September 2026

Two and a half minutes on seven of this month's changes: the Posting Register, seven new connections, AI-built steps you approve first, Weave run monitoring, Row Access on one screen, column profiling, and moving projects to Lakehouse v2.

Releases Shipped

Version	Released
v5.11.20	2026-09-05
v5.11.21	2026-09-08
v5.12.0	2026-09-11
v5.12.1	2026-09-11
v5.12.2	2026-09-16
v5.12.3	2026-09-17
v5.12.4	2026-09-21
v5.12.5	2026-09-25

Added

• The Posting Register gives finance and accounting one place to review ERP postings across every project, and to act on them in bulk. It opens on an exception queue of the postings most likely to need attention (in doubt, unconfirmed, pending or failed), with simulated postings counted separately.

  • One window. A navigation rail opens each screen in place: postings, exceptions, approvals, delegations, roll-up, trends, control totals, lineage, exports and alerts, and AI assist. Pick a project, connection or entity by name, then bookmark the view or save it.
  • Act on many at once. Select rows, or everything your filter finds, preview the impact, then clear simulated postings or retry only what the ERP rejected. A high-value action asks you to type the amount before it goes ahead.
  • Read a posting the way an accountant would. The Journal tab shows debit and credit lines by account, each line’s dimensions or worktags, and a balance check for each currency that never mixes currencies. An amount that wasn’t captured shows blank, not zero.
  • Reverse with segregation of duties. Reversals go to every qualified approver through an approvals inbox, with delegation and escalation. A whole intercompany run reverses at once and balances to zero in each currency.
  • See the pattern. Trends break failure rate and turnaround down by connection or ERP, Lineage traces a posting back to its run and out to the ERP document, and Control Totals give each batch a verdict and the reason for it. Export to CSV or JSON on demand, on a schedule or as an alert. Closing a period seals its postings into a snapshot that can’t be changed.

  Four permission levels (view, operate, approve and admin) govern the register, and you see only what your level allows. What you can see follows your existing connection and project access, and exports leave out personal details and mask sensitive values. See Review the Posting Register.

• Each workflow run can post to your ERP live, simulate the posting, or skip it. Simulate builds and checks exactly what each step would post without contacting the ERP or committing anything, and records the would-be entries in ERP Post History marked as simulated. Off skips posting steps entirely. A project can set its own limit, and a run always uses the more cautious of the two settings, including in sub-workflows, macros, loops and resumed runs. A posting step created without a Validate Only setting (through the API, the AI assistant or an imported bundle) now previews instead of posting live. SAP posting, attachment and RFC steps can’t simulate safely, so Simulate and Off skip them. See Choose a Posting Mode.

• Seven more systems connect to PlaidCloud natively. You connect each one from the Connections screen, it has its own workflow steps, and its objects, cubes, views or entities are listed live from the system.

