November 2026

New connectors

lemlist

lemlist is now available as a Weld ELT source for your outreach data.

  • Connect with an API key generated in lemlist under Settings > Integrations > API. The key is shown only once, so copy it when you create it.
  • activity syncs incrementally from the sync's start date; every other table is fully re-read on each sync.
  • campaign, sequence, and lead are read per campaign, so each campaign's schedules, steps, and leads come through.
  • sender and webhook support soft deletes, so removed records stay in your warehouse with _weld_deleted_at set.
  • Weld stays within lemlist's rate limit of 20 requests per 2 seconds and waits when lemlist asks it to slow down.

11 total changes

Table/columnChange typeNotes
activityNew table
campaignNew table
companyNew table
contactNew table
leadNew table
scheduleNew table
senderNew table
sequenceNew table
teamNew table
unsubscribeNew table
webhookNew table

Visual Crossing

Visual Crossing is now available as a Weld ELT source for weather data by latitude and longitude.

  • Add each location in the sync settings with its coordinates and an optional ID; the ID becomes location_id and, if left blank, Weld generates one and keeps it fixed even if you edit the coordinates later.
  • history backfills daily and hourly weather from the sync's start date (one year by default) up to yesterday, then re-reads the last 2 days on every run. Each location tracks its own progress, so a location added later backfills on its own.
  • forecast stores the 15-day forecast keyed on the date it was issued, so earlier forecasts are kept instead of overwritten.
  • One API call per location per sync covers both daily and hourly data, so hourly rows never add to the record count.
  • All measurements are metric.

8 total changes

Table/columnChange typeNotes
historyNew tableDaily weather, one row per location per day.
forecastNew table15-day forecast, kept per issue date instead of overwritten.
history
history_hourNew subtableHourly weather for the day.
history_preciptypeNew subtableUnwraps the day's precipitation type list.
history_hour_preciptypeNew subtableUnwraps the hour's precipitation type list.
forecast
forecast_hourNew subtableHourly forecast for the day.
forecast_preciptypeNew subtableUnwraps the day's precipitation type list.
forecast_hour_preciptypeNew subtableUnwraps the hour's precipitation type list.

New features

Restrict workspace members to your company email domains

Admins on a plan with Advanced Security (Business, Enterprise and the free trial) can now list the email domains their workspace accepts. While any domain is listed, only people with an address at one of those domains, or at a subdomain of one, can be invited to the workspace.

  • Case, a leading @ and a trailing dot are ignored, and the same domain is only listed once. A workspace can list up to 20 domains, and an empty list lifts the restriction.
  • The check runs when an invitation is sent, resent and accepted, so an invitation sent before the domains were set is refused when it is accepted.
  • Existing members are kept, even if their address is outside the listed domains.
  • Only admins can change the list, and changes are recorded in the audit log.
  • If your workspace moves to a plan without Advanced Security, the domains are kept but no longer enforced, and you can still remove them.
  • Domains aren't verified. Listing a domain only limits who can join your own workspace, and accepting an invitation still requires signing in with the invited address.
  • The get_account MCP server tool reports the allowed domains and whether your plan enforces them, read-only.

Manage the list in Settings → Security → Advanced security → Allowed email domains. Add several domains at once, separated by commas or spaces; each one is saved right away. Removing a domain asks you to confirm first, and removing the last one warns that anyone can be invited again. The card shows how many pending invitations to other addresses can no longer be accepted. The invite dialog names the allowed domains, and for an address outside them shows an error and keeps Send invitation disabled. See Allowed email domains.

Session length and idle timeout for your workspace

Admins on a plan with Advanced Security (Business, Enterprise and the free trial) can now control how long members stay signed in. Open Settings → Security, find Advanced security, and turn on either or both limits in the Session policy card:

  • Maximum session length: "Sign in again every" 8 hours, 12 hours, 24 hours, 7 days or 30 days.
  • Idle timeout: "Sign out after" 15 minutes, 30 minutes, 1 hour, 4 hours, 1 day, 3 days or 7 days without activity. When both are set, the idle timeout must be shorter than the maximum session length.

Details:

  • Before you save a new or shorter limit, Weld warns that members whose sign-in no longer meets it, possibly including you, are signed out on their next request. Removing a limit or making it longer doesn't ask.
  • A member counts as idle after 1 minute without mouse, keyboard, scroll-wheel or touch input. While idle, Weld pauses background refreshes and resumes them on the next input.
  • When a session has expired, a dialog explains why and Sign in again takes the member back to sign-in.
  • The policy applies to the workspace only. A member of several workspaces is signed out of the workspace whose limit was reached, not the others.
  • Changes apply right away, without members having to sign in again first.
  • MCP server clients are covered too. When a session has expired, the client has to sign in again. Connections that use an API key are not affected.
  • Idle time is counted separately for the web app and for MCP clients, so using one doesn't keep the other signed in.
  • If your workspace moves to a plan without Advanced Security, your settings are kept but no longer enforced. Settings > Security shows a notice, and you can still remove the saved policy.
  • The get_account MCP tool now reports the workspace's two-factor requirement and session policy, read-only, so AI tools can see them.

See Workspace settings.

Run a single ELT table in an orchestration

You can now attach individual tables of a data source to an orchestration, so the orchestration no longer has to wait for the rest of the sync. For example, a Postgres sync with 59 tables can put the 15 tables an hourly orchestration needs in it and leave the others on the sync's schedule.

