SoftTech AI Registry
Google · Google Workspace

Workspace Studio Agentic Flows

User-configured automations ("flows") built in Google Workspace Studio. A flow begins with a documented starter (schedule, incoming Gmail message, Google Chat event, Drive or Sheets change, Calendar meeting offset, Google Forms response, meeting outputs, or manual start) and then runs a sequence of steps that may include Gemini reasoning and persistent writes across Gmail, Chat, Docs, Sheets, Drive, Calendar, Tasks and Gemini Notebook, plus beta third-party integration and webhook steps.

Recorded characteristics

Function
Google documents that every flow has two main parts: a Starter ("the trigger that launches your flow" — "an event, schedule, or create a trigger manually") and Steps ("a task your flow performs after it starts"), with multiple steps permitted. Documented starters are: On a schedule; When I get an email (Gmail, with From/To, words, subject, attachment, size and default-label conditions, ignoring events created by other flows unless explicitly included); When I get a chat message; When I am mentioned; When someone joins a space; When an emoji reaction is added; When a sheet changes; When an item is added to a folder; When a file is edited; When an item in a folder is edited; Based on a meeting (a time offset before or after a Google Calendar event); When a form response comes in; When meeting outputs are ready (transcript or Gemini notes); and Start manually (limited to users in the Gemini Beta program). Custom starters built by developers are a further, separately controlled category. AI reasoning is a step type, not the whole capability. Google documents "Ask Gemini" (custom prompt to generate text, answer or transform information from earlier steps), Gemini Gems steps, "Deep Research (Beta)" (builds a research plan, fetches data and produces a Google Doc), "Ask Gemini Notebook", and logic steps "Check if" (continue only if conditions are met), "Filter a list" and "Repeat for each". Google states Gemini can "analyze, write, research, or even determine if a flow should run at all", and the Gmail labelling step offers AI-powered labels applied only when message content matches a label description. Documented persistent actions, by service. Gmail: Draft an email; Add or remove labels; Mark as read or unread; Star or remove star; Archive an email; Notify me by email; Send an email (compose and send a new email from the user's Gmail account); Send a reply (reply to sender or reply-all, with Cc and Bcc). Chat: Notify me in Chat (direct notification to the flow owner only, sent through a Chat app named after the flow); Send a message (direct message to one or more people); Post in a space; Reply to a message (in thread or quoting). Docs: Create a doc (with content and Drive location); Add to a doc (before or after existing content). Drive: Add email attachments to Drive; Create a folder; Copy a file or folder (folders up to 50 items and 3 levels of nesting); move an item to a selected Drive location. Sheets: Add a row; Update rows matching conditions; Clear rows matching conditions; Get sheet contents (read only). Calendar: Block time (create an event); Set up a meeting (with guests and description); Add guests to a meeting. Tasks: Create a task (added to the primary task list). Gemini Notebook: Ask Gemini Notebook (read and generate) and Add a source to a notebook (write). Third-party integration steps in Beta: Asana (create projects, sections, tasks, subtasks), Confluence (create pages), Jira (create issues and issue comments), Mailchimp (manage subscribers and campaigns), Slack (send messages), plus Hubspot, Quickbooks and Salesforce named by Google, and Webhooks ("Send HTTP requests to external endpoints and trigger actions in them"). Flows are created in three documented ways: from a template, from scratch in the flow editor, or by describing the automation in natural language for Gemini to assemble ("Tell Gemini what you want to automate, and it will build the flow for you"). Gemini constructing a flow definition is not execution of that flow. A flow is not live on creation: Google's tutorial ends with "When you're finished, click Turn on", and warns that the optional test "takes real actions. It will send messages, update files, and set up meetings." Once turned on, the flow runs when its starter fires, without the owner being present, except where an administrator approval policy or a DLP rule pauses a step.
Data access
Access depends entirely on the starter and steps a user configures and on the Google services their account can use: Google states "The templates available to you rely on your Google app access." Workspace Studio is documented as integrating with Gemini, Gmail, Google Calendar, Google Chat, Google Drive, Google Docs, Google Forms, Google Sheets, Google Tasks, and third-party apps. For AI steps Google documents the source setting: "By default, Gemini can reference the web, as well as all content that you can access in Google Workspace and connected integrations. To prevent Gemini from accessing your personal or organization's data, choose Web only." Google warns that "Integration steps can share your Google Account data (like Gmail or Calendar contents). Make sure you trust the service." No document reviewed states that every flow has access to all Workspace data; access is per flow, per step, and bounded by the owner's own access.
Actions
Can take actions
External actions
Conditional
Human confirmation
Conditional
Permission basis
User permissions
Administrative control
