Integrate.io Release Notes & Product Updates

September's release cycle put more of the platform's output where teams already work, from partner-ready files and spreadsheets to the observability tools they run. Easier downstream edits in the package designer and deeper marketing and CRM connectors filled out the month.

Downstream Field Propagation Keeps Every Component in Step

Renaming a field early in a package used to mean opening every component after it to find what broke, often after a job had already failed. Downstream Field Propagation tells you which components a field change leaves out of step and brings the compatible ones back in line in one click. Changes that alter meaning, such as a sort key or a join field, come to you for review instead of being rewritten silently. A refactor that used to cost an afternoon of clicking becomes a review of a short list.

Log Drains Stream Job Logs Into the Observability Stack You Already Run

Job logs have lived inside the platform, so lining up a failed pipeline with the rest of your systems meant copying errors out of the dashboard by hand or retrieving them through an API endpoint. Log Drains send job success and failure logs to any OpenTelemetry-compatible service (e.g. New Relic, Datadog, Grafana, etc), for selected packages, whole workspaces, or the entire account. Pipeline failures then alert, correlate, and get retained under the same rules as the rest of your infrastructure instead of in a separate tab someone has to remember to check. Log Drains are available on request, our team can turn it on for your account. The documentation has the full picture on supported endpoints.

XML File Destinations Write Partner-Ready Documents Straight From a Pipeline

Partners and older systems that only accept XML used to force a conversion script between a pipeline's output and the file drop. File Storage destinations on Amazon S3, SFTP, and FTPS write well-formed XML directly, including nested sections, element names drawn from your data, and repeated groups. The document a partner's spec describes comes out of the job itself, so there is no reformatting step to maintain or to break when the spec changes. See the documentation for the options that shape each document.

New Connectors

• Salesforce Data Cloud Destination: Pipelines load records into Data Cloud objects through customer-managed Salesforce connections, with upsert and delete support and rejected rows routed to Problem Records. Pipeline data reaches Data Cloud segmentation and activation without a custom Ingestion API integration to build and maintain (documentation).
• OneTouch Health Source: OneTouch Health brings client, carer, invoice, location, status, and tag data into a pipeline, paging through full result sets. Username and password connections renew access on their own, so long-running jobs finish instead of failing when a token expires (documentation).

New Features

• Excel Workbooks on the SharePoint Destination: The SharePoint destination writes .xlsx workbooks from a template, and several destinations in one package can fill sheets of the same workbook. Formatted reports come straight from the job instead of being manually pasted into last month's file (documentation). This functionality will be live on all file storage destination connectors in the coming weeks.
• NetSuite RESTlets: The NetSuite REST source reads from RESTlet scripts, addressed by script and deployment ID or by full URL. Custom business objects your NetSuite team exposes through a script become a pipeline source instead of a separate export.

Improvements

• Job errors name a real cause and points to a failed component, instead of an AI-written summary.
• Azure Blob Storage connections can authenticate with a SAS token, so you can grant access scoped to the containers and time window a pipeline needs.
• The HubSpot source adds Campaigns and Users as objects, and existing OAuth connections need reauthorizing before the new objects appear.
• The Ometria destination maps every standard field Ometria documents for Contact, Order, and Custom Event, including @force_optin, @collection, custom properties, and SMS consent.
• The Ometria destination can write an optional CSV of accepted records to cloud storage, so you have a record of what Ometria took in.
• The Trustpilot destination sends product review invitations alongside service reviews, with productReviewTemplateId, productReviewProducts, and the other product review fields available in mapping.
• File Mover delivers files to Google Drive and Shared Drives, including over service-account connections, so output lands where business teams already work.
• SharePoint source and destination components pick up an authorized connection when the job runs, so cloned packages move between accounts and environments without reconnecting.
• The Salesforce source offers Source Action selection in Query mode, so Bulk query all, archived records included, runs on existing SOQL without rebuilding the field list in Object mode.
• Long-running Salesforce jobs, previews, schema detection, and Bulk upserts recover on their own when the org's session expires mid-run.
• NetSuite source JDBC connections run on version 8.10.190.0 of the NetSuite JDBC driver, up from 8.10.184.0.
• The expression and variables editors accept functions in any capitalization, so tOdATE(test) is corrected to ToDate(test) automatically.
• Users added to an account through Single Sign-On for the first time join with the read-only Reader role instead of Operator, so assign a higher role where one is needed.
• New accounts start with fewer automatic retries and a 24-hour execution timeout by default

