Skip to content

November 2026

Version Released
v5.12.17 2026-11-03
v5.12.18 2026-11-06
v5.12.19 2026-11-07
v5.12.20 2026-11-07
  • A Trust Center in the control plane gives your security team PlaidCloud’s security documents. Admin and Security users open Trust Center for their organization, download public documents such as the subprocessor list and DPA right away, and sign the security-documentation NDA once for the organization to unlock restricted documents such as the SOC 2 Type II report. Other Admin and Security users see who signed and when. An organization whose contract already includes an NDA can have it marked on file so documents unlock without signing. Restricted downloads are watermarked and recorded, and a new NDA version needs accepting again. Admin and Security users can also opt in, per organization, to a daily email summarizing new or updated documents and new NDA versions; it never contains the documents themselves. See Trust Center.

  • PlaidXL builds, drills, explains and plans in Excel. The PlaidCloud ribbon tab is grouped into Connect, Insert, Refresh, Drill, Write and Share, the common commands are also on a PlaidXL right-click menu and a PlaidXL Cell tab, and the main commands have keyboard shortcuts you can change, which don’t fire while you type in a field. A command that fails says which one and why. A command run before you sign in opens the sign-in and runs once you’re in, a sign-in works for one workspace so switching workspaces asks you to sign in to the new one (and switching back doesn’t), and with a Microsoft Entra work account in Office, PlaidXL finds your workspace for you. See the PlaidXL reference.

    • Build a report in a few clicks. The Report tab lays out a live report from a table, with a dimension drilled down the rows and another across the columns, and a CELL value at every intersection. Build my first report makes a sample from your own data. As you drill, a report inserts and removes whole rows and columns, and values you’ve typed, notes and formats stay with their row. See Build a Report.
    • Drill from any cell. Drill down, Drill up, Keep only and Remove only turn a node into a live DRILL formula and edit it for you, from a node, a dimension heading or a report value. Custom drill previews the rows before you insert them, and the member picker inserts members or adds a dropdown of a node’s children to cells. DRILL takes nodes to exclude or open, and can run across a row. See Drill, Pick Members and Explain Numbers.
    • Explain a number. Explain this number shows what a value calculates, whether it matches PlaidCloud now, and the table and workflow step it comes from. Show detail rows puts the rows behind a value on a sheet of their own, with a check that they add up.
    • Governed metrics in Excel. PLAIDXL.METRIC and PLAIDXL.METRIC_BREAKDOWN return a business metric as dashboards show it, and a CELL or CELLS with nothing to aggregate shows 0 for a sum and otherwise #VALUE! with the reason (blank in a CELLS spill), and Metric… lists your project’s metrics with their current values. A metric can filter on more than one alternate hierarchy and mix an alternate-hierarchy member with members of other dimensions. See Metrics.
    • Plan by typing into a report. Type a number over a CELL value and it’s written to PlaidCloud: spread in proportion, equally or in the shape of another member, adjusted by an amount or a percentage, with held cells left alone. Mean, median, minimum, maximum, standard deviation and variance cells take typed values too, and a typed value follows its cell when rows or columns are inserted or deleted (values in deleted cells are reported). Undo last write reverses typed values and saves, newest first, and only the person and workspace that made a write can undo it. Undoing a sum, or a mean typed into an empty cell, keeps what has been written into the same cells since, and a write that was interrupted before it finished can’t be undone. See Type Values Into PlaidCloud.
    • Refresh just what you need. Refresh a sheet, a selection or every table, with each table’s progress and Cancel in the task pane, or recalculate only the PlaidXL formulas. Large tables retrieve much faster in Excel on the web and in current desktop Excel (a 143,000-row, 77-column table takes about a minute instead of over ten; older Excel versions work as before), and text longer than an Excel cell holds is flagged and kept from being saved cut. Retrieved tables arrive as Excel Tables formatted by data type, keep your own columns and number formats when they refresh, and refresh in their own place. A retrieve or refresh never overwrites other cells or overlaps another Excel Table, and whole-number IDs too large for Excel and dates before 1900 arrive as text. Retrieving a table again doesn’t add it to the refresh list twice, Stop refreshing takes one off, and dimensions refresh with your tables. See Refresh All Tables.
    • Save writes straight away and can be undone. Save to PlaidCloud writes without a confirmation, says what it did, and offers Undo, which works for up to 7 days after each save on Lakehouse tables. A sheet retrieved with a filter, a detail-rows sheet, or someone whose access covers only some rows replaces only those rows, so other people’s rows are kept. A drill-through sheet whose members changed since you retrieved it isn’t saved, and a save that fails names the restore point so the table can be put back. See Save a Table to PlaidCloud.
    • Ask Weave in Excel. On the Weave tab, ask in plain words for a report, a formula, a drill or a table, or choose Ask Weave about this cell. Weave reads PlaidCloud as you and changes only the workbook, asks before replacing anything you’ve entered, never writes a formula that reaches outside the workbook, and each change can be reverted until you reload the pane. See Ask Weave in Excel.
    • Share a values-only copy of one sheet or the whole workbook. Convert to values converts a whole workbook, one sheet, or values with no way back (which asks first), changes only PlaidXL formulas and their spills, leaves formulas still calculating for a second try, and Restore PlaidXL formulas still works after you rename a sheet.
    • The task pane has a new look. It matches the PlaidCloud web app, with brand blue, rail-style tabs and lists, and tables, dimensions, projects and columns in bounded scrolling cards. Keyboard focus shows a clear ring, and the pane follows your light or dark Office theme. The tabs sit three across and two down in a narrow pane and in one row from 560 pixels wide. See Light and Dark Theme.
    • Numbers in the pane follow Excel’s display language. Row and column counts and the refresh, recalculation, Explain, Formulas, Report and planning messages show numbers the way your Excel does, so German Excel shows 4.812.233. Values written to cells don’t change.
    • A sheet’s tab turns yellow while it refreshes. During Refresh All Tables, whether from the pane, the ribbon or Weave, each sheet’s tab is yellow while that sheet refreshes, then returns to its previous colour, including your own, when it finishes, fails or is cancelled. If you save and close the workbook part-way, the tabs are put back the next time PlaidXL starts in it. See Refresh All Tables.
  • Google Drive document accounts work in workflows. A Google Drive account, set up with a Google Cloud service account whose key JSON you enter as the account’s private key, can list, import, export, copy, move, rename and delete files and create folders from a workflow. Its top-level locations are the folders shared with the service account and the Shared Drives it belongs to, and a Google Sheet reads as an .xlsx export. A folder in someone’s My Drive accepts overwrites of existing files but not new ones, so create the file first or use a Shared Drive. A missing folder lists as empty, and throttling and temporary Google errors are retried automatically. See Add Google Drive Account.

  • {date_iso} gives the run’s time in ISO 8601. It holds the current UTC time as 2026-11-12T22:49:56+00:00, so an expression can take its first 7 or 10 characters for the month or the day. It works wherever {date}, the long form, does: step settings and expressions, notification email templates including their preview and Send Test to Me, and Forecast, Parity and Solver steps. See Built-in Variables.

  • The Posting Register opens on what needs you. It opens on Exceptions when something needs a look, on Approvals when requests are waiting on you, and on All postings otherwise. The rail groups its screens into Needs me, Register, Insight and Settings, with counts on Exceptions and Approvals, and every screen shares one header and scope bar: pick a period (This month, Last month, Last 7 days, Last 30 days or Custom), a connection, project, environment or entity, and Live, Simulated or All, then Reset or keep a Saved scope. Your last scope is remembered.

