Agent Studio

Hey,Our organization is using Moveworks Copilot and Gemini. I am wondering if it is possible to merge both to build something similar to 'Quick GPT,' where users can go to our Moveworks Copilot and ask Gemini for answers.I am thinking of building an Agent Studio plugin and adding triggers like 'Ask Gemini,' but I am wondering if there is a better and more straightforward way.Also, is it possible to build a fallback mechanism where, if the bot cannot find an answer in the internal KB or Forms, it suggests the Gemini plugin?

Hi Moveworks Community,Our organization is transitioning from managing distribution lists (DLs) in Active Directory (AD) to managing them entirely in the cloud. While we know Moveworks has native functionality for DL management in AD, we are seeking guidance on whether similar capabilities are available for cloud-based DLs or if they can be adapted to meet our needs.Specifically, we are looking for the following capabilities for cloud-based DLs: Provide the list of owner(s) of a distribution list to a user Allow users to request and receive details of the owner(s) of a specific DL. Provide the list of members of a distribution list to a user Allow users to request and receive the list of members for a specific DL. Delete a distribution list If the owner requests deletion, the DL should be deleted immediately. If

I have a few plugins that collect information to fill out a ticket. The different fields to fill out are slots in the conversational process. The slots have a resolver strategy to pull the possible options from a api. I have the slots set to Infer slot value if possible. However inferring never actually works when a resolver strategy is applied.For a access management plugin, I can say I want access to <exact application name in system> and the plugin will show the first x options. It normally ends with the correct option and says things like, if you mean <exact application name in system> type “yes” or the number in the list. It however always makes the user confirm the option instead of inferring.I am getting around this by throwing a generate_structured_value_action at the beginning after asking for a description of the ask, but it is possible the user won’t provide enough information to determine the correct option and it feels like a hack when slots sh

Hello, Nevermind

ScenarioWe’re seeing a small but recurring UX gap in our Moveworks → UKG plugin flow and wanted to check if others have solved this or if there’s a recommended pattern.Current behavior User asks Moveworks to perform an action (example: approve timecard). Plugin detects expired/invalid OAuth token for UKG. Moveworks prompts the user to re-authorize via an auth link. After successful authorization, the user must explicitly type something like “retry approve timecard” to continue. Only then does the plugin re-execute the original action. From the user’s point of view, it feels broken or incomplete — they believe the task should continue automatically once acces

Hey everyone 👋We are excited to announce User Consent Auth in Agent Studio, which lets you securely deploy Agent Studio use cases that respect user permissions and access levels, is now in Preview!What are the capabilities of this launch?Permission Consistency. Ensure users only access data they are permitted to see. Any user interacting with an app through Moveworks does not gain privileged access through a generic service account. User Control. Individual users can grant access to their own data. Unlike JWT where a developer can impersonate any user, this allows users to control exactly what types of data and actions the connector is allowed to interact with. Tokenized Security. Short-lived access tokens are generated and continuously rotated with longer lived refresh tokens to reduce the risk of damage from a privileged access token bei

We’re working on a use case where users request specific software, and our organization maintains a list of available software on a SharePoint site. Our solution uses a compound action: it fetches the software list via HTTP and script actions, then filters and displays results based on what the user asks. We’ve added both positive and negative utterances as well.The issue is that when someone asks for a particular software (for instance, “I need Teams”), our workflow works sometimes. But if the user says something like “I need to install Teams,” the system often triggers an internal plugin—such as installation help forms or knowledge base articles—instead of using our custom plugin. Even though our internal team has disabled some pre-built plugins, we still get unwanted results from forms. This means the intended custom plugin either fails or returns unnecessary or irrelevant information.Sometimes it works (“I need Teams” gives the right result), while other times, lik

Background:In June, before Webhooks were available, setting up instant approval notifications through the Concur API was difficult because Concur doesn’t have native automation tools. At that time, we were told we’d need to regularly poll Concur records to check for updates, since Webhooks weren’t an option yet.Current Situation and Questions: Webhooks Availability: Now that Webhooks are available, can we use them to automatically trigger approval notifications in SAP Concur? In other words: Instead of our bot checking for updates all the time, can Webhooks now handle this automatically? Impact on Our Bot Performance: Given that we would like to use real-time Webhooks (and not scheduled ones), what does this mean for our bot’s performance? Specifically:

Form update conditions don't support $LENGTH() for string validation. Is this intentional or a limitation? Any workarounds?this is in reference to Rich Ticket Filing. I was trying to do some validation, because ServiceNow cuts off the short description after 160 characters. And we have folks that use that field for the description. Basically, my intention was to show the helper label that explained the limit if the user entered more than a certain number of characters in that field. I tried a few different conditions like:$LENGTH(field_short_description) > 143field_short_description.$LENGTH() > 143field_short_description.$LENGTH > 143and also the one you see in the screenshot, all of them seemed to result in the error “Invalid DSL expression” which makes me think that length isnt available to u

Hey builders!Office Hours #2 is in the books and we went FULL dev mode today 🔥The questions were chef’s kiss, we went deep on the exact pain points you’re hitting in production right now.Here’s everything we crushed:Using a single webhook listener to trigger multiple ambient agent plugins → Use event filters to route one webhook to the right plugin (PTO vs Approval demo’d live) → No more “one webhook per use case”— one gateway listener to handle all plugin triggers “My API returns a giant payload and it has no filtering capabilities” survival guide → Chunking + parallel blocks + LLM actions = 🏃‍♂️ → Pagination patterns in compound actions that work “My connector has way too many permissions” → Always. Provide. Access. Rules Default slot values without prompt engineering? → Optional slots with default fallbacks ar

Here's the situation we face when a user submits time through the Workday Portal: our bot can successfully approve or deny the time off. However, the problem happens when a user submits time directly through the bot. In these instances, the approval/denial process doesn't seem to work as expected. The error we get back is “This event step is not awaiting your action”. When I review the data through the RaaS report, it appears that working time-off requests have two true approvers, whereas those that are not working only have one. The API call we are using to request time off is /api/absenceManagement/v1/{{tenant}}/workers/{{worker_id}}/requestTimeOff.Has anyone experienced this issue before?