Fixes

• Salesforce connections save and reconnect without an "At least one attribute is not allowed" error.
• Opening or editing a Join no longer changes the input field names that downstream selection, validation, and previews rely on.
• SugarCRM paginated pulls deliver each record exactly once, with no duplicates or gaps.
• File sources with underscore-prefixed column headers work with Detect header row automatically, and re-detecting existing schemas picks up the real names.
• Detect header row automatically handles files with colliding column names.
• Packages built through MCP inside a package group with connection restrictions save without an error.
• Packages built in an MCP chat open in the dashboard with their components laid out correctly.
• Salesforce SOAP destination jobs that write date fields complete instead of failing.
• SFTP and cloud storage destinations set to copy files to an existing directory overwrite on re-runs instead of failing or leaving partial temporary files.
• Redshift destinations with Automatically add missing columns enabled handle existing mixed-case columns instead of failing.
• Job failure summaries keep the log lines that identify the cause, including deep stack frames and long line tails.

Tell us which partner file specs and observability tools you want covered next, and where field changes still catch you out.

September's release cycle went into keeping data flowing when part of a load goes wrong, and into clean recovery when replication sources are interrupted. Ad platform coverage and initial-sync data fidelity took the rest of the month.

The Snowflake Destination Keeps Healthy Tables Loading When One Fails

One table's failed load into Snowflake used to stop the whole pipeline, so a single bad table held up every healthy one behind it. With a per-pipeline setting, the Snowflake destination keeps the pipeline running while that table is set aside, and a Destination Load Failed notification names the table that needs attention. Failed tables retry on their own, so a transient problem clears without anyone stepping in. The rest of your warehouse stays current while the one problem gets fixed, instead of the whole sync waiting on it.

The OneTouch Health Source Puts Care Operations Data in the Warehouse

Home care teams running on OneTouch Health have had client, carer, and scheduling data locked inside the care management system, reachable only through exports. The OneTouch Health source replicates operational, care, client, scheduling, financial, and residential data into your warehouse. Scheduling, billing, and staffing questions get answered alongside payroll and finance data instead of in a spreadsheet stitched together each month. Operations and finance teams work from the same numbers, refreshed on every sync.

New Features

• Facebook Ads Lookback Windows: Facebook Ads pipelines can re-read the last 1, 7, 14, or 28 days of Insights on every sync. Conversions Meta attributes after the original sync land in the warehouse instead of leaving spend and return figures short.
• Source Change Time for DB2 CDC: DB2 CDC pipelines can carry an _iio_changed_at column holding each row's change time at the source. Ordering and freshness checks use when a change happened instead of when it loaded, and our team enables it on request.

Improvements

• Facebook Pages connect with page-level access, so a connection no longer needs business portfolio administrator permissions.
• The Facebook connector's schemas follow Meta's new API version, so streams keep loading as Meta retires older fields.
• Facebook Ads adds daily_budget, lifetime_budget, and budget_remaining on campaigns, and bid_amount and bid_strategy on ad sets.
• Snapchat discovers every organization and ad account a connection can reach and syncs them all, instead of one organization at a time.
• Initial syncs of pipelines with many tables allow schema setup time in proportion to the table count, so wide pipelines start instead of timing out.
• Snowflake connection validation checks only the configured warehouse, database, and schema, so roles with many grants validate instead of timing out.
• Redshift and Snowflake destinations never run two load cycles against the same table at once, so catch-up after a delay finishes without duplicate work.