    • Exceptions sorts out what needs a look. Chips for Needs a look, Unconfirmed, Pending, Failed and Reversal outstanding show their counts, a connection health line reads “All N connections posting normally”, and every posting has a plain-English state with a tooltip on what to do next. All clear shows when nothing is open. Paste a ledger id into Find a posting by ledger id, or the command palette, to open that posting directly.
    • Approvals, delegations and watches are inbox screens. Switch between Waiting on me and Submitted by me, Approve or Reject with a reason, approve several at once, and Undo an approval from the toast. Delegate approvals is a short dialog, and My watches lists what you follow. Reverse posting and the bulk retry and clear dialogs share one layout.
    • A posting’s drawer says what happened. It shows the state and its meaning, the key facts by name, and Lines, History and Sign-off tabs, with Mark reviewed, Retry, Reverse…, Watch connection, Watch run and Copy link, which opens that posting directly.
    • Assist, exports and retention follow the same design. Assist has an ask box and insight cards, advice marked Draft — nothing applied, and anomaly cards on All postings with Show me these. Exports & alerts has New schedule, and Retention is one card.
    • The register stays current. Rail badges update as postings change, open lists offer Refresh when postings have changed, and an open drawer updates in place. Home shows a Posting Register card under Needs attention. See Review the Posting Register.
  • Project home has seven pages. A text-only rail lists Overview, Variables, Schedules & sensors, Access & sharing, Data & connections, Data model and Apps & branding, with badges for the variable count, for active schedules and sensors, and for the element count. Overview shows the project’s description and labels, lock status, tiles for tables, workflows, steps, variables, published tables and Data size (which opens Storage), the same total as the Projects screen shows, earlier versions kept for restore included, with a tooltip that splits current data from earlier versions, the next three scheduled runs, and the project’s defaults, with Project report as a PDF or interactive HTML and Export archive. The other pages manage variables, schedules and sensors (with Calendar view and Activate or Deactivate), who can open the project and its dashboards and which projects can read its tables, connection and table-storage defaults with ERP posting mode, and Panel apps, branding and the project’s tile image. A locked project shows a banner and disables its controls. See Project Home.

  • Describe a shared column once, as a project element, and see how your tables connect. A column name several tables share, such as Period or Org, becomes an element with a description, type, mask, dimension and aliases. Every table with a column of that name shows the element’s description, mask and dimension wherever the column has none of its own, and nothing stored on a table changes. The new Data model page on Project home lists the elements, shows the tables that share each as a graph or a matrix, flags tables whose type differs, and suggests names worth making elements. The graph zooms with the mouse wheel or the zoom buttons, has Fit, 1:1 and a minimap, and exports to PDF or PNG. An element’s name appears beside its columns in Table Explorer, Column Properties and a step’s Data Mapping tab. Data Foundation projects come with elements for their shared Common Finance Model columns. The REST API and the element_find, element_upsert and element_delete MCP tools manage elements, and AI assistants see each column’s element meaning. See Data Model and Project Elements.

  • Documents opens on an Overview, with a rail of pages. Overview, Browse, Your accounts and Manage sit in a rail, and pages have a title, often with a subtitle, and their actions in the header. Overview shows tiles for your accounts, Pinned files that open Browse at their folder with the file selected, accounts by service and who can use them. Your accounts is one list of every account you can use, with a Shared or Private pill on each row. Manage lists the accounts you own, and a workspace admin sees every document account there and can manage any account they can otherwise manage (system-managed accounts are refused, and managed-bucket accounts can’t be cloned). A row’s ⋯ menu has Open, Edit, Clone…, Who can use it, Where used…, Version history, Enable or Disable and Delete. An account opens as a page with General, Access, Credentials and, for SFTP, SSH cards, and asks before you leave with unsaved changes. Browse opens your favorite account, or the workspace’s default, and has labelled toolbars with icons, an Upload button, a row menu, drag-and-drop upload (drop on the file list, a folder or a breadcrumb segment to upload there, or on the Root row or empty space in the tree for the account root), folders you can drag onto Root, and an Uploads and imports panel that shows each upload and import to a table with its progress and actions. Pop out opens the folder in a new browser tab. See Documents and Using Document Accounts.

  • Dashboards opens on an Overview, with a rail of pages. Overview, Dashboards, Apps, Notebooks and, for workspace admins, Access review sit in a rail, with Open navigator ↗ at the bottom, marked because it opens its own window. Pages have a title, often with a subtitle, and their actions in the header. Overview shows tiles for dashboards and apps that open the matching list already filtered, what you opened or starred lately, failed apps and access requests waiting on you, and the most opened dashboards for members with usage access. Dashboards filters by All, You manage and Favorites, and Apps shows each app’s Kind and live status. See Using Dashboards.

  • Identity opens on an Overview, with a rail of pages. The rail lists Overview, Members, Directory, Security groups and Distribution lists, then Sessions, Security log, API clients, Group audit and All scopes. Pages have a title, often with a subtitle, and their actions in the header. Overview shows tiles for members, active and never-signed-in members, builder seats and who is signed in now, an Active members chart, Needs attention with Resend welcome email and Review, and the last 7 days of security events. A member opens as a page with Profile, Access and Groups and data cards, and Invite member works from the Overview and from Home’s first-run cards. See Member Management.

  • Lists can show as Standard rows, Roomy rows or Cards. The View menu in the toolbar of lists such as Dashboards, Apps, the ERP register and Identity Members lets you choose between them. Cards keep everything the list does: selection, sorting, columns, right-click Copy and, where a list has favorites, the favorite star. Dashboards cards show a thumbnail. See Standard, Roomy and Cards Views.

  • The Lakehouse window has a side rail, table detail and time travel. Queries, Profile, Load and Catalog sit in a rail, with Lakehouse catalogs, Database users and Compute under Administration. Database users and Compute show only to members with the Lakehouse admin or Lakehouse cluster role. Queries filters history by SQL text, user, kind and duration, copies a statement with Copy SQL, and opens a query in Profile with Profile this query. Catalog opens a table’s columns, current snapshot, partitioning, location, properties and snapshot history, and Copy query as of this snapshot copies a FOR VERSION AS OF query that reads the table as it was then. See The Lakehouse Window.

  • Preview on step forms. Import: SQL, Execute: SQL, Extract: SQL, Export: SQL and Salesforce Query show up to 100 rows without running the step, and the Anaplan view and list reads show the columns and types they would produce. Click Preview on the Source tab. Variables are filled in as the run would fill them, but variable formats such as {x:>9} aren’t supported. Preview only runs standard SQL functions; otherwise it tells you to run the step. It isn’t available on locked projects, or for the Agent-based Execute SQL step. See Import SQL and Extract SQL.

  • Usage Reports open inside Identity, show what changed, narrow to a group, and flow into workflows. Usage reports under Insights in the Identity rail opens as a page beside the rail, not a separate window, with tabs for Overview, Activity, Paths, Members, Content, Exports and Funnel. Overview adds trend arrows against the previous period and a Highlights list that points to inactive members and to dashboards and apps nobody opened. Deactivated members aren’t counted or listed as inactive. Group limits Overview, Content and the inactive list to one security group’s members, and the date range adds Previous month and This month and is shared by every tab and remembered with your account. Every figure now has an explanation, tables use plain names such as Members and Area, rows open with Enter, loading and errors are clearer, and the last Funnel step shows how many members completed it. The Get Workspace Usage step writes usage into a table on a schedule, as daily or monthly totals with no member names, or as reporting-ready datasets: Members, Daily per member, Content and Content per member. The member datasets carry name, email, groups and activity, so you can join them to your own org table on email and report active minutes or dashboard opens by department. Each run, like Get Workspace Members, is recorded in the Security Log. See Usage Reports and Get Workspace Usage.