  • Xero. Sign in and pick your organisation; you don’t need a developer account. Import eleven accounting entities filtered by status, date and last change. Post manual journals, invoices, bills, payments and credit notes back, with a test mode that checks the batch first and protection against posting the same document twice. If Xero doesn’t confirm an outcome, PlaidCloud holds it for you to review. See Connect to Xero.
  • Salesforce. Connect with your External Client App’s credentials, and Test Connection shows which organization and user your data comes from. Extract any object in bulk, see each field’s type before you save, and filter at the source. Tell a run which period to load, and a re-run recognizes when it already has that period. See Connect to Salesforce.
  • SAP ECC and S/4HANA. Build an extract yourself by searching for a table, transaction code or saved SAP Query, and SAP’s own definitions set up the step. Columns land with the right types (dates, exact decimals, whole numbers and timestamps), and extracts run through a PlaidLink agent under SAP’s own authorization checks. Advanced filters, such as exclusions, pattern matching or more than one condition on a column, need the current PlaidCloud SAP package on the PlaidLink agent’s system; an older package still runs simple equals and range filters. A column too large for its decimal type lands as text, and a column whose SAP type changes replaces the whole table, each with a warning naming the column. See Ad-Hoc Extracts from SAP ECC / S/4HANA.
  • Oracle HFM/FCCS. Pick a point of view in a workflow and get that slice of your consolidation data back as a table, without raising a ticket with the HFM team. Repeat pulls come from cache, results are ready for row access, and any slice can become a scheduled load. See Connect to Oracle HFM/FCCS.
  • IBM TM1 / Planning Analytics. Run a saved cube view or your own query, or load a TM1 hierarchy into a PlaidCloud dimension. TM1 doesn’t track changes, so tie the point of view to a period variable and let a schedule build up history one period at a time. A hierarchy PlaidCloud can’t represent faithfully is refused, not loaded wrong. The TM1 server needs a network path PlaidCloud can reach, or you can run cube queries through a PlaidLink agent. See Connect to TM1.
  • Oracle Essbase. Run a live query, or load an outline into a dimension. A hierarchy that can’t be represented is refused by name. If you already use Essbase, three things change: a cell you can’t access now raises an error instead of landing as text, labels come from the connection’s alias table (so label text may differ), and the row limit defaults to 250,000 instead of unlimited. See Connect to Oracle Essbase, Essbase: Query Cube, and Essbase: Read Dimension.
  • Anaplan. Read-only, so nothing is ever written back. Read a saved view, run an existing export, or read lists into a dimension. Anaplan limits requests across your whole tenant, so stagger your schedules. See Connect to Anaplan.

• The Project (Time & Expense) app is now one workspace for time, trips, approvals and billing. Every screen opens in place from a navigation rail, and opening a record shows its detail in a panel on the right.

  • Time is a weekly grid that suggests rows from your recent work, with a note on any entry and Copy last week. It flags a long day, a weekend or a project over its hours without blocking the save, and keeps a history of who changed what. A sheet your approver returns reopens as a draft with their note attached.
  • Approvals shows each member’s hours and billable share beside their sheet. Approve several sheets at once, Return for Changes with a note, or Reopen a sheet you’ve approved to correct it. Nobody can approve their own sheet.
  • Acting as lets an assistant enter a member’s time and trips for them, with every change credited to the member.
  • Trips shows your trips and receipts, with a receipts PDF for each trip. Finance assigns unassigned trips to projects and marks them billable.
  • Setup covers master projects, each with a bar showing how much of its PO is used, plus their sub-projects, billing roles, team assignments and work locations. People shows every member’s approvers and assistants.
  • Billing covers accounts, invoices and voids. Add unbilled time or trips to a draft invoice in one click, or void a finalized invoice to return its items to unbilled. Home shows finance what’s ready to bill.
  • Submitting, approving, returning and reopening a sheet each send an email, and anyone who hasn’t finished the current week gets a reminder on Friday. Every list exports to CSV.

  Rates and amounts are visible only to members with billing permissions. On every screen, old or new, you must be someone’s approver to read their time, expenses or receipts, and nobody can approve their own timesheet or expense report. See Track Time and Expenses with the Project App.

• Move projects to Lakehouse v2 one at a time, from the Tools menu. Migrate Project to Lakehouse v2… moves a project’s tables along with everything built on them: queries, expressions, views, dashboard datasets, calculated columns, metrics and row access rules. The rest of your workspace stays on v1 while you work through it. Check Migration Readiness… reports first without changing anything, and names anything that can’t move or will change. Migration is one way, and anything written to the project during the move is lost, so stop scheduled workflows and keep people out of the project until it finishes. The original data is kept for seven days, so you can return a project to v1 within that window. One difference is by design: v2 starts weeks on Sunday rather than Monday, so weekly totals shift by a day. Check any weekly figure you rely on. See Migrate a Project to Lakehouse v2.

