Agent Studio

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

The first red box is a script activity that will return a list of systems the user has a locked account in.The second red box I need to run against that list, up to 4 times. So, if a user has a locked account in 2 different systems, I need to run the unlock action for each one. Does anyone have any ideas how I might achieve this?

We are using OAuth 2.0 with Authorisation Grant in our Moveworks plugin integration. On first authorization, we successfully receive both an access token and a refresh token. However, the access token expires after 24 hours, and when users query the plugin after that, the plugin directly fails instead of prompting the user to re-authorize.Expected behaviour :  When the access token expires, the plugin should use the refresh token to fetch a new access token silently. If that also fails (e.g., refresh token expired), it should prompt the user to re-authorize.Current behaviour :  The plugin fails immediately on expired access token without asking users for re-authorization, resulting in poor user experience.Request:  Could you please advise the best practice or configuration in Moveworks to handle access token expiration and ensure the plugin prompts users for rea

Hi ​@Ajay Merchia  Are the other two content type activities disabled?Knowledge and Form Referencing the 

Hello Community!I wanted to share a useful architecture for managing Agent Studio Notifications that allows end-users to opt-in or opt-out of specific proactive messages.The Solution: A Moveworks Notifications TableTo support custom user notifications, this strategy involves provisioning a Moveworks Notifications Table (e.g., in ServiceNow or any iPaas tool) dedicated to housing user notification settings. This table tracks user preferences for specific proactive use case notifications (e.g., approval reminders, training due reminders, direct report birthday reminders). Any proactive message sent—whether triggered by a Moveworks Scheduled Job, a ServiceNow Flow Designer Flow, an Azure Function, or any other system—has the ability to cross-check this Moveworks Notifications Table. The message is only sent if the user has actively

Does Movework allows to create and access a datastore, this data store can be - SQL, NOSQL or VECTOR STORE. This would allow users to store and access data from persistent store which is indexed.Use case: If we refresh CMDB and pick CI directly from MW Platform and doesnt have to make any API call to ServiceNow or other ITSM platform to fetch CI Data.

Hello, We are facing an issue with the MW generic approval template(mw.create_generic_approval_request). We are able to build the approval request successfully, but when we trigger it to the user, the notification contains the header as approval_details. Whatever we pass as “approval_details” appears as the header. This is fine when we add one or two lines of approval details, but when we add a list of information for approval, it looks bad for the end user and it’s duplication valuesIs there any way to eliminate or truncate these details in the header?  

Sharing an important update for everyone using the Snowflake Dynamic Query / Snowflake Cortex Analyst plugin.Snowflake has announced that they are deprecating single-factor password sign-ins, including for service accounts. Once this enforcement begins, password-only authentication will stop working, and Snowflake will require MFA or non-password mechanisms for programmatic access.Reference:🔗 https://docs.snowflake.com/en/user-guide/security-mfa-rollout  Our Snowflake connectors today are set up using username + password. If these connectors remain unchanged, they may start failing with authentication errors when Snowflake completes this rollout. Snowflake recommends migrating to one of the supported authentication methods for automated integrations:Key-pair authentication (public/private key)

Ambient agents anticipate needs and take action autonomously, bringing humans into the loop when it matters. They’re always-on, system-driven AI agents that operates in the background and spring into action the moment they’re needed. They don’t wait for a user prompt. They listen. They act. They serve as an invisible, intelligent workforce.Learn all about what this means by clicking HERE. And then come back to share in the comments what you would build with an ambient agent 😄

How do I keep the bot from returning the resolved input arguments in a reply back to the user? I have tried adding the instruction to “NOT show resolved input arguments in the message to the user if information needed to be resolved” in the plugin description.  Haven’t had any luck with this though in Agent Studio. This was not created via a compound action, but does involve a decision policy with 2 actions. The required slot for the decision policy is is_department. Any help would be appreciated! 

Hi community,I am having an issue where the llm instructions and guidance are being shown to the user? “Need to call tool again with chat_history. No need to ask user per tool instr; we should summarize ourselves.” “Need to provide chat_history back to tool. They told: first ask user for missing args, then after they provide, call tool before messaging—but system also says don't message before tool except greetings. But here tool explicitly requires ask user. So ask: briefly summarize history ourselves and send as chat_history without needing user? But instructions say ask user and then call tool after user provides. So I must now ask user something, then when they respond, call tool. For now I should ask them to confirm / describe issue and mention reinstall option will come later. Question only.”Does anyone has this issue also and any workaround - I think this is related with slot resolvers.

Hi All, I am trying to add an approval workflow using the inbuilt Action “mw.create_generic_approval_request” however I am confused where to add the approver email address?. I am getting only two input parameters as given below. approval_details: <!-- Placeholder for approval_details -->users_requested_for: <!-- Placeholder for users_requested_for → Any help on this topic will be appreciated.

You can now collect multiple values for slots from users and pass conversation context directly into dynamic resolvers in Agent Studio. This is enabled for all customersWith List[t](Multiple-Value Slots), you can let users provide several entries at once like multiple users, strings, IDs, or any supported data type without repeated prompts. Collect multiple values for a single slot Works across all supported slot types (String, User, Email, etc.) Returns a clean array for use in resolvers and actions With Dynamic Resolver Context Passing, you can pass data gathered in conversation straight into your dynamic resolvers. Pass slot values, action outputs, and metadata into resolvers Chain resolvers together to create hierarchical slot resolution No more manual workaround

Hi Team,I’ve recently noticed that Knowledge and Forms features in Agent Studio Content Activity seem to be unstable for me. Specifically,When I add a knowledge article in a content activity, the bot displays empty “” results. For forms, instead of  the usual buttons to open them, I only see a hyperlink.Is this a bug? Has the product team already identified this issue? Thanks in advance.