  • AI assistants can profile a table. Over MCP, table_profile returns the column statistics and Data Quality Radar checks that Table Explorer shows, plus each column’s most common values, for up to 50 columns at a time and only over rows you can see. See Reading the Profile From an AI Assistant.

  • AI assistants can unpublish a table and set a project’s default dashboard audience. publish_manage unpublishes a table along with its published view and any dashboard dataset no chart uses; a dataset a chart still uses stays, and so do the view and dataset when a report view is built on the published view, while the table still unpublishes. project_upsert sets dashboard_default_audience, the groups that new dashboards in the project are shared with, and [] clears it. See Unpublish and Set the Default Audience Through the AI Assistant or MCP.

  • Studio and App Runner apps live in your apps repository. A new Panel Studio app, and an App Runner app on the managed PlaidCloud Git connection, is created as a folder named for the app inside the workspace’s apps repository, beside your other Panel apps, instead of in a repository of its own. You can find and edit a Studio app’s source there. Studio keeps work that isn’t live yet on a draft branch named studio-<app-slug>, and Publish copies the app’s folder onto the main branch. Existing Studio and App Runner apps move into their folder automatically, and their old repository is archived: kept read-only, never deleted. A push to apps rebuilds only the app whose folder changed (or every app when the root requirements.txt changes). See Panel Studio.

  • A Git app lists your repositories and tells you what you can do. Git in the app list shows your role (Administrator, Contributor or Viewer) and the workspace’s repositories, each with its description, last-updated date, a clone address with a copy button, and Open. A Set up Git on your computer card walks through installing Git, creating a token and cloning, and How to clone reopens it. View all in Git opens the full Git interface and Manage who can push opens group management. Access is Viewer, Contributor (push app code to apps) or Administrator (push everywhere, including udfs), and a Viewer sees up to three workspace administrators to ask. The new Git Contributors group grants Contributor and can be a default group, so new members get it automatically; Git Administration grants Admin. A member’s Git tab shows their Git username, level and the group that grants it. A member’s Git account follows their membership: deactivating or removing a member blocks their Git account within about an hour, tokens and SSH keys included, and reactivating them unblocks it, with their repositories and history kept. Signing in to Git with single sign-on opens the member’s own account. Contributors don’t create repositories directly in the workspace’s Git organization; repositories are created through PlaidCloud. App windows show a Repository section and UDFs have Open, and Git has a ← Back to your workspace link. Every repository the platform creates, including app repositories and the built-in apps and udfs repositories, is readable by every signed-in member of the workspace, so members can open and republish apps without per-repository setup. A Git administrator can still make a repository private, which hides it from repository pickers and stops members publishing from it or running Git-backed UDF code stored in it. See PlaidCloud Git and Who can see repositories.

  • Cost to Serve and Operational Transfer Pricing templates build on the Common Finance Model. Start a project from Cost to Serve to spread operating expense to customers by the activity and driver you set for each department, and see each customer’s margin after the cost of serving them, tied back to the ledger. Start from Operational Transfer Pricing to test each entity’s profit level (return on sales, net cost plus or resale minus) against your policy ranges, with the true-up that brings an out-of-range entity back to target. Return on sales is EBIT divided by tested-party sales. See Cost to Serve and Operational Transfer Pricing.

  • OneStream and Anaplan feed Data Foundation, and OneStream has a read step. Data Foundation reads OneStream consolidated results through a data adapter your OneStream administrator builds, and an Anaplan plan through an export action, and reconciles the group profit and loss to them. The new OneStream: Read Adapter step reads an adapter, one call for each consolidation member, into a table in any workflow. See OneStream for Data Foundation, Anaplan as a Data Foundation Source and OneStream: Read Adapter.

  • Choose how Data Foundation spreads shared costs. Actions > Allocations… walks through which costs to spread, what to spread them by, where they land and what happens to a negative driver, for overhead to departments and for departments to products and customers, with a preview of one month that shows what is pooled, allocated and left unallocated. Rules you edit by hand are kept. See Allocation Rules.

  • Explain a number and name things your way in Data Foundation. Explain this number splits a profit and loss figure into what the ledger holds, what was allocated in and out, and what was eliminated in the group, with the largest source documents, under your own row access. Business names give categories, P&L lines, metrics and columns the names your business uses, in reports, metrics and the assistant. See Explain a Number and Business Names.

  • Data Foundation writes a board pack with every load and lets you go back. Every Load writes an Excel workbook and a PDF of the profit and loss dashboard to a folder of a document account; with reporting off the pack is the workbook alone, and the dialog says so. A restore point is made as each load starts and before each change to mappings or sources, and Actions > Restore… shows how many tables each one puts back and puts the project, dimensions included, back. The Foundation tab has its own rail with Status, Answers needed, Source coverage and Sources pages, with badges for what needs attention, one bar above every page to apply changes, and is clearer about what is happening: it shows an update that is still running or stopped part way, names each source and each hierarchy that failed in the last load, says why a folder of tables isn’t a source, keeps Check that totals tie out pending until something has tied out, and recovers when a status read fails. Applying an update keeps your own edits to generated steps and notes any conflict. See Board Pack and Restore Points.

  • Data Foundation consolidates several ERPs and systems into one model, and you choose which companies each one loads. Connect a system on the Sources page offers a tile for each kind of system, and Connect another system adds another ERP or another instance without replacing the ones you have. In Load order, Edit on a source has a Companies list of its subsidiaries: tick the ones to load, or none for all. Leaving a company out asks you to confirm first, because its loaded rows are cleared when you save, and the next Load pulls the ticked companies again. If the source’s connection now reads a different account, the save is refused with a message instead of changing another source. See Sources.

  • Clone a Data Foundation project, or save one as a template. Actions > Clone… copies a project with your model edits, metrics and analysis paths. Save as my template… saves a setup to a document account, and From a template file… starts a new private project from it, with its sources waiting to be connected. See Clone a Project and Save as My Template.

  • Data Foundation’s setup checklist runs through handover, and the assistant can resolve what’s left. The checklist adds the nightly refresh, the evidence pack and handing the project to its owner, and says who set the project up. Ask the assistant to propose answers for accounts without a type, codes several sources use and file columns that matched nothing; it writes nothing, and you apply the answers. See Get Started With Data Foundation and Resolve Open Questions With Ask.

  • Converted workflows open tidied, and Tidy all lifts loose notes above the flow. Workflows converted from Alteryx, KNIME, Prophecy or Savant open already laid out on the Advanced workflow canvas, with their groups kept and their notes moved to just above the flow. A group nested inside another in the source becomes a group of its own, named by its path, such as “Outer › Inner”. Arrange → Tidy all also moves notes and highlights that aren’t on or near any step or group to just above the flow, keeping their arrangement. See Arrange the Canvas and Containers.