• Store embeddings in an ordinary column, and find the nearest rows with an ordinary workflow step. A Vector column holds the embeddings you already produce, in the same table as the data they describe. It loads and joins like any other column, so you don’t need a separate vector store. Set Vector Width on the column and every load is checked against it: a load with the wrong length is refused, naming the file, column and row (CSV imports are not checked). The AI: Vector Search step returns the rows nearest to one or more query vectors, by cosine or Euclidean distance. The search is exact, so filter the table to keep it fast. A search too large to finish is refused before it starts. PlaidCloud stores and searches vectors, and you generate them with the embedding model of your choice. Requires Lakehouse v2. See Vector Data Type and AI: Vector Search.

• Weave now watches how your workflows run, and every project has a Weave tab listing what it finds.

  • Runs. Weave compares each workflow’s latest run with the last 30 days: a failed run and how many have failed in a row, steps that took much longer than usual, output tables with far more or fewer rows than usual, and columns that came out empty much more or less often than usual. A workflow without enough history is reported as unchecked, not as healthy.
  • Scheduled events. Weave reports an event whose schedule can’t be read, one that came due without starting a run (and why, where it can tell), one that came due while its workflow was still running, and one that no longer matches what’s scheduled.
  • The Weave tab lists findings in plain language: problems first, then anything Weave couldn’t check and why. Mark findings read one at a time or all at once, and Go To opens the step or workflow a finding is about. A workflow’s Weave mark opens the tab on that workflow, and you can bookmark or share the tab’s link.
  • When your run ends, a message lists the problems Weave has found in that workflow that you haven’t read yet, with an Open Weave link.
  • Clear coverage. Each report says how many steps and workflows it checked out of the total, and never marks something as passed if it couldn’t check it. Weave is rule-based, with no AI model behind it, so running it costs nothing.

  See What Weave Checks, The Weave Tab, and Being Told When a Run Ends.

• Ask the AI assistant to build a workflow step, and approve it before it exists. Describe what you want: totals by region for a year, a filtered copy of a table, a lookup into a second table, a calculated column, a fixed value, or a column changed to a new type. The assistant proposes the step, showing the table it reads, every column it writes, any filter or grouping, the table it writes to, and whether it joins a workflow. Nothing is created until you reply, and what gets built is exactly what you were shown. For a lookup, the summary explains what the match does to your rows: every row carries through whether or not it matched, no row is ever duplicated, and if the second table holds a key more than once, it tells you whether that affects your data. Everything the step reads is checked first, so a missing column is caught before you approve. A calculated column or fixed value is stored as text unless you give it a type, and the summary names any column without one. See Building a Workflow Step.

• Connecting an AI assistant is guided, with ready-made tasks and a clear consent screen. Connect Your AI in the header has setup steps for Claude, ChatGPT, Grok, Claude Code, Cursor, Microsoft 365 Copilot and others, along with your connection URL. The Microsoft 365 Copilot tab walks your workspace admin, your Microsoft 365 admin and each user through their part. Copy request for admin sends the setup to whoever can approve it, and first-time visitors to Home see a short banner pointing to it. Assistants that support them offer task starters: import a CSV, explain a cost change, debug a failed workflow, or explore project data. Before you authorize an assistant, an Application Access Request page names the app that’s asking. Assistants already connected see it the next time they need to sign in again. See Getting Started with AI Coding Agents and Authentication.

• Version history backs up in the background, on every plan including Free. Saving a step, user-defined function, workflow, dimension, layout, document, data editor or project setting never waits for a backup. Once an object has gone 30 seconds without another save, one backup captures it, so a burst of edits makes one version instead of many. Back up now captures the current state right away. Restoring a version is refused in a locked project, like any other change there. Through the AI assistant, you can also restore and hold dimension versions, set up and write to data editors, and unlock a locked project, without overwriting a colleague’s edit made at the same moment. See Version History and Managing Data Editors.

• Getting started is quicker, and your progress follows you to any device.