Fixes

• MySQL change tracking resumes after a source restart instead of stalling silently while reporting as healthy.
• DB2 sources write _iio_start without a container time zone shift, and an unreadable source change time falls back instead of failing the row.
• Oracle syncs of large tables keep their connection alive through long idle stretches instead of failing on a dropped connection.
• Oracle unconstrained NUMBER columns keep fractional values as decimals instead of truncating them.
• Oracle CDC reads text values that look like numbers instead of failing to parse them.
• BigQuery pipelines treat a transient error as transient instead of failing with a permissions error.
• PostgreSQL ARRAY columns load into Snowflake as text, and existing affected tables need a resync to pick up the corrected type.
• MySQL JSON columns load into Redshift during initial syncs that use parallel chunk workers.
• Snowflake initial syncs complete on tables holding zero dates or text longer than 16 MB.

Performance Enhancements

• Oracle initial syncs of large tables without a numeric primary key chunk on a suitable indexed column, so the first load runs in parallel instead of as one stream.
• BigQuery destinations share one schema-propagation wait across tables, so a column change across many tables costs one 5-minute wait instead of 5 minutes per table.

Tell us how table-level failure handling holds up on your busiest Snowflake pipelines, and which ad platforms you want covered more deeply next.

August's release cycle focused on regulated file workflows, encryption at the storage boundary, and improved job failure debugging through component attribution. Authentication enhancements, connector coverage, and a broad range of other updates took up the rest of the month.

Component Error Names the Step That Failed

A failed job used to hand you a stack trace and leave you to work out which of a dozen components produced it. Job errors now name the component that failed and explain it in language you can act on, so triage starts at the cause instead of a search through the log. A stable error code makes the failure recognizable across runs, so a one-off debugging session becomes something that can be answered in a sentence.

User-Defined Schemas: Centralize the Layouts You Map To

Describing a destination layout used to mean rebuilding it field by field in every package that wrote to it, with no shared definition to point at. User-defined Schemas make that layout a first-class object on the account, owned and versioned in one place, so a change to the destination is one edit rather than a sweep through every package that feeds it. Smart Mapping then has something authoritative to map to: field matching, renames, and type conversions run against the layout you published rather than one inferred from whatever sits upstream. Our team can enable schema management for your account upon request.

PGP Encryption Closes the Gap at the Storage Boundary

Exchanging encrypted files with a partner meant a decryption or encryption step outside the platform, so plaintext sat on intermediate storage and a private key lived on somebody's laptop. File storage sources now read PGP-encrypted files directly and destinations encrypt output before it lands, so the data is only ever in the clear inside a running job. That closes the gap auditors ask about on regulated exchanges, and it removes the local scripting step that used to sit between a partner drop and a pipeline. Check out the documentation for the file storage source and the file storage destination for key selection on each side.

The EDI (X12) Source Puts Enrollment Data Where the Rest of Your Data Lives

Benefit enrollment has lived outside the data stack, locked in a format that only a dedicated EDI translator or a clearinghouse subscription could read. The EDI (X12) source brings 834 enrollment into the platform as ordinary rows, so member and coverage data lands in the same warehouse, reports, and pipelines as everything else. Eligibility reconciliation against HR and payroll, coverage audits, and downstream loads stop being a separate manual process that lives in a spreadsheet. For a health plan or a benefits administrator, enrollment becomes something the data team can own rather than something routed around them. The documentation has the full picture on what the source parses.

New Connectors

• Ometria Source and Destination: A native Ometria source imports profiles, orders, products, and lists from the Data API, and a matching destination pushes contacts, orders, and custom events back with batching, throttling, and merge control.
  (source and destination documentation).
• Trustpilot Destination: July's Trustpilot source covered reading reviews, and the destination adds the write side: service-review invitations sent from an ETL job with templates, field mappings, rate controls, sender validation, and an audit file of what went out. Trustpilot works in both directions on the same platform instead of a read-only feed (documentation).