  • Tell Weave what you want, and it builds and runs it. The dialogs that create a project, workflow, step, import or dimension, and empty lists and canvases, start with Tell Weave what you want, with Fill in the form instead beside it. Describe a workflow and Weave answers with one plan card; approve it and watch the steps appear on the canvas, marked Built by Weave, with Stop to end the build. Ask Weave to run a workflow, a range of steps or one step, to schedule a workflow or change its schedules, or to change or add a step: each is a card you approve, every run has ERP posting off, and Weave leaves steps that post to an ERP or write outside PlaidCloud to you. Drop a data file on a project’s Overview or a canvas for a table or an import step, drop or paste a SQL script to have it rebuilt as standard steps, and drop an Alteryx workflow, or name a folder of them, to have Weave convert them into a new project, run them and compare the results with the files they wrote in a move-in report. After an import, Suggested next step offers a first chart or clean-up to build. See Build and Run With Weave.

  • Weave sits beside every app, and you can ask it from anywhere. The Weave dock on the right of the window holds your Weave conversations, the changes Weave proposes, and your jobs. It opens with a click or Ctrl+J (Cmd+J on a Mac), remembers its width, and starts from what you have open, shown as chips you can remove. The Weave button, with its count of unread problems and its menu, sits in the dock’s header, and the folded dock shows a count of what is waiting for you. The current conversation’s title in the header switches to another conversation. Right-click a step, several steps, a workflow, a table, a column, a dashboard, or an error line in the log to choose Explain, Where is this used, What changed, Check, Fix, Change…, Document or Ask Weave. The quick answers say how Weave knows. The command palette answers questions with Ask Weave, help topics and Weave actions, and has a close button beside its search box. Log entries, error messages, Panel Studio and Support open the dock too. In the browser, click the microphone to ask out loud and hear a summary of the answer. Correct Weave and it offers to remember the term for everyone in the project. Nothing is scanned until you ask. See The Weave Dock.

  • Weave helps while you build a workflow. Empty fields in step forms show a suggestion you keep with Tab or Use: join keys, a new step’s column mapping, a filter or an expression, with where it came from. What next? beside a new step suggests the next one. Count rows counts what a step would write before you save, against its usual range, and resting the pointer on a step shows its table’s rows. Running a workflow checks its settings, and says something only when a problem is found, with Review and Stop run. A failed step’s card in the Inspector offers Fix or Ask Weave. See Weave in the Workflow Editor.

  • Fix a failed step in one press. Right-click the step and choose Fix, or use Fix on its log entry or in the Inspector. Weave shows a card with what will change, how far it reaches and what happens next. One approval applies it, even across several steps, re-runs the fixed steps with ERP posting off, and says plainly whether the problem is gone. When a scheduled run fails, Weave prepares the fix straight away and the card waits on the failed run in Monitor, on the log entry and in the dock, with Apply and re-run, and its owner is told in PlaidCloud and on their phone. Approval happens only in the browser. A card Weave prepared while you were away waits in its conversation as a Waiting for you row you open, and shows you how things look now if anything changed. Every change Weave makes that can be undone has an Undo, one press for a whole card, and it shows what it will restore first. See Fixing Workflow Errors with AI.

  • Weave jobs keep working when you leave. Choose Weave jobs… from the Weave button in the dock to see each job’s state, follow its progress, open its conversation or stop it. A job continues if you close the dock or the tab, and you get a message when it finishes or needs you, and, for a job that ran over a minute and wasn’t stopped, a phone notification if you use the app. See Jobs That Keep Working.

  • Weave has a Monitor. The Weave tab’s Findings page is now Monitor, with a rail that groups Findings, Team, Activity, Monthly summary, Expected and Requests, and the close checks, Readiness and Reconciliations, under Close checks. Findings opens on one line with Weave’s verdict, such as 3 open problems, 1 is yours., a note of what happened while you were away, and where your close stands. Tiles show open problems, unassigned problems and the oldest open one. Findings are grouped by severity and ranked by how much else they reach, and a failed run is one row, with the workflows and dashboards it affects folded under it. Each row has its owner, how long it has been open and one action, with Fix or Open first and the rest under ⋯. Open a row to read Weave’s sentence, the evidence, how Weave knows and comments, and filter by Owner and Workflow. Under Suggestions, a workflow someone runs by hand on a pattern gets Automate this, which proposes the schedule. This is expected marks a range as normal from now on, and Expected also lists the terms and conventions Weave has learned about the project. See The Weave Tab.

  • The close is prepared overnight. For a readiness set with a due time, Weave checks the close each day and after the runs it depends on, prepares a fix or a re-run for each problem it can, and shows one card, such as Month-end close is 92% done. Two things need you., with each decision numbered and its own card to approve. See The Close Prepared Overnight.

  • Keep Weave’s answers, and ask for reports and alerts. Keep ▾ under an answer keeps it as a standing request, a metric, a what-if scenario, a team recipe or a pinned answer. Ask Weave in your own words for a report or an alert on a schedule or after a workflow runs, and it tells you in PlaidCloud, on your phone and, without figures, by email. Manage your requests on Monitor → Requests with Run now, Pause, Change and Remove. See Keep Answers and Standing Requests.

  • Ask why a number is what it is. Right-click a cell in Table Explorer, or a point or cell on a dashboard chart, and choose Why this number: Weave names the step that writes it and what it comes from, and how it changed since the run before and since last period. In Table Explorer, As of lists a table’s earlier versions by the run that wrote them, and Compare with now says which rows and columns changed since. Ask a what-if about an allocation’s result in words for an instant estimate. A reconciliation’s last result is kept, such as Tied out to Cost Pool at 06:02. See Why a Number Is What It Is.

  • Share Weave with your team. Pin an answer and everyone in the project sees it, re-run with their own access. Save a question as a recipe that anyone can run from the dock with /, or on a trigger. Both are listed on the project’s Overview under From your team. Steps, workflows, tables and findings have comment threads where you can mention a teammate or ask @Weave, and Weave shows on the canvas while it works on a workflow. See Share Weave With Your Team.

  • See what Weave was worth. Some findings ask whether they were worth telling you, with a thumbs-up and a thumbs-down. The Monthly summary shows how useful Weave’s findings and answers were, the failed runs Weave fixed and the pinned answers teammates reused, with an estimate of the time saved, for the project or for you. See Weave Monthly Summary.

  • Voice in Weave on mobile takes one tap. Tap the microphone, ask your question, and Weave sends it once you stop talking. Tap the microphone again to stop early, and it turns off by itself if nothing is heard. If you leave the Weave tab or background the app mid-sentence, your words stay in the question box unsent so you can review them. Run, Pause, Stop and Resume ignore an accidental double tap on their confirmation. Requires app version 1.3.0 or later. See Use Your Voice.

  • Weave emails you about problems. Turn on, for yourself, an email when a workflow run ends with a serious problem for you to look at, and a daily summary, at a time you choose, of the problems Weave told you about that you haven’t read. Emails name the project and each workflow, step, table or metric with the kind of problem, link back to the Weave tab, and never include figures or values from your data. Workspace administrators can turn Weave email off for everyone and set quiet hours, in each member’s own time zone, that hold alerts until they end. See Weave Email.

  • How Weave uses your data. A new panel, in the Weave menu, below every AI chat and in AI Features, says what Weave sends to its AI provider and how the data is handled. The provider retains none of your data, your data is isolated per workspace, and it is never used to train AI models. See How Weave Uses Your Data.