  • Sign-up is one short flow that ends in your workspace. Confirm your email with Google, GitHub, Microsoft or a six-digit code, answer one optional question, and name your workspace. Free needs no credit card and builds immediately. Starter and Team go to checkout, with an optional seven-day trial. A live status screen shows progress, and you can invite up to ten teammates while it builds. See Sign Up.
  • A Getting started card on Home tracks your setup. It shows a progress meter, leads with your next step, and lists the rest: five steps for admins and three for builders. Viewers don’t see it. Hide tucks it into a toolbar button, and when you finish every step it retires with a one-time “You’re all set.” Your progress, tours, tips and anything you’ve dismissed are saved to your account, so they carry over to a new device or a cleared browser. See Quickstart.
  • The sample project is shared with everyone in the workspace as soon as it’s built, and a viewer in an empty workspace gets a welcome.

• Table Explorer profiles every column and flags data quality issues. The column details panel adds null and blank counts, the share of distinct values, and for numeric columns, skew, a histogram and correlations with up to ten other columns. A quality check covering outliers, format mismatches, duplicate keys, orphaned dimension members and constant columns shows as badges with a ranked summary. It respects row access. See The Details View.

• Import a named range or a cell range from Excel. The Import Excel step can read a named range or a cell range as well as a whole worksheet, and you can mix all three in one import. A range that’s out of bounds or covers separate areas is rejected, not imported wrong. See Import Excel.

• Workspaces upgrade themselves on their release channel, with a week’s notice. When a release reaches your channel (Rapid, Regular or Stable), your organization gets an email at least seven days ahead, and PlaidCloud upgrades during the maintenance window (Sunday 02:00 UTC by default). To stay on your current version, set the channel to No Release. See Automatic Release Updates.

• Give a project a default document account. Set it on the project home, and every file selector in the project offers it as Project Default. New file steps choose it automatically, and existing steps keep the account they were saved with. LLM steps and document conditions on execution groups use it too, and tell you plainly if the project has none. See Setting a Default Document Account.

• Limit a dashboard display control to the columns you choose. Each display control has an Available columns setting that narrows its Group By list to the columns you pick, without hiding them from charts and filters elsewhere on the dashboard. See Display Controls.

• Email templates can use variables and one-click formatting. The Notify Via Email template editor adds project and workflow variables from an Insert menu, and a Style menu applies formatting that displays correctly in email. An email sent without a subject is titled “PlaidCloud Notification.” See Send Templated HTML Email Notifications.

Changed

• Row Access is set up on one screen, takes effect as soon as you save, and holds everywhere a member reads. Open it from Table Explorer, a column’s menu, a project’s Security tab, or a member’s or group’s Row Access tab, and choose where the values come from, which values each group sees, and a default for everyone else. Saving applies to PlaidCloud right away and to dashboards within about a minute. See Managing Row Access.

  • One column name covers every table. A restricted column filters every table in the project that has it, in PlaidCloud and in dashboards, and publishing such a table sets up its dashboard filter at the same time.
  • A status you can trust. The sync status reads Waiting, Not applied yet, Applied or Error, then checks dashboards directly and reports In sync with dashboards or Dashboards differ. Retry sends the restrictions again, and a table that can’t be filtered is named, with what to do about it. See Keeping Dashboards in Sync.
  • Dimension grants follow the dimension. Grant a node and the group sees everything under it, including members added later. A member moved out stops being granted, without anyone reopening the grid. See Dimension Grants.
  • Only Architects change what a restricted project does. Workflows run as the project itself, which can read every row, so once a project restricts rows only its Architects and workspace admins can change its workflows, steps, functions, schedules, variables and notification templates, or the dimensions and tables that decide who sees what, or copy the project. Everyone else can still run workflows, and sees the editors read-only with a banner naming who to ask. Row Access tells you how many members this affects before you add the first restriction. See Who Can Change a Project That Restricts Rows.
  • Set it up once. A Set Row Access step ties a column’s grants to a source table. Restrictions can be saved as workspace-wide templates, grants can come from a group attribute in your identity provider, and Stop restricting removes a restriction in one step.
  • See what a member sees. A member’s Row Access tab shows the rows they actually see through all their groups, and Table Explorer shows Rows limited by your access when your own grants filtered the grid or a download.
  • Clear messages, and fewer blocks. When Row Access blocks an action, such as running SQL it can’t safely filter, or changing a restricted project’s automation, the message gives the reason and the fix and names the Architects to ask. A restricted member’s column filters in Table Explorer now show their own values instead of being unavailable, and an Architect can export their own SQL results in any format. See Row Access and Queries You Write Yourself.