New Features

• Ask Helm Auto Mode: Ask, Build, and Auto sit on one Ask Helm surface, and Auto routes a natural-language request to a documentation answer or a package-building action. Picking the right mode before you type is no longer part of the job (documentation).
• Fixed-Width File Destinations: File storage destinations write positional records with per-field widths, alignment, padding, overflow handling, and numeric formatting, and a preview renders the layout against real data. Fixed-width exports for banks and clearinghouses stop needing a downstream reformatting script (documentation).
• Benefits Hub: Trading partner management arrived in Account Settings with delivery settings, profile versions, effective dates, inheritance, and filename previews. Partner delivery rules live on the account rather than inside every package that serves them, and our team can turn the hub on for your account.
• Snowflake Database Overrides: Snowflake source and destination components each take an optional database value, and the connection form accepts an empty database. One connection covers every database on the account instead of one connection per database.
• Schedule Skipped Hook Event: A Schedule Skipped hook event fires when a scheduled run is passed over because the previous run is still active, or after repeated cluster failures. Skipped runs surface as a notification instead of a gap in the job history.
• NormalizeState Function: NormalizeState converts a full US state or territory name, the District of Columbia, or an already-valid two-letter code into its USPS abbreviation, including values carrying pasted whitespace. Address normalization fits in one expression instead of a nested conditional (documentation).

Improvements

• Salesforce customer-managed OAuth connections coordinate refresh-token rotation, so schema discovery, tests, previews, and jobs share short-lived access tokens instead of competing for a single-use token.
• New Salesforce OAuth connections use the configured proxy server.
• The Universal OAuth connector tests Custom Auth connections against a custom token URL and detects expired tokens correctly.
• Excel sources select a worksheet by tab position rather than by name, and previews, schema detection, and job runs all read the same sheet.
• Cloud storage and file upload sources accept an optional workbook password, so previews and scheduled jobs read protected .xlsx and .xlsm files.
• The HubSpot source adds Calls, Emails, Meetings, Notes, and Tasks as objects, and Contacts incremental loads sync on lastmodifieddate.
• The GA4 source adds Session, First user, and Event channel-group dimensions to the field picker.
• The Xero connector requests the accounting.budgets.read scope, so budget data comes through on new connections.
• The Oracle source resolves DB links in Table mode, so a table on a remote database is addressable with the @ character.
• The Oracle destination writes non-key text columns as NVARCHAR2 rather than NCLOB when the account sets _ORACLE_STRING_DATATYPE to NVARCHAR2.
• File Storage and File Upload sources append a literal string, a job variable, or both to filenames when archiving processed files, and keep the original if renaming fails.
• File destinations gained a third step for reordering, removing, and adding fields, with a five-row preview of the generated file.
• The REST API destination previews real payloads and headers from component previewer data.
• The Select component shows a warning when a mapped field is no longer available because a parent component changed.
• A failed job on the write-files-directly and replace-files-in-directory destination actions leaves the target directory in place instead of deleting it.
• Smart Transform runs by default on every account.
• Changing a BigQuery V2 connection's region or service account restores temporary-file access automatically, and administrators get tools to review and repair affected connections.

Fixes

• Package groups delete correctly when a custom role uses the group as a restriction, and permission scopes cannot widen when their last workspace is removed.
• The Join component passes the correct input fields to the Select component.
• Join type and optimization changes appear in the component previewer without a save first.
• The Universal OAuth connector's Test Connection button works for the Custom Auth type.
• Universal OAuth tokens refresh correctly on long-running jobs and ahead of schema-importer calls.
• The Salesforce Own Auth connector stays authorised instead of transitioning to an unauthorised state.
• Job variables such as _WORKFLOW_PACKAGE_LAST_SUCCESSFUL_JOB_SUBMISSION_TIMESTAMP carry their intended value rather than an incorrect timestamp.
• The workflow job details Variables tab shows a plain value rather than a DeserializeFromJson wrapper.
• The Aggregate and Cube components save after an upstream field is removed, and unused calculations clear automatically.
• The AI Transformation component works for users holding Editor and Operator roles.
• The REST API source drops other auth parameters when following an HTTP redirect, as do other components that make outbound API calls.
• File source components apply Detect header row automatically during job runs, not only in the schema importer and previewer.
• File Mover with a wildcard source and delete-from-source leaves files in subfolders it was never pointed at.
• Database source and destination, REST API source, and SharePoint source components report missing connections or record settings as a validation error, not a crash.
• The Region field in the Admin dropdown and in Account Settings shows regions outside the default list.
• The HubSpot source syncs objects carrying large numbers of custom fields, and the full-load previewer returns actual records.
• The SugarCRM source refreshes access tokens that expire in the middle of a job run.
• Ask Helm stays visible beside component modals and the expression editor rather than sitting underneath them.
• Reopening a Helm conversation restores its history without adding a blank assistant bubble.
• The API Transformation component's JSONPath expression input evaluates correctly, and response fields no longer collide with same-named fields from the parent component.
• The Slack webhook form works correctly when the browser is in fullscreen mode.
• The schema importer keeps read-only aliases aligned without stretching row height.