  • Tidy all chooses the order within each column. Arrange → Tidy all no longer keeps the top-to-bottom order you had: steps still flow left to right and the layout stays where it was on the canvas, but the order within each column is chosen to reduce crossing links. Tidy selection keeps the order of the items you select. See Arrange the Canvas.
  • One filterable checklist assigns people, groups and workflows everywhere. Security group members, a member’s group associations, distribution list members, document account members, groups, agents and owners, connection access and a Weave readiness set’s workflows use the same picker instead of two lists to drag between. Type to filter by name or email, including a domain such as @example.com, paste a list of emails to match exactly those people, switch between All, Assigned and Not assigned, and use Assign all shown or Unassign all shown. Rows you change carry a Will add or Will remove pill, Undo changes puts them back, and the save button is available only when something changed. Enter in the filter never saves, and Space ticks the selected row. Each row shows a name and an email in columns, and the summary shows how many are assigned and what is pending. Where you can see access but not change it, the list is read-only, with a note and each person’s initials in place of checkboxes.
  • Weave on mobile asks you to agree to its AI notice again. The notice now says Weave is powered by a third-party AI provider that PlaidCloud calls on your behalf. The provider retains none of your data, keeps it isolated and doesn’t use it for training. See Privacy and Consent.
  • The What if window reads one period at a time. When an Allocation Split step’s output has a period column, the window opens on the latest period, and Period: lets you choose another or All periods. A step whose driver has a line for every period, too large to read across all of them, now opens and answers for one period. A saved scenario keeps the period it was saved for, if any, and opening it moves the window to that period while it is still listed. See Choose a Period.
  • PlaidCloud’s web app and Python client library use the REST API, and JSON-RPC is deprecated. The web app sends its requests to the REST API. With plaidcloud-rpc 1.25.0 or later, SimpleRPC calls go to the REST API with no change to your code; a call the REST API doesn’t cover yet, or a workspace still on an older release, uses JSON-RPC as before. JSON-RPC will be removed in a later release, so upgrade your scripts and notebooks to plaidcloud-rpc 1.25.0 or later and write new integrations against the REST API. See Command Line.
  • A step follows its project’s connection environment unless you fix it to another. Switching a project between environments, such as QA and Production, moves every step. A step that must stay on one environment can be given a Fixed environment instead: the form marks it with a warning, its log notes each run on an environment other than the project’s, and switching the project names the steps that won’t follow. ERP posting steps always use the project’s environment. An environment can’t be deleted while a project, a fixed step or an agent uses it, and the agent form’s field is now Environment this agent serves. See Fixing a Step to One Environment.
  • Step forms follow one layout. Tabs have standard names in a standard order: General, Settings, Source, Columns, Filters, Output and Advanced, with Header columns and Line columns on ERP posting forms. Required fields are marked with *, and a form only blocks saving for what the step needs.
  • Acumatica, Bill.com, Dynamics, Eclipse, HFM/FCCS, NetSuite, Workday and Workday Financials connections can be created, edited and picked in steps without losing settings. They are saved as REST connections, and the Connections list shows their own type. Right-click a connection and choose Edit REST settings to open its headers, authentication and pagination. Test Connection works on saved NetSuite and HFM/FCCS connections. A Workday Financials connection signs in with an integration system user and its password, given together, or with an OAuth client, and the Workday, Bill.com, Dynamics and Eclipse forms have an API host field. See Create and Manage a Connection.
  • Data-entry forms put each field beside its label. Fields are left-aligned at one consistent width, not pinned to the far right, in Data Foundation Connect and Sources, document account settings and the member profile.
  • Copy Project chooses table data and dimension members separately, copies data faster, and keeps a private copy’s ERP write authorizations. Under Data to copy, choose Table data — No table data (the default), Full or Sample — and, separately, Copy dimension members, which is on by default, so a copy without table rows still arrives with populated dimensions and is quick to make. Full and Sample copies copy several tables at once, so projects with many tables finish faster. Copy Results lists each dimension’s outcome alongside the tables, and Retry failed items retries failed tables and dimensions together. A new copy made with data is private to you and keeps the source’s ERP write authorizations that you could grant yourself; if the source posts Live, the copy arrives in Simulate, so nothing reaches the ERP until you switch it to Live. The REST API and the project_copy MCP tool take dimension_members too. See Choosing How Much Data to Copy and ERP Write Authorizations in the Copy.
  • A table’s details panel shows the steps that use it sooner. Selecting another table clears the previous table’s steps straight away, and steps in other projects that share this project’s data are added a moment after the rest instead of holding up the whole panel. See See Where a Table Is Used.
  • The Time & Expenses Invoices list opens fast and loads more on request. It shows the 100 newest invoices matching your status and search, and Load more adds the next 100 when more exist. Choosing All or Paid applies for that visit only; the next visit opens on Open.
  • Every table has the same layout, filter and row actions. Your own actions come first on the left, such as New and Open, then Actions, View where a table has one, and Refresh as an icon at the far right. No toolbar has a More menu: less-used commands live in Actions, such as Storage and Download on Tables, Stop All and Clear Statuses on Workflows, and Move selected item, Diagnostic mode, Lock workflow, Dependencies and Flashback on a workflow’s detail window, and the move arrows stay on the toolbar. The Filter sits at the bottom of every table, beside the row count; press / or Ctrl/Cmd+F to jump to it, and in tree views one filter drives both the tree and the flat list. The Tables list has an All | Published switch beside the filter, and a dimension’s hierarchy has a Main | Both | Alternate switch choosing which tree the filter searches, with Show Alternate Hierarchy, columns and layout in its View menu. A locked workflow shows a Locked marker on its detail toolbar; click it to unlock if you can edit the workflow. Connections and a project’s Security list gain a filter. Data shares, Scheduled events, Flashback, a workflow’s run history and a member’s Row Access list now filter, refresh, show a row menu and open with Enter like the rest, with F5 to refresh. Delete asks you to confirm, which data shares didn’t before. In every dialog Cancel, or Close, is the rightmost button. Project home’s Variables page, Members, the Posting Register’s postings list and the Lakehouse catalog’s tables card no longer show a second count of their rows in a footer or subtitle.
  • Put things at the top level of a project. Folder pickers for dimensions, tables, UDFs, workflows and data editors offer /, the project’s top level. New Folder, when the selection is inside a folder, has Create at the top level. In a project’s explorer trees, drag items onto the empty space below the last row to move them to the top level. See Move.
  • Delete and clear steps won’t match everything at the top level. Table Drop, Table Clear and Dimension Delete, set to match a search, won’t save with empty search text while the search path is the project’s top level, which would delete or clear every table or dimension in the project. Enter text to match, or choose a folder. See Table Drop, Table Clear and Dimension Delete.
  • The Multi-Table Join designer zooms like a canvas and exports to PDF or PNG. The mouse wheel zooms without a key held, the toolbar shows the zoom level with zoom buttons, Fit and 1:1, a minimap shows the whole graph, and the export button downloads a PDF, with Export as PNG in its menu. See Designer Toolbar.
  • Row actions are the same everywhere. Open a row’s menu from the ⋯ button at the start of the row, by right-clicking it, or from the toolbar’s Actions menu. It always lists Open, Details and Edit first, then the list’s own actions, then Delete or Remove, with an action that doesn’t apply to a row greyed out in its place. Enter and Del are shown on the items they run. A list whose rows have a single action shows that action’s icon instead of ⋯, and Dashboards, Apps and Notebooks show the Open icon again. See The Row Actions Menu.
  • The New app menu names each kind of app, and your account details are in the avatar menu. New app on Apps has a Panel app heading over Studio (write and publish in PlaidCloud’s editor with the standard layout; your code is kept for you), Browser only (your own repository; runs in the viewer’s browser) and Server based (your own repository; runs on a PlaidCloud server), then Dash app and From gallery…, and the Type column uses the same names. Your name, email and Connect Your AI are in the avatar menu instead of the header bar. See Creating a Panel App and Connect Your AI Agent.