• The AI assistant’s allocation answers say what they cover, and the figures add up. An answer says whether its figures cover the whole table or one member, warns before the figure if a period may still be loading, and names any part of your question it couldn’t apply. Its confidence rating is about the breakdown, not the cause. The figures reconcile: a stated change matches the totals beside it, and anything it couldn’t measure is labeled unmeasured. A “what if this input changed” question gets a written answer saying how much of the change it can account for, separating what it measured directly from what it measured partly or not at all. If you don’t name a field to break a change down by, the assistant finds the one that best explains it. Asking the same question twice gets the same answer, and the assistant never claims an independent check confirmed its answer when none was possible. On a project that restricts rows, tracing is available only to the project’s Architects, since anyone else would get figures mixing restricted and unrestricted data. See Analyze Allocations With an AI Agent, Tracing Allocations, and How the Assistant Stays Honest.

• Dimensions load and export much faster, and refuse what they can’t load correctly. A large hierarchy load takes seconds instead of minutes, and an export with alternate hierarchies is written in one pass. A flat load honours its default consolidation, and loading aliases, properties or values names any members it doesn’t recognize instead of skipping them silently. A parent-child load refuses an invalid default consolidation, or a target hierarchy with the same name as one of the loaded columns. Backing up or restoring all of a project’s dimensions as a single file is no longer offered, because large projects produced files too big to handle. Backing up and restoring a single dimension and version history are unchanged, a file you already hold still restores, and the AI assistant can still roll every dimension in a project back to its last backup. See Loading and Unloading Dimensions and Dimension Export.

• Allocation: Split using Dimension and Rule-Based Tagging give the same results however they run. Ticking Apply Rules using SQL, not Pandas used to change what these steps wrote, not just how they ran. Now a missing or blank value is always written as a null rather than an empty value, and the Mapping, Paths and Summary tables have the same columns either way. If a later filter or join looks for an empty value in these results, change it to look for a null. If a record can match more than one rule and the rules are split into iterations, the step now stops and asks you to tick Only match records to one rule or run each iteration as a separate step. A rule’s record count now includes records whose first column is empty. See Rule-Based Tagging, Matching Records to More Than One Rule, and Intermediate Tables.

• Builder-seat limits are enforced. Inviting, promoting or reactivating a Builder is blocked with a clear message once you reach your plan’s limit. Viewers stay free and unlimited on every plan. See Viewer vs. Builder.

• The LLM Provider connection form asks only what applies to your provider, and Anthropic Claude connections can be edited again. The Model list comes live from your provider, so new models appear without waiting for a PlaidCloud release, and the form gains Test Connection and the standard sharing and security settings. Editing an Anthropic Claude connection had been broken, which blocked rotating its key. See LLM Providers and Connections.

• A table’s usage lists include steps that reach it through a variable path, such as a path containing a period variable. A step whose variable can’t be worked out appears under May Reference This Table, so an empty Table Contributes To list now really means nothing uses the table. See Managing Tables and Views.

• PlaidCloud now needs Chrome or Edge 94, Firefox 93, or Safari 16.4 or later, and iPadOS or iOS 16.4 or later on iPad and iPhone (this excludes only the iPad Air 2 and iPad mini 4). An older browser shows an unsupported-version page. See Supported Browsers.

