputs, or the need to decide what should happen next. In these situations, the value of an agent comes from its ability to interpret intent, reason across available information, and coordinate an appropriate response through first and third-party resources when applicable. #### AI agents can be paired with Knowledge sources for additional context and more relevant responses AI agents can be paired with selected Knowledge sources to give them more contextual grounding when responding to requests or carrying out tasks. These Knowledge sources can include materials such as Zoom Canvas documents, chat channels, recurring meetings, and other supported Zoom-native resources that provide relevant background, history, or reference information. By pairing an agent with these sources, organizations and users can help ensure that the agent’s responses are better informed by the specific work context in which it is expected to operate. For example, an agent paired with a recurring meeting may be able to draw on the summaries and context generated from that meeting series, helping it answer questions, synthesize patterns over increased context, or perform related tasks with greater continuity. In the same way, pairing an agent with a Zoom Canvas document or a chat channel can help it work from a richer body of knowledge that reflects the team’s ongoing materials, discussions, and priorities. #### AI agents are commonly triggered through conversation, but can also operate on a fixed schedule AI agents are most often initiated through natural-language interaction or contextual user requests, allowing users to ask an agent to investigate an issue, prepare a deliverable, or help complete a task. In these cases, the agent can determine how best to proceed based on the request and the available context. At the same time, agents can also be configured to perform tasks on a fixed schedule when the use case calls for recurring or time-based execution. This makes agents useful both for work that begins with a conversational request and for work that benefits from regular, automated follow-through. #### AI agents can orchestrate workflows, tools, and connected knowledge sources as part of a larger task AI agents can function as coordinators across the broader ZoomMate environment. In more advanced scenarios, an agent may invoke workflows, retrieve information from connected systems, use available tools, and combine those capabilities into a larger multi-step response. This allows workflows to provide reliable execution for defined tasks while agents provide the reasoning and orchestration needed for more flexible end-to-end experiences. ### Custom Avatars #### Bring your scripts to life with dynamic AI avatars Custom Avatars enable users to create a personalized virtual avatar using their recorded video and voice, or select from a list of pre-provided avatars. Once an avatar is selected or generated, users can upload a script that is transformed into a Zoom Clip, with the avatar narrating the content in a natural voice and, if applicable, the user’s likeness.

The custom avatar clip creation interface

This feature is especially useful for businesses looking to scale video production efficiently, allowing users to create training materials, lessons, and presentations with just a transcript—reducing production time while maintaining quality, personalized video content. For example, a compliance officer may need to remind employees of quarterly training requirements. Rather than recording a new message or sending another lengthy email, they use their custom avatar to generate a brief, direct video reminder from a script—quickly reinforcing the message while keeping communication efficient and repeatable. Custom avatars also support multiple languages, enabling users to upload a script in any supported language and generate output in the same language, or produce a copy in an alternative language. For example, if a user records their custom avatar likeness in English, they can upload a script in Spanish or Russian and will receive a clip spoken in the corresponding language that reflects their authentic voice. Alternatively, a user can upload a script in English, and also receive copies of the clip in Spanish or Russian. The current list of supported languages includes: | | | | | ------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------ | Refer to Zoom’s support center for more information on [creating clips with Custom Avatars](https://support.zoom.com/hc/en/article?id=zm_kb\&sysparm_article=KB0080200). ### Custom Dictionary #### Provide Zoom with your business’s or industry’s language and jargon for better captions and transcripts Custom Dictionary enhances Zoom’s ability to recognize and incorporate your organization's specialized terms, industry jargon, and acronyms, resulting in more accurate and contextually relevant meeting captions, transcripts, and summaries. When utilized, Zoom’s automatic speech recognition service will reference the account’s Custom Dictionary during the captioning process, resulting in more accurate meeting captions, transcripts, and summaries. This feature is especially beneficial for industries that rely on specialized language, such as healthcare, finance, or manufacturing sectors, where precision in communication is crucial. A Custom Dictionary is designed to help Zoom AI adopt the unique vocabulary of the organization, improving the relevance and quality of AI-generated responses. For example, imagine a legal team using Zoom AI to generate a meeting summary from a strategy session where one attorney says, “We need to prepare a memorandum regarding the Daubert standard and its applicability in this tort case.” Without a legal custom dictionary in place, it can be difficult to accurately identify uncommon words like “Daubert,” potentially rendering it as “Dahlberg” or “dog bird,” which could lead to confusion or misinterpretation. However, with a domain-specific dictionary that defines terms like “Daubert,” “tort,” and “voir dire,” Zoom AI is better equipped to recognize and transcribe specialized language.

Adding words to a custom dictionary

Refer to Zoom’s support center for more information on [creating and managing Custom Dictionaries.](https://support.zoom.com/hc/en/article?id=zm_kb\&sysparm_article=KB0080313) ### Custom Meeting Summary Templates #### Standardized summaries defined at the account or group level Custom Meeting Summary Templates allow account administrators to create and manage standardized meeting-summary formats for use across their organization. Unlike Personal Meeting Summary Templates, which are created and managed by individual users for their own preferences, Custom Meeting Summary Templates are defined at the account or group level to support consistent summary structure, language, and emphasis across users who have access to them. These templates are useful when an organization wants meeting summaries to follow a specific format that reflects internal standards, business processes, or reporting expectations. For example, a company may want every meeting summary to include sections for business decisions, customer risks, follow-up owners, compliance considerations, or other categories that are not fully reflected in Zoom’s default templates. In that case, an account administrator can configure a custom template so that eligible users can generate post-meeting summaries that align with the organization’s preferred structure.

Example of a new custom template

This model helps organizations promote greater consistency in how meeting outcomes are documented and shared. Rather than relying on each user to create their own preferred format, the account can define a common summary framework that supports more standardized communication, review, and follow-through across teams. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://library.zoom.com/ai-whitepaper/zoommate.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.