  • PlaidCloud has a refreshed blue look, in light and dark. Buttons, checkboxes, selected rows, tabs, links and the rails that list a workspace’s pages use one PlaidCloud blue, with the contrast needed to read them in both themes. App tab icons are one consistent set.
  • New loading screens. PlaidCloud opens on a cleaner loading screen that follows your light or dark theme. A Panel or Dash app waking from idle shows its progress through Starting, Preparing and Loading app, and if it doesn’t start, says which step it stopped at and offers Try again. See Using Panel Apps.
  • Saved ERP write steps with no validate setting open as validate-only. This covers SAP Post, SAP Attach, NetSuite Post, Oracle EBS GL and AP Post, and Xero export. Steps explicitly set to run live stay live.
  • Long-running jobs show progress the same way everywhere.
  • Workflows stop, fail and report honestly. Stopping a workflow stops everything it started, a run or step that can’t finish fails with a reason instead of hanging, and run status matches what actually ran.
    • Stopping a workflow stops everything it started, and a run or step that can’t finish fails instead of hanging. Stop cancels import and dimension steps that were still queued, so later imports and dimension loads in the workspace start normally, and it also stops the workflow’s queued or running connector (Singer) syncs. If a step’s job can’t start, the step fails with the reason instead of waiting forever: no worker starts it within about 40 minutes, a worker started it and then went away, its pod is stuck Pending for about 20 minutes, its image can’t be pulled for about 5 minutes, or no pod is created for about 20 minutes. A connector sync that no runner picks up within about 30 minutes fails with a message naming the sync. While a step waits for a worker, its status message reads “Waiting for a worker…” with the minutes so far, and clears once the work starts. Stopping a macro stops every running copy of it, while stopping one copy, or a Macro Concurrent step giving up on a copy that stopped responding, leaves the workflow that called it running. A macro running only inside other workflows shows as Running in the workflow list, with Stop available and each step’s status, start and end time from its latest run. A run whose server is lost ends as failed, and one still running after 12 hours fails with “Run failed: still running at the 12-hour limit. Re-run the workflow.” See Run a Workflow.
    • Resume, run history and run status match what actually ran. Resuming a parent workflow runs everything a child workflow had left, even after you re-ran one of the child’s steps on its own, and each call of a child workflow within one run has its own run-history entry. A step that restarts because its server stopped says so in its log, and workflow_run_status and the REST API’s run outcome list the steps that failed but continued, in continued_step_ids and continued_step_count, instead of reporting the run as clean. See Pause, Stop, and Resume and Continue on Error.
    • Step errors show their own message. An error PlaidCloud didn’t anticipate reads “The step stopped on an unexpected error (type): message”. The Needs attention cards on the Launcher show the error in one line and leave out runs of workflows that have since been deleted. See Managing Step Errors.
    • An ERP extract that doesn’t finish fails its step, and Resume continues it. When SAP: Ad-Hoc Extract pauses at its row or time limit or its agent doesn’t reply in time, or a Salesforce, Oracle Fusion bulk or HFM/FCCS export job is still running when the step stops waiting, the step fails so the steps after it don’t run on a partial table. Resume the workflow or re-run the step and the extract continues from where it stopped, on the same job. See Run Limits and Salesforce: Query Object.
    • Cancelling an export stops its query in the warehouse. When an export or download is cancelled, including by stopping its workflow, the query reading its data stops too on Lakehouse v2, Snowflake, Databricks, PostgreSQL and Greenplum, and an export to a remote database stops reading its source, instead of running on to the end.
  • REST, Import REST and Import Postman steps are converted to REST Request when your workspace upgrades. A step that can’t be converted cleanly stays as it was. See REST Request Step.
  • A dashboard link opened after your session has expired takes you to that dashboard once you sign in. See Using Dashboards.
  • MCP tools refuse a field they don’t recognise and name it in the error, instead of ignoring it. workflow_upsert takes memo and accepts description as another name for it. See Unknown Fields Are Refused.
  • On Lakehouse v2, the lakehouse writes Parquet exports to Google Cloud Storage itself. An export to a Google Cloud Storage document account goes straight into the account’s bucket, written with the account’s own credentials, instead of every row passing through PlaidCloud first. Exports to Amazon S3 and Azure Blob Storage accounts work as before. See Export to Parquet.
  • Slack, Microsoft Teams and Web Hook steps say why a message wasn’t sent. The step’s error names the server it tried to reach and, for Slack and Teams, the connection, and says what went wrong: for example, Slack no longer recognises the webhook, the channel can’t be posted to, the server can’t be found or doesn’t answer within 60 seconds, or the connection has been deleted. A Slack or Teams webhook URL is never shown, because anyone who has it can post as your integration. See Notify via Slack.
  • Access tokens for a project’s service account last at least 24 hours. Workflow jobs that run as the project, such as UDF steps, get tokens that last 24 hours. A shorter lifetime is raised when your workspace upgrades, and a longer one you set is kept.
  • A workspace’s organization, owner, release and version are set by PlaidCloud, and you choose its data center. In the control plane, a workspace you create is owned by you and placed on the release channel PlaidCloud sets, and you choose its Data Center from the regions offered at signup. Organization, owner, release and version are read-only when you edit a workspace. Use Set Version to change versions, and ask PlaidCloud Support to move a workspace, change its owner or channel, or pin it. See Workspace Configuration.
  • The encoding conversion steps can keep a folder’s subfolders. Document: Convert File to ASCII Encoding, Document: Convert File to Unicode (UTF-8) Encoding and Document: Convert File to Unicode (UTF-16) Encoding have a Keep Subfolders box. Tick it to write each file from a subfolder of the input folder into the matching subfolder of the destination; the contents of a compressed file go into the subfolder it was in. Left clear, every file is written into the destination itself, as before. See Convert Document Encoding to UTF-16.
  • A step whose source tables don’t match its settings stops before writing anything, and names every problem at once. Table Extract, Table Lookup, Table Inner Join, Table Cross Join, Table Outer Join, Table Anti Join, Multi-Table Join, Table Union Distinct, Table Union All, Table Pivot, Table Melt, Table In-Place Delete, Table Upsert and Table Append check their source tables before they write. A column the step reads that the table doesn’t have, a join key that is missing or not in the step’s source columns, or a source table the step reads that hasn’t been built stops the step with one error that names each problem and how to fix it, and a target table that already existed keeps what it held. Table Union Distinct and Table Union All are the exception for a source table that hasn’t been built: as before, they leave it out without a warning and combine the rest, and with none built they finish with a warning instead of failing. A column the step lists but doesn’t read, and the table no longer has, is a warning instead, so a step that ran before still runs. For a step whose target is a view, the columns and join keys it lists are checked against its tables only where the warehouse would refuse to create the view anyway. These failures used to come from the database, in its own words and one column at a time, and a step that replaces its target table had already emptied it. If a workflow’s error handling matches the database’s text, update it. See When the Step Runs.
  • An export with no tables stops instead of finishing. An Export to Parquet, Export to ORC or Export to Avro step with no tables finished without writing a file, and an Export to XML step with no tables wrote a file holding none. These steps now stop before writing anything, with No Tables Specified. A step saved with no list of tables at all, which failed with an unexpected error naming an internal setting, now stops with the same message.
  • A workflow started by a Workflow: Run step can use the values its caller’s run was started with. Answers given in the App Runner, and values passed when a run is started, reach the workflows the run starts, for any variable the called workflow and its project don’t define themselves. See Run Workflow.
  • Tidy all keeps steps inside the groups of converted workflows. In a workflow converted with groups nested inside other groups, Tidy all pulled steps out of their groups. It no longer does, including in most workflows converted earlier, whose nested groups have been separated, keeping their names and layout. See Containers.