• Searching by name matches the words you type, whatever the punctuation or order. “invoice header detail” finds Invoice_Header_Detail, closest match first, in Home search and the AI assistant. Dashboard and app search and workflow step pickers still match exactly as typed.

• The pencil column in a workflow’s Steps table is hidden by default. The gear icon opens the step form, Edit Step Details on the right-click menu opens it on the General tab, and the column chooser brings the pencil back. A column layout you’ve saved keeps it if it showed it. See Where Are Workflows.

Fixed

• The AI assistant asks before it acts, and checks what it builds. See Weave AI Chat and Fixing Workflow Errors with AI.

  • Nothing is saved until you approve it. The assistant no longer imports a file, creates an import or export step, or saves a step correction before you approve it in a reply of your own. What it saves is exactly what you were shown, once, and it always shows you a draft to approve rather than going quiet. See Importing a File.
  • The fix assistant corrects the step you opened it on, where it could refuse as if you had no access to your own project, or write an empty step instead.
  • Steps it builds are real and keep their types. It can’t save a step type that doesn’t exist, or a setting the step doesn’t have, and suggests the closest match instead. A copied column keeps its table’s type unless you ask to convert it.
  • Every expression it suggests is checked first, against your project’s Lakehouse. An expression that would fail or give a wrong answer, such as a divide-by-zero guard written the wrong way round, is withheld and the reason given. PlaidCloud functions it used to reject are accepted. See Expression AI.
  • It knows when your view is filtered. It’s told when row access narrowed what it could see, so it no longer describes your view as the whole table. See Row Access.
  • It answers about real connections, offers only connection types that can actually be saved, and searches text containing a colon correctly. See Connect Your AI Agent.
  • It reports problems honestly. When the language model provider is busy or unavailable, it says so and suggests trying again in a few minutes, rather than reporting an internal error. A tool call that failed is shown as failed.
  • Replies are readable in the Light theme after switching from Dark without reloading.

• Document accounts and document steps work as they should.

  • Sub-folders are followed on every account type. On OneDrive, SharePoint, SFTP and Azure Blob, anything below the top level came back incomplete or empty. Every account now follows sub-folders to any depth.
  • Download as ZIP brings down everything in the folder, and a folder name no longer matches similarly named folders (Report no longer pulls in Report2024). Redo any affected download. See Download Folder Contents.
  • Copying and moving work on OneDrive and SFTP. SFTP’s Copy Folder had moved the folder instead, and SFTP file copies and moves reported errors even when they succeeded. If a recent SFTP folder copy seems to be missing from its source, check the destination, and if an SFTP file move reported a failure, check whether the file is now in both places. OneDrive copy, move and rename work again. See Move or Copy a Folder.
  • Delete Old Files reaches sub-folders, so check any you’ve scheduled. With Recursive on, it had missed sub-folders on several account types and could delete a folder itself. It now deletes only files, and may delete considerably more than before. Use Test first. See Delete Old Files.
  • Large folders list completely and quickly. Browsing a large Google Cloud Storage folder no longer slows down PlaidCloud, S3 and Wasabi folders with more than a thousand entries show them all, and an app’s file picker lists each sub-folder once.
  • Document steps produce what their settings say. Convert to PDF, Convert to JPEG, Merge Multiple PDFs, Compress PDF and Crop to Headshot produce their output again. Convert to PDF uses the page size and compression you choose, Fix File Extension renames files to the extension their contents need, the UTF-16 conversion writes UTF-16 and keeps accented letters and symbols intact, and Text Substitution handles large files. See Document Workflow Steps.

• Workflow steps no longer fail, hang or quietly do the wrong thing.