Hi All,I am working on a simple compound action use case where I am pulling data from one of the ServiceNow tables. Even though the use case is quite straightforward, it doesn't seem to be working as expected.This is one example, but I have observed similar behavior with other use cases as well.I requested fields like number, justification, and status. I have also mentioned these fields in the use case description, but the result is not aligned with the code written. Compound actionResult Anyone experienced similar behavior with Agent Studio.  Regards,Rajeev

Hey Developers!Last week we kicked off our very first Developer Office Hours and wow – the energy was 🔥 many of you showed up live, we crushed a bunch of real-world questions you submitted ahead of time, and we even squeezed in a live demo of the brand-new File Slots feature that just dropped into Limited Preview.Here were the hottest topics we tackled together: Slot descriptions & smart defaults → How to make the assistant infer “today” vs “tomorrow” timestamps without always prompting the user (hint: clever slot description + decision policy + DSL date functions = 🤝) OAuth 2.0 user consent flows (Google Workspace, GitHub, etc.) → Live walkthrough of the authorization-code grant + the exact connector settings you need so your agents can act on behalf of real end-users Dynamic slot resolvers & output mappers →

Is there a way to customize the bot response or use the decision policy to route the response based on error status code received from an API call? Example:I have an action that submits an email. If the email is valid the API returns the required data if not valid it returns Status Code 400 error. Status Code: 400Response:{ "code": "error", "message": "Email not valid."}Email not found in the system means their account doesn’t exist and the bot should relay access request information. I’ve attempted using the decision policy using if data is valid or not but with no luck.

You can now collect files from users right inside Moveworks!With File Slots, you can design plugins that request and send files (like receipts, ID documents, invoices.) directly to external systems.Single-file uploads via a built-in “Upload File” button in chat Integration with HTTP actions through flexible file mapping Supports raw upload, multipart/form, multipart/related, application/octet-stream, and base64Want to learn more about the feature (& see a demo)? Check out our latest docs here:https://help.moveworks.com/docs/file-slots#/ https://help.moveworks.com/docs/http-action-files#/ Limited PreviewWe’re launching File Slots in a l

I am trying to build a conversational process that can target either a user's email or computer name. Is it possible to configure a slot to automatically handle user input as follows?If the input is an email address, use it as the slot value. If the input is not in an email address format, attempt to find a user by that name and use the matching user's email address as the slot value. If no matching user is found, use the original input text as the slot value.I plan to use a decision policy to process this slot differently based on whether it’s an email address or a computer name.Thank you.

We are looking to open up the ability to users to drag and drop documents into the chat bot so that they can be sent to other destinations.  What we are finding (Chat Bot is in Teams) is that when someone drops a file in chat, it puts that file in the users own OneDrive but the chatbot only knows about the URL and Name of the document not and of the Data in the document.Has anyone else attempted this and how have you done it?  If not is this something that may be coming in the future?I feel like this could be really userful even if it was something like how Brief Me does it and the document is stored for X amount of time then removed.  But atleast you could still process the documents.  Im thinking something like HR documents, Claims Documents, etc.

I want to ingest files in sharepoint folder and build custom plugin to search dedicated folder using compund action. As per the Agent Architect, this is feasible but i am not able find the steps suggested by tool.Step 1-- I cannot find so please help me navigate and also verify if this solution will work or any better solution if you have. 

Hello Team,I am working on a CMDB search skill, when there is no data i would like to show few default messages, somehow its now working.. but with data case is working well… below is my default  instructionany tips and tricks would be appreciated.  display_instructions: '"You must always respond with exactly one complete sentence in plain English.  Do not use bullet points, line breaks, or labels under any circumstances.  Do not explain the search process or mention that you looked for keywords.  If there is application data available, combine the application name, owner,  description, tower, and platform executive into a single user-friendly summary.  If there are no results or data is missing, you must respond only with this text,  \"No application information found based on exact keyword search.\"  In all cases, after the sentence, you must append the following text exactly and without modification

Hi communityI have noticed now we have live suggestions/data entries:In the conversational process there is a slowness because this is always saving and removing me from typing.Furthermore while creating a script in the script actions I have the same situation I cannot code there anymore - since this is always suggesting entries and stops my typing and everything breaks and is slow.Is anyone having the same issue?I understand this is a new feature but it would also a possibility to have the option to disable it? or solve it by increase the speed of the instance. This might also be a issue with the browser - dont know - I am using EdgeThank you

I have two plugins built in Agent Studio—View Addresses and Update Address—that are meant to work back-to-back: users typically view their current address, then say “update address” to start the edit flow. When the update is invoked immediately after a view, the handoff does not preserve context: the update flow behaves as if state is lost, and it doesn’t re-display the primary address as part of the update confirmation steps (we must display the primary address to show the user what they are updating). Notably, if the user starts with “update my address” directly (without viewing first), the flow works correctly, which suggests the issue is specific to inter-plugin state/intent handoff or session variable carryover. 

we already have movework chat-bot application integrated within Teams.and Now, we have created our own AI-Agent application. what is the best possible way to integrate it with movework chat-bot.