  • Invalid project settings are reported with a clear message. An unknown environment or document account, a slug that is taken or reserved, or an invalid storage type or posting mode now returns a message naming the problem instead of a generic server error. Through the REST API this is a 400 rather than a 500, and through the AI assistant or MCP it is an invalid-argument error.
  • The REST API’s dimension values request works. POST /rest/v1/analyze/query/retrieve-dimension-values was answered as a grid request and refused as invalid. It now returns one aggregated value for each column in value_columns, and a grid, one value for each pair of a row filter and a column filter, has its own POST /rest/v1/analyze/query/retrieve-dimension-cells.
  • Slow table reads and file handling no longer hold up other requests. Reading a table’s columns from the warehouse, registering a Data Foundation file, exporting to Excel, reading a REST or SOAP import’s response, finding where a document account is used, and checking a Panel app’s code or logo could each pause every other request the same server was handling until it finished. They now run alongside those requests.
  • The project and workflow variable dialogs scroll. On a small screen or low resolution the add and edit dialog was taller than the window, so the Create and Update buttons could be out of reach. The form now scrolls and the buttons stay at the bottom.
  • Files over 256 MiB in an Azure Blob Storage document account can be copied, moved and renamed. Each of these used to fail on a file that size, and so did moving or renaming a folder holding one.
  • On Lakehouse v2, a table’s row count reflects what was just written to it. A step or import that wrote a table could record the row count from before the write, so a table just loaded showed 0 rows and an import could report 0 rows imported. The count is now read once the new rows are visible everywhere, so the steps after it read them too. Switching a table to another branch, or merging a branch into it, also updates its row and column counts.
  • Weave’s empty-values check compares each run with what earlier runs really left. A step’s empty values were often counted just after the step emptied its table to refill it, so its earlier runs looked as if they had none, and a column that had always been partly empty was reported as if it had only just become so. Each run is now counted as the step left its table. Counts recorded before can’t be told apart from good ones, so they aren’t used: each step’s empty-values check starts its history again, and is reported as unchecked until five runs have been counted. The rows and spread-of-values checks also no longer credit a step with a write another step makes to the same table more than two seconds after it finishes. See What Weave Checks.
  • Data Foundation keeps the precision of inverted exchange rates and decimal thresholds. A rate PlaidCloud turns around, because your rates are quoted from the reporting currency into another currency, could be rounded to six decimal places, which left amounts in a currency with a small rate, such as the Indonesian rupiah, around 1% off. Inverted rates now keep 12 decimal places. A group_amber or group_red threshold with decimals, such as 0.5, was rounded to a whole number. Your next Load uses the precise values. See Exchange Rates.
  • Alerts and digests stop for a deactivated member. A Panel app alert or digest whose owner was deactivated kept checking and sending. It now pauses while they’re deactivated and resumes if they’re reactivated. See Alerts and Digests.
  • Asking the REST API for a deleted connection returns 404. Requesting a connection that has been deleted, or an ID that doesn’t exist, answered with a generic server error (500). It now returns 404 with a message naming the connection ID.
  • Templated table exports work through the AI assistant and MCP. The templated format of table_data_export failed whenever it was given a table. It now exports the table through your template, limited to the rows you can read. In the template, a table’s rows are named by its table ID with each - written as _.
  • Amazon S3 and S3-compatible document accounts keep their Region. Saving one of these accounts discarded the Region you entered, so the field was blank the next time you opened the account. The Region is now saved. On an account saved before this fix, enter the Region again and save. See Add AWS S3 Account.
  • Converting a folder’s encoding no longer overwrites a file with another of the same name. Two files bound for the same place in the destination — two files of the same name from different subfolders, say — left only the last one written. The step now stops before writing the second, names both files and where they would go, and says when ticking Keep Subfolders would keep them apart.
  • Clearing Replace characters in Document: Convert File to ASCII Encoding takes effect. A step saved with Replace characters the target encoding cannot hold with ? ticked stayed that way after the box was cleared and the step saved again. Clearing it now turns replacement off.
  • The AI assistant and JSON-RPC no longer name the steps that use a table your role hides. Asked where such a table is used, they named each step your role can see that reads or writes it, with its settings and workflow. They now say that steps use the table without naming them, the same as the project’s table references. See What Your Role Hides.
  • The default AI connection you choose stays the default. Marking an AI connection as the workspace default appeared to succeed but was never saved, so agents and steps that rely on the default kept using whichever connection was updated most recently. Your choice is now saved. See LLM Providers.
  • Right-clicking a conversation in Weave’s history no longer opens Rename on top of Delete Thread. Rename is now on the chat’s tab: right-click the tab, which until now offered nothing. See The Tabs in the Weave Tab.
  • A blank Memo no longer stops you saving. Creating or saving a project, dimension, table, warehouse, editor, scheduled workflow event or distribution list with the Memo box left empty failed with “Memo: Input should be a valid string”. A blank memo is now accepted.
  • An expression that gives several values where one is expected is refused, and a Table Extract or Table Union All step keeps its target table’s data. The usual cause is a trailing comma, as in table.amount,, or a comparison with several values, such as table.region == ('EU', 'US'). On Databend the step failed, and a Table Extract or Table Union All step had already emptied its target table by then. Some other warehouses ran such an expression, but not always as written, so a step that has been running without errors may now be refused. The step form refuses the expression when you save the step, and the message shows what to change. See Several Values Where One Is Expected.
  • A step that names an output column twice stops before emptying its target. In Table Lookup, Table Extract or a join, two Target Output Columns with the same name made the step empty its target table and then fail with an unclear error. The step now stops first, with a message naming the repeated column, so you can give each output column its own name. Extract SQL, which writes columns in the order your query returns them, is unchanged. See Table Lookup.
  • A union source saved without its source columns says what’s missing. A Table Union All or Table Union Distinct source saved through the REST API, JSON-RPC, or the AI assistant or MCP with no list of source columns failed when the step ran, with an unexpected error naming an internal setting. Saving one is now refused, naming the source, and a step that reads one saved before stops before it writes anything, naming the source. A converted Table Union All that takes every column from its tables, or stacks them by position, needs no list and isn’t refused for having none.
  • A column’s constant can be a number, 0 or false. A step whose output column had a number as its constant failed with an unexpected error. A constant of 0 or false was treated as no constant at all, so the column took its expression or source column instead, or the step failed because the column had nothing to read. These constants are now written to every row, converted to the column’s data type, and the AI assistant can propose a step with one.
  • Converted Alteryx Data Cleansing tools remove null rows. A Data Cleansing or Data Cleanse Pro tool with Remove Null Rows ticked was flagged for review in the conversion report, and its converted step kept every row. The step now removes each row that is null in every column, as Alteryx does. A row holding an empty text value is kept. Data Cleanse Pro checks each row after its other options, so a row whose nulls it replaces with blanks is kept too. Remove Null Columns is still flagged for review. Import the workflow again to pick this up.
  • Email notifications with no columns listed include every column. A Notify Via Email frame with no columns chosen used to fail the run. In a template body, {variable} placeholders in the subject are filled in for scheduled sends, preview and Send Test to Me.
  • Expression validation accepts the functions a run accepts. Valid expressions that call functions such as int(...) were refused when you checked them.