Performance Enhancements

• Worksheet selection reads an Excel workbook once rather than repeatedly, and large shared-string tables can be sampled for schema inference without exhausting memory.
• Error summaries load without transferring the full job log, and REST API and fixed-width previews run on real sample data.

Tell us which file formats and partner exchanges you need next, and where the new error pages still leave you guessing.

August's release cycle went into Oracle replication depth and keeping destination schemas current as sources change. Security hardening and large-volume load reliability took the rest of the month.

Automatic Column Detection Keeps Destination Schemas Current

A column added at the source stayed invisible until someone noticed a report was missing it, worked out which pipeline was responsible, and edited the schema by hand. Singer-based connectors detect new source fields and carry them into the destination on the next sync, so the warehouse tracks the source instead of drifting from it. The gap between a change upstream and a change in your tables becomes one sync interval instead of however long it takes someone to notice. Past rows keep an empty value for a new column, so a backfill still needs an explicit resync.

Parallel Chunk Workers and History Mode Deepen Oracle Replication

A large Oracle initial sync ran as one extraction stream, and a record's earlier state vanished the moment it changed. Splitting extraction across parallel chunk workers shortens a large first load, and a failed setup releases its database sessions instead of leaving them for a DBA to clean up. History Mode keeps prior versions of a row with the source's change times, so a point-in-time question about what a record looked like last quarter has an answer in the warehouse instead of in a backup. Replication also drops rolled-back changes and replays partially acknowledged ones after an interruption, so updates and deletes stop going missing quietly.

New Features

• Configurable Salesforce Lookback: Each Salesforce pipeline can re-check recent changes on a window from thirty minutes up to twenty-eight days. Records that land after their original sync come through on the re-check instead of going missing, and the setting stays off until you enable it.
• Initial Sync Only Override: A pipeline can complete its initial snapshot, load the staged data into the warehouse, and stop before continuous replication starts, configured per pipeline by our team. One-time migrations finish without a CDC stream you then have to turn off.
• Validation Overrides: Pipeline creation and edits can proceed without validating the source and destination, and our team can apply an account-specific override for an overly strict Snowflake check. Bypassed checks stay visible and auditable, while connection, credential, and provisioning failures still block deployment.

Improvements

• TikTok Ads syncs ingest reservation campaigns, including TopView and Reach & Frequency buys, alongside the auction campaigns already covered.
• TikTok Ads tracks report progress separately for each advertiser, so a stalled advertiser stops hiding missing data elsewhere.
• TikTok Ads fetches partially complete recent report windows after a resume or a restart, so the newest data is not skipped.
• TikTok Ads handles more error responses from the provider rather than failing the sync outright.
• Oracle destinations support SCD Type 2, so a dimension table keeps its change history instead of overwriting a row in place.
• Singer source validation messages carry the underlying provider error, such as a missing permission, instead of a generic credentials message.
• The Snowflake connector runs on a newer Snowflake SDK driver version that carries the provider's latest fixes.
• Resumed and retried initial syncs report steadily advancing progress, and BigQuery receives completion notifications during large initial loads.
• Chunk workers wait until the schema record is fully applied before they start, so chunked uploads complete more consistently.
• Large pipelines wait for destination preparation to finish, and the platform validates timing and retry settings before use.
• DB2 setup queries double-quote database, schema, and table names consistently on the generated setup SQL.
• Alert setup consistently rejects malformed, unauthorized, or excessive notification-hook references when an alert is created or edited.