At present, there is no support for analytics related to ambient agents. Although logs of successful notifications are available in the logs section, corresponding analytics data for ambient agent use cases is not accessible. This limitation is impacting the integrations we have implemented to monitor success metrics.Could you please let us know when will you introduce analytics support for ambient agents?

If you use a slot resolver for a conversational slot and set the Slot Inference Policy to Infer slot value if available, it will pull from the User Profile.. for example, their location from ingested user data.  This can be an issue for some plugins where similar data exist but has nothing to do with what the plugin is asking for.  It would also be beneficial to be able to define what each ingested element from the User Profile is, allowing context for Moveworks to better understand when to use those slots.Example:We’ve built a Conversation Flow where one input is a state, and we provide a list of states. If a state matches the user’s profile, the Infer slot value will automatically select that state instead of presenting the choice to the user—even if we add notes in the plugin description, conversation process, and input variable advising not to use the user profile location. This matters because the state they select is inte

Would be helpful to have an SDA on/off toggle on each plugin as we’ve noticed that it calls on some plugins that we don’t want it to run on. 

Hi Team,We have a usecase to build a solution to provision servers in Zscaler using MCP servers. We want to know if we have existing frameworks / connectors from Moveworks to MCP server to achieve these kind of usecases.

I found Moveworks documentation showing that we can export plugins between environments, but I don’t see anything about exporting HTTP actions, conversational processes, script actions, compound actions, etc.Is there a way to export everything related to a use case—not just the plugin—so we can move our work between environments?Right now, we usually build directly in production and just use test users, but sometimes we build and test in sandbox first. When we do that, it’s a hassle to manually copy and paste everything from sandbox to production. It would be extremely helpful to have an easy way to export all of this work between environments.

Hey builders!We wrapped 2025’s Office Hours series yesterday with a cozy, high-signal session – I was out sick (stay safe from the flu!), so Steve Rankich stepped in as host with Ryan Geroulo riding shotgun. Even with holidays in full swing, the questions were gold – all about making your agents smarter, more reliable, and less noisy.Here’s the recap of what we unpacked:Available Data Bank Pop-Up – Your New DSL Best Friend  → No more hunting through actions to remember output keys! The new “Available Data” modal shows exactly what you can reference: action outputs, slots, and the all the user fields (time zone, employee ID, etc.). → Drag-and-drop or copy-paste – instant data.user.time_zone wins. Huge quality-of-life upgrade for chaining actions cleanly. GPT-5.2 Now Live in LLM Actions → Fresh model support dropped: GPT-5.2 (pinned & unpinned), plus 5-Mini/Nano/04 flavors.

Hi, I was looking into the Agent Marketplace use case for Workday – Notify on Time Off Request Approval. I got stuck at a point where we need to fetch event steps using the URL/ccx/api/businessProcess/v1/{tenant}/eventStepsThis endpoint requires parameters such as businessProcess and stepType, which are mentioned as IDs in the documentation. Could you please clarify from where we can fetch these IDs for our tenant/environment?

Our current manual process is as follows: Step 1: create user accountStep 2: license the email account with a G1 licenseStep 3: convert the mailbox to a shared mailbox Hoping somebody may have done this or used a combination of moveworks and powershell scripting.  

I have been getting error since this morning while testing a compound action: "AttributeError('link_context_infos')"This was working perfectly fine until Thursday night before I logged off for the week; nothing changed since then but running into this weird error. I tried to build from scratch but still same issue.  any insights are greatly appreciated. 

Referencing the March 2025 Product Release Notes and implementing the following compound action:steps:  - action:      action_name: mw.get_user_by_email      output_key: output_user_to_notify      input_args:        user_email: data.email      delay_config:        seconds: data.delay_in_seconds  - notify:      output_key: output_test      recipient_id: data.output_user_to_notify.user.id      message: data.message We cannot suppress the message after the reminder notification is sent.  The message states: “you’ll receive: <message content>” after the no

Hey Agent builders!Session #3 dropped today and we were cooking – we dove into fresh releases, live demos, and those "aha!" unblocks that make building feel like magic.Quick hits from the hour: List Slots – No more prompt gymnastics → Tick the "Data type is expected to be a list" checkbox and voila: collect arrays of strings, ints, booleans, or custom types without hand-holding the assistant. → We did a live demo showcasing the power of this new feature! Dynamic Resolver Context Passing – Databank to Resolver, Seamlessly → Feed data bank outputs (e.g., activity results from your data bank) directly into resolvers without brittle input args. → We also did few live demos of this new feature! Taming Giant Payloads: Data Mappers + DSL Filter Magic → API spewing massive JSON? Slice it down with output mappers and DSL functions like $FILTER – trim n