  • A calculated column in a file import with no Source no longer loads empty. Clear any Source you set as a workaround, and reload any table loaded while it was blank. See Advanced Data Mapper Usage.
  • A workflow can’t list the same step twice, which had left runs stuck in Running. An existing duplicate now fails the run with a clear message; delete it to fix the workflow. See Choosing a Workflow Type and Copy & Paste Steps.
  • Interrupted imports retry automatically, whether an Import SQL step was interrupted or the warehouse lost a machine partway through a load. PlaidCloud checks first that nothing is loaded twice, and a real error still fails right away. Allow about thirty extra minutes before deciding a retrying Import SQL step is stuck. See Import SQL.
  • Clear messages instead of server errors. Running a workflow that’s already running tells you so. A step whose source table was deleted or renamed names the missing table, and a column of an unrecognized type is named too. A step reading a decimal or floating-point column can be edited again. See Run a Workflow.
  • Loops and containers run. An iterating container in an Advanced workflow runs again, and Workflow Loop, Macro Concurrent and iterating containers run over tables with date, time or interval columns. See Iterating a Container, Workflow Loop and Macro Concurrent.
  • Conditions behave. Workflow: Conditional Run fills in the date, project and model variables like any other step, and a condition that doesn’t say what to check is skipped with a reason instead of failing. See Conditional Step Execution.
  • Import External Database Tables lands common column types correctly, including exact decimals, money, times, binary and JSON. A value that won’t fit its column now stops the import and names the column, instead of being silently truncated or rounded. See Import External Database Tables.
  • Settings are checked when you save. An Import: CSV step whose file yields no columns can’t be saved and tells you why (Import CSV), cancelling the unusual-quote-character prompt simply cancels the save, and ML: Train Model refuses an unknown algorithm or setting.
  • Set Project Variable and Set Workflow Variable show every variable’s current value and memo. See Set Project Variable and Set Workflow Variable.
  • SAP Post refuses a field mapping set on only one side (the header but not the lines, or the reverse), which had posted lines under the wrong header.
  • Output columns keep the memo, mask and dimension you set, where the next run could erase them.
  • Smaller fixes. Empty retry, resource or row-limit settings use the defaults. A REST import body containing its own template markers is sent as written. An Agent that picks up a step late no longer runs work already reported as failed. On StarRocks, columns named with reserved words such as text work, and on Snowflake a % in a column name is kept.

• A REST Request step keeps its method when you save it, so check any you’ve saved recently. A step saved in the earlier editor with a method other than GET could open showing GET, and saving it, even unchanged, made it send GET from then on. The editor has its Target Table picker back, Row format and Pagination mode are lists again, and choosing a catalog endpoint fills in its headers, parameters and body (asking before it replaces your edits). Inspect Source → Populate fills the column mapping from a live response. A single-request step saved with a blank row format had run as Array; if your API returns each record as a list, choose Array before saving. See REST Request Step.

• A project archive keeps its table data, both when it’s taken and when it’s restored. Archives taken on most warehouses had left out table data while reporting success, restores on the Lakehouse had the same problem, and projects using row access couldn’t archive table data at all. If you have a full archive taken before this release, or restored one and found it empty, do it again. The closing message now says how many tables went in, or came back, without data. See Archive a Project.

• Table snapshots and keys work on every Lakehouse. Restoring a snapshot in place, and holding one, now work on Databend as well as StarRocks. Restoring to a different table creates a new table beside the current one, and a held snapshot keeps everything written after it until you release the hold, so storage grows until then. Replacing a table outright, with an upsert or an SCD-2 replace, starts a new snapshot history and drops any hold. On Databend, an in-place restore is refused if the table’s columns have changed since the snapshot, and suggests restoring to a new table instead. Partition, distribution and update keys can be set on a table, and partition keys are kept when a table is created or comes from a bundle. See Table Snapshots.

• Dimension fixes. See Loading and Unloading Dimensions.