Fixes

• Snowflake operations quote reserved-word columns and names containing double quotes without a pipeline change.
• Chunked COPY statements on Snowflake run in sequence, and chunk stream keys resolve before they leave the copy handler.
• The TikTok Ads pixel_instant_page_events stream completes on parameter combinations the provider does not support.
• DB2 CDC setup SQL runs on the DB2 CLP, and it handles SAP-style identifiers, unusual punctuation, and repeated execution.
• Initial-sync progress reports current and total in the right order, and null fields return a result rather than a server error.
• Auto restart completes without firing a false error event hook.
• Oracle pipelines handle a duplicate record on an index column instead of failing.
• User invitations stay scoped to the correct organization, and moving an existing account requires the invited user to accept.
• Observability alert validation on create and edit checks object ownership.

Performance Enhancements

• Snowflake imports of more than 1,000 files complete instead of failing partway, so a high-volume load no longer needs splitting into smaller batches.
• TikTok Ads backfills use thirty-day windows and larger pages, so a historical load needs fewer requests and still fetches the latest partial range.

Tell us how the Oracle changes hold up on your largest tables, and which sources should get column detection next.

July's release cycle put AI assistance at the center of the package designer, covering pipeline building, field mapping, and expression writing. Alongside that, the editor gained safer handling of in-progress work, workflow orchestration went a level deeper, and destination coverage expanded.

Helm: Prompt to Pipeline

Helm is your copilot in Integrate.io. It builds pipelines, answers questions, and manages your packages.

• Build a pipeline from a prompt, or start from a ready-to-run package and adapt it
• Answer questions about how a component behaves, with a link to the source
• Manage packages through edits that land in a draft, not the live version

Helm designs and builds; the Integrate.io platform executes. A pipeline that ships stays deterministic and runs the same way every time. Nothing becomes production until you review and approve.

Helm Ask is on for every account, Helm Build is enabled per account by our team, and the documentation has the full picture.

Smart Map and Smart Transform Take the Slowest Step Out of Pipeline Delivery

Smart Map removes field-by-field setup. Point the platform at your destination schema and it automaps your source data to the matched schema output, including renames and type conversions. Smart Transform works at the field level: describe the transformation in a prompt or give a before and after example, and it generates the Integrate.io transformation expression for you. Suggestions are validated against your own data before you accept them, so what you review is what runs. Both are available on request, and the documentation covers what they can do.

Package Drafts and Autosave Keep In-Progress Edits Safe

Losing package edits to a closed tab or a dropped connection used to be a real risk, since nothing was stored until you explicitly saved a version. The designer now autosaves your work to a private per-user draft after you stop editing, and again when you leave the editor. Reopening a package that has a draft asks whether to resume the draft or open the last saved version, and a Discard Draft button in the toolbar throws the draft away when you want to start clean. Draft autosaves stay out of version history, so only explicit saves create a version. Full setup details are in the documentation.

The Run Workflow Task Runs Workflows Inside Workflows

Reusing one workflow as a step inside another previously meant duplicating its tasks by hand and keeping both copies in sync. The Run Workflow task executes another workflow as a sub-run of the current one, so a long orchestration can be split into smaller workflows that also run on their own. Read the documentation for the nesting limit and variable scoping rules.

New Connectors

• Salesforce REST API Destination: A new Salesforce destination built on the REST API supports insert, upsert, update, and delete across throttled single-record and batched write modes, and produces per-record success and failure CSV outputs. It works with both standard and customer-managed OAuth Salesforce connections, giving teams an alternative to the Bulk API for smaller or latency-sensitive writes (documentation).
• Databricks Destination: Pipelines can now write to Databricks, with staging and load handled for you so output lands in your tables without a manual copy step. Teams running on the lakehouse can send ETL output to Databricks directly instead of routing it through an intermediate warehouse first (documentation).
• Trustpilot Source: A native Trustpilot source pulls reviews, review categories, and private reviews for a business unit, with incremental load supported. Review data can now be joined against the rest of a customer dataset without maintaining a custom API pull (documentation).