  • Loading a dimension from a table works on projects with Row Access. On a project that restricts rows, the load failed, so Data Foundation’s hierarchies weren’t rebuilt. Project Architects and workflow runs now load it. Anyone whose Row Access limits what they can read of the source table is told to ask a project Architect or to load it from a workflow, because everyone in the project sees the members a load makes. See Loading Data Into a Restricted Table.
  • Data Foundation loads, and makes the Pipeline dashboard, on projects with Row Access. On a project that restricts rows, Load and Apply changes could stop while building the pipeline by stage, and the Pipeline dashboard wasn’t made, because its table couldn’t be published for reporting.
  • Salesforce, SAP, Oracle E-Business Suite and Oracle Fusion sources load their data into Data Foundation. Reading them during Load failed, although the load itself finished.
  • Accented and other non-English characters read from Salesforce come through intact. Names such as Hernández came out garbled from the Salesforce: Query Object step and from Data Foundation’s Salesforce sources.
  • Key marks on the Data model graph no longer run together. When several tables were keyed on the same element, the 1 marking each table’s key was drawn where their lines left the element, so the marks ran together, as in “11”. Each 1 is now drawn at the table end of its line, just left of the table card. See Read the Data Model.
  • Weave no longer sends your name and email address to its AI provider at the start of each conversation. It never used them.
  • Password sign-ins need an authenticator app, and passwords are stronger. A member who signs in with a PlaidCloud password is asked to set up an authenticator app at their next sign-in, and enters its code each time after that. Single sign-on, Google, Microsoft and GitHub sign-in, API tokens and service accounts are unchanged. A new password needs at least 12 characters and can’t be your username or one of your last six passwords, and after repeated failed sign-in attempts the account is locked for a short time. A stricter rule your workspace already had is kept. See Member Authentication.
  • Connections point at public and private-network addresses. A connection can’t point at a loopback, link-local or cloud metadata address, and the error names the host and the address it resolved to. Addresses on your private network are unaffected. See Addresses a Connection Can’t Reach.
  • Saved connection credentials are shown only to members who can edit connections or workflows. Through the REST API and MCP, a saved connection’s passwords, tokens and keys are masked for other members, who can still open and browse it. See Who Sees a Connection’s Credentials.
  • App builds run without access to your workspace’s Git credentials.
  • An agent’s private key is shown only to members who can edit agents. The key no longer appears in agent lists or details through the REST API and MCP, and the Agent form shows the Private Key and View RPC Config only to members who can edit agents. Service accounts, project and agent tokens, and AI-delegated tokens can’t retrieve the key or an RPC config. Running agents are unaffected, and the public key is still shown. See Obtaining Agent Credentials.
  • Proxy download links are valid for one hour. With Use PlaidCloud Proxy Download on, open the download again for a new link when one has expired. See Services.
  • Adding people to security groups follows the permissions you hold. You can add a member to a security group, choose a member’s groups or invite an Architect when you hold every permission the group grants. Workspace admins are not limited. See Managing Security Groups.
  • A connection’s owners, or a workspace administrator, change its settings and switch it on or off. Changing a connection’s settings through the REST API, JSON-RPC and the AI assistant or MCP takes one of its owners, or a workspace administrator for a connection with no owner. Enabling, disabling, locking and unlocking a connection take an owner or a workspace administrator, and a request naming several connections applies only when you can change all of them. Only a workspace administrator can choose the workspace’s default AI connection, and a connection a member owns is the default only when no connection owned by an administrator, or with no owner, could be. See Edit a Connection and LLM Providers.
  • Row Access applies to everything a member does with a restricted table. Members see and change only the rows their grants allow, in dashboards, in their own SQL, and in loads.
    • Project Architects see all rows of row-restricted tables in dashboards, matching what they see in Analyze. Taking away Architect access takes effect in dashboards immediately — whether you change someone’s role, remove them or their group from the project, change the project’s access type, restore an earlier project version, or delete a security group. See Managing Row Access.
    • Members limited by Row Access can run their own SQL queries and export the results. On a project with Row Access, a member who isn’t an Architect gets exactly the rows their grants allow from their own queries on query screens, and can export them to CSV, Excel or JSON, with the Rows limited by your access notice; before, these were refused. A grant value containing a percent sign (%) now returns its rows on query screens, and downloading such a query as CSV works. Their own free-form SQL on Snowflake or Databricks, exports of their own SQL from Postgres-family warehouses, and reading a governed table as of a snapshot are still refused, and a view must keep the governed column under its original name to be filtered. See Row Access and Queries You Write Yourself.
    • Saving Row Access keeps a granted value that its source doesn’t have. A value a security group was granted stays in the grid, marked, when the table has no rows with it, the dimension doesn’t have the member, or the list doesn’t have it. Save asks whether to keep each such value or remove it. See The Row Access Grid.
    • DataFrame writes, record saves and clears into a row-restricted table change only the rows you can see. When Row Access limits your rows, writing a DataFrame to a restricted table or clearing it with the Python client, or saving records to it or clearing it through the REST API, changes only your own rows, and a load carrying a row outside your access is refused before anything changes, naming the values it found. Project Architects, members who see every row, and workflow runs write the whole table. Each of these writes that a person makes, to any table, is recorded in the Security Log. See Loading Data Into a Restricted Table.
    • Imports, copies, restores and deletes on a row-restricted table need every row to be visible to you. When Row Access limits your rows, importing a file into a restricted table, copying a table into it from another project, switching or merging its branch, restoring it, deleting it, bringing it back after it was deleted, or any other action that replaces the whole table runs only when you can see every row of the table; otherwise it is refused and nothing changes. Saving a table’s rows through the REST API changes only your own rows, as the other loads do. Project Architects, members who see every row, and workflow runs are unaffected, and these actions are recorded in the Security Log too. See Loading Data Into a Restricted Table.
    • ERP post steps run on projects with Row Access. A step that posts to an ERP, such as Workday, runs on a row-restricted project, and snapshot, branch and optimization-candidate listings work there too.
  • Only an account’s owners can restore its earlier versions. Restoring a document account to an earlier version, holding or releasing a version, and taking a backup take one of the account’s owners, and none of them is available on a system-managed account. Seeing an account’s version history needs access to the account.
  • query_run runs read-only statements. The query_run tool that the AI assistant and MCP clients use accepts one SELECT or WITH statement, an EXPLAIN of one, or a SHOW; any other statement is not run. To change data, use query_transform. The SQL in an MCP table export follows the same rule. Through the REST and JSON-RPC APIs, it also applies to running a query, building column facets, downloading a query as CSV, exporting a query to a remote database and exporting table data with your own query, for members who can’t change project data; a refused query returns 403. Explain Analyze runs the query to measure it, so in the query log and the lakehouse it accepts only a query that reads, and on a project with row access only a project Architect can run it from the query log.
  • Table Explorer filters, queries sent to PlaidCloud and the query that finds a Table Extract step’s Dynamic Columns can’t call functions that run SQL or read the server’s files. A call to one, such as query_to_xml, dblink or pg_read_file, is refused, and the error names the function. In the same places, .op() accepts only an operator and a literal_column can’t use Unicode-escaped quoting. The Dynamic Columns query carries the step’s filter and the expressions that read the column names from the source, so one that breaks a rule fails the step when it runs. See Functions and Operators That Are Refused.
  • Data Foundation allocation rules refuse a department or statistic name that contains a single quote, a backslash or a line break. The name is written into the allocation steps’ filters, so Save and a preview of stage 1 refuse it and name the value. See Allocation Rules.