  • Ancestry-format exports no longer repeat rows, which had made large dimensions run out of memory. See Dimension Export.
  • Filtering finds members by property values again, the Advanced filter offers properties, and filters survive a tree rebuild. See Finding Nodes in a Hierarchy and Advanced Filters.
  • Referenced in Tables shows every table that uses a dimension, and setting a column’s dimension refuses one that doesn’t exist.
  • Rolling every dimension in a project back to its last backup tries every one, and names any it couldn’t restore.
  • Working with dimensions through the API is reliable. Reading a member’s last child or its ancestors works again, very deep dimensions can be traversed, properties and values round-trip correctly, and alias loads take effect immediately.

• The SCD-2 Master Builder and SCD-2 As-Of Join steps run on real tables. Both had reported every column missing and couldn’t run. A table with no data yet now says so, and the Master Builder checks that typed-in dates are valid and in the right order before saving. See SCD-2 Master Builder and SCD-2 As-Of Join.

• Scheduled events run, and are shown, at the right time. Events set to UTC -12, UTC +2 or UTC +6 now run at exactly that offset all year, and existing ones move to the correct time automatically. Next Run and the upcoming-runs calendar no longer show times an hour out under daylight saving. See Event Scheduler and Upcoming Runs Calendar.

• Inviting and managing members works more smoothly. In a workspace that uses single sign-on, invites no longer send a set-your-password email. The welcome email’s Set your password link opens the right page, and its buttons display properly in Outlook. The Identity tab appears only for people who administer identity. Member, group and distribution-list forms show progress, prevent double submission, confirm deletes, and say plainly when a member no longer exists. Deleting or updating a member no longer stalls when dashboards are briefly unavailable. See New Member Welcome Email and Member Management.

• A project whose name contains an & can be created again. If creating one failed, delete any half-created entry before trying again. See Creating New Projects.

• Dash apps open for signed-in users again, where some users saw “Request Header Fields Too Large.” Rebuild any app you’ve already published. See Rebuilding a Dash App.

• Paycor connections can be authorized again. Save the connection, then click Connect to Paycor. See Paycor REST Connector.

• Smaller fixes around the app. Pages no longer occasionally time out while PlaidCloud checks for a newer version. The Labels box has its Add button back and keeps labels typed on a new step (Organizing Work with Labels). A column added to a grid since you saved its layout, such as the step number in the Steps grid, appears in its place instead of off the right-hand edge (Change the Order of Steps). The first-win celebration card no longer covers the header and its confetti plays through, and Home’s toolbar is shaded evenly.

Security

• Row access now holds everywhere a restricted member reads. See Managing Row Access and Row Counts and Table Sizes.

  • Downloads and dashboards. Table Explorer downloads return only the member’s rows. Every published table with a restricted column is filtered in dashboards, not only the tables picked when the restriction was set up, and a group granted some values sees only those. Adding or removing a member, or deleting a group, reaches dashboards within about a minute.
  • Counts and sizes. Row counts and storage sizes for tables and projects, and the rows read and written in workflow run history, are hidden from members whose access is filtered, since they revealed how many rows exist beyond what they may see. Project Architects see them as before.
  • Dashboard tools that bypass row access. The dashboard roles every member holds no longer include SQL Lab or dataset editing. Grant these through a separate role only to people who may see every row.
  • Project access. Only a project’s Architects and workspace admins can change its access lists and data shares. Everyone else sees them read-only. See Managing Project Access.
  • AI tools. When an AI tool makes several calls in a row, each one is now checked on its own.

• AI assistants can no longer change who may post to an ERP. Adding or removing a connection’s owners or users, and changing a project’s posting mode, are now for people only, as project access and credentials already were. A Read-only AI connection can no longer list a project’s member access or its app deployments. See Agent Access.

• Connection credentials are never shown again once saved, and saving a connection never erases one by accident. Every credential field shows blank when you reopen it and keeps its stored value if you leave it blank. Twelve credentials that had been shown in plain text, including several tokens, passwords and secrets, are now masked. See Create and Manage a Connection.

• PlaidCloud’s server components are updated to fix the critical and high-severity vulnerabilities found by our security scanning. Affected third-party libraries are on their fixed releases, and an unused bundled tool has been removed.