New Features

• Live Preview in the Expression Editor: The expression editor now runs the current expression against five real sample rows from the previous component and shows each input value next to its result. Expressions can be checked while you write them instead of after a job fails.
• Field Data Types in the Select Component: The Select component's field list now shows a data type for every output field, and the expression editor shows the return type for most functions, fields, and variables. Where a type cannot be inferred statically, running a preview fills the column in.
• Separate Read and Write Access for AI Tools: Helm and MCP capabilities are now split into distinct read and write tool sets, so inspection, previews, and searches can be opened up broadly while package edits, schedules, and job runs stay restricted. Write access is closed by default, and our team can open it up for an account on request.
• Package Group to Connection Group Binding: A package group can now be bound to specific connection groups, restricting its packages to connections from those groups. Admins get a direct way to keep one team's pipelines from reaching another team's credentials.
• AddToDate and ToDateOrNull Functions: Two new expression editor functions cover date arithmetic and parsing that returns null rather than failing on an unparseable value. Date handling that previously needed nested conditionals now fits in a single call.

Improvements

• SFTP connections now support private key plus password authentication, covering servers that require a multi-factor login.
• Job logs, the logs API, and AI error summaries mask connection credentials, and account-level global secrets are masked in job and workflow variable views.
• Chatmeter, Corrigo, and GoTo sources now paginate through results, so large accounts return their full dataset instead of a first page.
• The Email Report destination accepts variables in its subject and body.
• HubSpot destination has a new Clear when empty flag that clears a HubSpot property when the incoming value is null.
• File and cloud storage destinations can skip writing an output file when a run returns zero rows, an opt-in setting for CSV and JSON.
• The Google Sheets source imports the full sheet schema without manual selection, and can fail a job when the sheet contains no data rows instead of finishing with nothing loaded.
• The Google Ads source adds ConversionsByConversionDate and ConversionsValueByConversionDate to its available metric fields.
• The REST API destination supports a sleep interval between batched requests, giving rate-limited endpoints room between calls.
• Custom role management adds a filter on the roles list and a form for assigning users to a role directly, and the packages and connections lists gained a Show ungrouped only switch.

Fixes

• Package variables are no longer removed when a variables save and a package save run at the same time.
• Package validation succeeds when ExecuteSql or ExecuteSqlDateTime is used in a package variable.
• Draft variables are now executed during schema preview calls.
• The Select component opens correctly when an integer is used as a package variable.
• Source fields with commas in their names, such as long Google Sheets headers, now carry their values into downstream components instead of previewing as null.
• File Storage jobs run correctly when a filename contains parentheses or the source path contains spaces.
• Packages whose variable substitutions produce long file paths no longer crash mid-run.
• SFTP and FTPS sources using a bare wildcard resolve the same file in preview and at job runtime.
• The DB Lookup component's + New button now opens the connection form.
• The Salesforce destination writes error and success record files to shared Google Drive directories when a directory ID is specified.
• The AI error summary reports Salesforce write failures correctly instead of describing them as filesystem handler errors.
• Packages inside a package group with connection group restrictions now save correctly.
• Global secret variables evaluate correctly in the cases that previously failed.
• Invitations work for users assigned a custom role, and the user management page is restricted to administrators.
• The owner role shows its label and filter on the user list and is assigned correctly on the roles view.

Performance Enhancements

• The Google Ads source can split a long date range into parallel chunks, cutting pull times on wide reporting windows. Set _ADWORDS_API_DATE_SPLIT_ENABLED to true in your package variables to switch it on, with _ADWORDS_API_DATE_SPLIT_CHUNK_DAYS and _ADWORDS_API_DATE_SPLIT_MAX_CHUNKS controlling the split.
• The Google Sheets source reads large sheets in batches rather than loading a whole sheet in a single pass, clearing the out-of-memory, read-timeout, and missing-row failures that wide or long sheets used to hit.

Give the new AI tooling a run this month and tell us where it saves you time and where it still needs work.