Administrators control Workspace Studio as a service in the Admin console (Apps > Google Workspace > Workspace Studio > Service status), on or off for everyone, per organisational unit, or by access group; turning it off for a user means "any active flows they own stop immediately and they can no longer access the app". Google documents a separate admin page for managing access to individual steps and starters, by Workspace service or individually. Google's 17 August 2026 Workspace Updates post names five built-in enterprise security controls: agent identities; agent access management; agent auditing and observability; admin settings / human-in-the-loop (HiTL); and runtime protections. It states that an "Agent access management dashboard gives admins the ability to suspend all flows or targeted OAuth scopes for individual flows, such as revoking Drive access", and that the security investigation tool can move from an audit event to that page. Additional admin settings "support disabling specific step types for flows, disabling Gemini data access, enforcing end-user confirmation for steps that share data externally, and disabling webhook integrations". A Beta identity-attribution setting lets admins decide whether flow actions show the owner's identity or are attributed to the flow itself with owner information visible; Google says this defaults to on, meaning actions are attributed to the flow. Google notes the least-privilege agent identity, the attribution setting and agent context in audit logs apply to newly created flows only, with existing flows supported later. DLP for Studio is a distinct runtime control from Gemini DLP. Google describes Gemini DLP as restricting "Gemini's ability to access Drive data based on content conditions and labels", while DLP for Studio controls flow execution: rules are evaluated per step before it runs and the rule action is block ("Stops the current run of the flow"), require user approval ("Pauses the run and requires the user to review and approve the step on the Workspace Studio Approvals page"), or audit only. It can scan referenced sources, step inputs and output visibility, with audience conditions "Block sharing with external users and agents" or "Block sharing with all users and agents". Documented limitations include a 10 MB extracted-text analysis limit per step, no scanning of resources linked inside AI-powered steps, no inheritance of classification labels by AI-generated text, no scanning of files owned outside the organisation, and no policy check when a flow stars, marks read or unread, or archives an email. Custom starters, custom steps, third-party integrations and webhooks are, per Google's 17 September 2026 post, "OFF by default" and enabled by admins; webhook access can be restricted by an admin URL allowlist on supported editions, and approval requirements for custom steps, integrations and sensitive steps are set in the Admin console.
Default state
Conditional
Availability
Google announced general availability of Workspace Studio on 3 December 2025 and noted on 19 March 2026 that Workspace Studio had launched to all Rapid Release and Scheduled Release domains. The user Help Center lists eligible work or school editions: Business Starter, Standard and Plus; Enterprise Standard and Plus; Education Fundamentals, Standard and Plus; the Teaching and Learning add-on; and Google AI Pro for Education. Admin documentation for turning the service on or off lists the same edition set. Personal accounts are available only "for users with Google Workspace Experiments". Flows must be created on a computer with a supported browser — "Workspace Studio isn't available on mobile devices at this time" — though "Once active, flows can run on any device." Parts of the system remain preview or beta while the system itself is generally available: the "Start manually" starter is limited to the Gemini Beta program; Deep Research is marked Beta; third-party integration steps are marked Beta; custom starters and steps were introduced on 17 September 2026 with admin controls and are off by default; and identity attribution is Beta. DLP for Studio is available only on Frontline Standard and Plus, Enterprise Standard and Plus, Education Fundamentals, Standard and Plus, and Enterprise Essentials Plus.
Licensing
Included with the Workspace editions and add-ons Google lists for the feature (Business Starter, Standard and Plus; Enterprise Standard and Plus; Education Fundamentals, Standard and Plus; Teaching and Learning add-on; Google AI Pro for Education), and, per the August 2026 availability note, also Enterprise Starter and Google AI Ultra for Business. DLP for Studio carries its own narrower edition list. No separate price for Workspace Studio is stated in the documentation reviewed and none is inferred here.
External model or provider
Google's December 2025 general-availability announcement states Workspace Studio was built "Harnessing the reasoning power and multimodal understanding of Gemini 3", and the Workspace Studio product page states that Gemini 3 creates a flow from a plain-language description. Those statements concern the system's construction and flow creation. Current Help Center documentation for runtime AI steps refers only to "Gemini", Gemini Gems and Gemini Notebook without naming a model version, so the specific model executing any given step in September 2026 is not established. No third-party model provider is documented.
Limitations and uncertainty
Not established: whether the Workspace Studio Admin console service status is on or off by default for a newly eligible organisation (the admin article documents how to set it, not its factory value); precise run-history retention, retry behaviour and partial-execution semantics for failed runs; whether a shared flow's third-party connections are carried over, stripped, or must be re-authorised by the recipient (Google documents only that a teammate who copies a shared flow receives that copy's approval requests when their flow runs, and that AI steps are removed for under-18 school accounts); the exact OAuth scopes requested per step or integration and the runtime effect of revoking one; and any per-organisation limits on the number of flows, steps or runs. Documented limits and constraints: the "Start manually" starter supports up to 10 input fields; copied folders may contain up to 50 items and 3 levels of nesting; Gmail starter filtering supports only default Gmail labels, not custom labels; only 50 labels are shown in the labelling step; the Gmail starter ignores events created by other flows unless explicitly included; the "When a file is edited" starter does not fire on rename, move or new comment; the Gemini Notebook step provides text responses without citations; flow creation requires a desktop browser; and under-18 school accounts cannot use AI features. The Get Started guidance also states that flows fail where Shared Drives, shared folders or IMPORTRANGE spreadsheets are used and that "Files must be private to you", which stands in tension with the collaboration steps introduced in August 2026 for shared files; the current boundary between the two statements is not resolved by the documentation reviewed. This record covers Workspace Studio flows only. It does not cover Workspace skills (reusable instructions for Gemini), scheduled actions in the Gemini app, or Gemini Spark, which Google documents as a personal-account agent unavailable on work or school accounts.

Evidence