Product Forums

We've just launched a public archive of The Agent Builder newsletter with technical breakdowns, implementation patterns, and real builds from our dev community. Click here to head to the archive. What's inside: Technical deep-dives, real-world examples, code snippets, best practices, and platform updates.Browse the archive anytime or subscribe for inbox delivery at aka.mw/newsletterShout out to ​@Kevin Mok for this amazing resource.Happy building!

📊 How Feedback Is Collected & How to Get All Feedback Data Today, users can give feedback on a bot response in two steps:1️⃣ Thumbs Up / Thumbs Down (Quick Feedback) Users react with 👍 or 👎 on the bot response This action always gets recorded, even if the user does nothing else 📍 Where this data lives Table: interactions Filters to apply: interaction_type = link_click interaction_content = helpful or not helpful ➡️ This captures all feedback clicks, including cases where users do not submit written feedback.2️⃣ Optional Written Feedback (Form Submission) After clicking 👍/👎, users are taken to a feedback form. Submitting the f

Notifications API is listed as part of the future roadmap for Data API here. I am looking for insights into AI Assistant notifications such as Concierge and CSAT. When will this be supported?

Do you have questions about building with Agent Studio? Want to connect directly with our Developer Advocate team?We’re hosting live Developer Office Hours — an open space where you can get help, share feedback, and learn best practices from our developer advocates.These sessions are your chance to:Get hands-on guidance from our Developer Advocate team Troubleshoot challenges in real time Exchange ideas and tips🗓 Upcoming sessions:Nov 13, 2025 — 1:00 PM PT / 3:00 PM CT / 4:00 PM ET Nov 20, 2025 — 1:00 PM PT / 3:00 PM CT / 4:00 PM ET Dec 11, 2025 — 1:00 PM PT / 3:00 PM CT / 4:00 PM ET Dec 18, 2025 — 1:00 PM PT / 3:00 PM CT / 4:00 PM ET👉 Register here to join any sessio

I wanted to get clarification on lifecycle management for custom views created. I created a pilot HR Custom View and shared it only with myself for testing, but I’m not seeing any self‑service way to edit, delete, or re-share that view after it has been created/shared. Based on what I’m seeing, it appears I would need Moveworks support to update/remove/reassign views—which could result in multiple duplicate views as we expand sharing to Little Joe domain admins.   Could you please confirm the following? Is this expected behavior? Once a Custom View is created (and/or shared), is it intentionally not editable/deletable/re-shareable by admins? Is there an admin role or feature we’re missing?

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

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?

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:

ProblemA failure in ingestion from a single application currently stops the entire ingestion flow, preventing updates even when other sources are healthy.Request:Introduce an active–passive failover switch for each application connector within the ingestion pipeline. So if there is an issue with one application the whole ingestion process still completes and only attributes used from the failed application is not updatedBenefitsPrevents full ingestion stoppages Ensures maximum data freshness from available sources

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

Hello Moveworks Admins,We're excited to announce that User Geocodes will be available in the Imported Users screen on 7th January 2026. This new feature provides visibility into user location data, making it easier to troubleshoot location-based answers not getting served and validate identity configuration. 🌏 What are Geocodes?Geocodes are structured location coordinates extracted from location-related fields (city, state, country, zip code) in your identity source during user ingestion. This allows Moveworks to know “where this user is”, which is then used in search and personalization—for example, boosting KB documents that match the user’s region when the user asks a generic question like “what are my benefits?”Read more about Geocodes in User Identity.

 Hey Team , User ID and Email will be added to the new Interactions table starting Wednesday, Jan 28th'2026. (3 weeks from now). This means that anyone with access to the new Interactions table will be able to see User ID and Email associated with each interaction.Since the legacy data-masking behaviour will not carry over, your governance controls will now come from our new Custom Views framework. Below are some important links for you to get familiar with the new governance framework.Academy video Governance best practices

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.

Data API throwing “API not found” errors for multiple Gov endpoints:https://api.moveworksgov.ai/export/v1beta2/records/interactions  https://api.moveworksgov.ai/export/v1beta1/records/interactions https://api.moveworksgov.ai/export/v1/records/interactions { "error": { "code": "NOT_FOUND", "message": "API not found" }}OAuth 2.0 token is working as expected (works with other APIs endpoints such as Events (previously) and Webhooks)This also fails from Data API tool: 

I’m facing an issue with the search functionality in the Moveworks setup when trying to check ingested files.Currently, when I type a file name in the search field, the cursor automatically jumps out of the search box. Because of this, I have to click back into the search box repeatedly to continue typing. This makes it very difficult to type the full file name, and the experience becomes quite frustrating.At the moment, the only workaround I’ve found is to copy and paste the file name directly into the search box to avoid the cursor jumping issue. Can you please help look into this or suggest if there’s a known fix.

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?