  • In the table's menu, Configure schedule now asks how the table runs: Follow the sync (default), Run in an orchestration, or Use its own schedule. The top line shows what applies today, such as "Currently runs in Hourly, on its own."
  • Attaching a table to an orchestration clears its own interval. Choosing Follow the sync again detaches it. Run in an orchestration is disabled on plans without orchestration.
  • A table attached on its own shows an orchestration icon next to its next run, with a tooltip naming the orchestration.
  • In the orchestration's job list, a data source that has only some tables attached is labelled Selected tables instead of Removed, and the tooltip lists them. A table attached after the last run shows Recently added until the next run includes it.
  • The activity feed now records these changes, for example "changed the schedule of table orders", with details such as "now runs in an orchestration on its own".

An orchestration run's failures are now grouped under its alert

When an orchestration run fails, the alerts panel used to list the run's alert and, next to it, one alert for every model, dbt Cloud job, and webhook that failed in that run. Those failures are now nested under the run's alert, collapsed behind Show N failures in this run, matching the single email Weld already sends for the run.

  • A failure whose orchestration has no open run alert stays at the top level of the panel.
  • The bell counts top-level alerts, so one failed run counts once.
  • Removing a run's alert also removes the failures nested under it.
  • Model pages and other per-entity views still show the individual alerts.

Name and remove each two-factor method

Two-factor authentication in your user profile (Settings > User Profile > Security) now shows each method you have set up with an icon and a name: Authenticator app, Biometrics (a fingerprint or face on your device), or Security key with the device name, such as "Ada's Mac". Any other method shows as Other method. When known, each row also shows when the method was added and when it was last used.

  • Remove biometrics or a security key on its own. Open the row's menu and choose Remove, then confirm in the dialog that names the device. Your other methods keep working, and this is also available while a workspace requires 2FA, because your authenticator app stays.
  • Turn off 2FA. The authenticator app row keeps the option to turn off 2FA, now labelled Turn off 2FA (previously Disable 2FA method). It is still hidden while a workspace requires 2FA.
  • Biometrics hint. While 2FA is on and no biometrics method is set up, the page notes that biometrics can be your second step and that Weld offers it when you sign in on a device that supports it.

Source outages are retried for about two weeks before a sync is locked

When a source on your side is down, Weld no longer locks or quarantines the sync after a few runs. This covers a custom connector returning 404, 408, 5xx, timeouts, or connection resets, and database connection failures for PostgreSQL, MySQL, SQL Server, and MongoDB. Weld retries with growing gaps instead:

Failed runsWhat you getNext retry after
1 to 3Nothing, so a short blip recovers silently10 minutes, 30 minutes, 90 minutes
4 to 10One email saying Weld keeps retrying until a date; no lock3h, 6h, 12h, 1d, 2d, 4d, 8d
11The usual lock notification, which stays until you clear itThe usual back-off, so a fixed source runs soon after you clear the lock
  • A successful run resets the count and resolves an open database connection notice. Any other kind of failure also resets the count and follows the normal retry and quarantine behavior.
  • A sync in an outage is never quarantined.
  • Open notifications in the Weld Connect API and MCP server have a new retrying_until field: the date of the last automatic retry while Weld is still retrying, and unset otherwise.

Close several editor tabs at once

You can now select several editor tabs in the model tool and close them together. Cmd/Ctrl-click a tab to add it to (or remove it from) a selection without switching to it; a normal click clears the selection and switches tabs as before. Right-click a tab that's part of a selection of two or more to see Close selected (N) at the top of the context menu.

Ed can change several models at once

Ed can now take on a change that spans more than one model, like "rename customer_id in stg_customers and fix everything that reads from it". Instead of a single query for the open tab, it proposes a changeset: edits to existing models and new models, which you review and publish together.

To find the models a change touches, Ed can now read the SQL of your other models, search the SQL of your published models, and look up which models read from a given one.

Every model in a changeset is checked against your warehouse before it reaches you, with the rest of the changeset in place: a model that reads from one Ed edited or added is checked against the new version, not the published one. The changeset appears in the chat with each model marked Valid or not valid.

Review opens Review Ed's changes:

  • A diff of each model against its published version, and the new models with the folder they will be created in.
  • The published models outside the changeset that read from it, each marked as a view or a table. Views read the new versions right away; tables keep their data until they next run.
  • Reject on each model. A model that reads from a rejected one can't be published without it, so it is left out too, and the review says which model it was waiting on.

Accept publishes the models you kept in dependency order, upstream models first, and you can close the review while it runs. If a model fails, the models that read from it are skipped and the review reports a partial publish. Models already published stay published; nothing is rolled back. Afterwards the review lists any model outside the changeset that no longer passes a check against your warehouse, and any table that holds data from before the publish until its next run. Active Reverse ETL syncs on a changed model are updated the same way as when you publish from the editor.

If someone edits one of the models in the editor after Ed proposed the changeset, that model is marked Edited since the proposal: reject it to accept the rest, or ask Ed to propose again. The check runs once more right before publishing, so a changeset never overwrites an edit.

Loading into several ELT destinations is out of beta

Loading a sync into several ELT destinations has left beta, so the Beta badge is gone from the ELT Destinations item in the Settings menu, the ELT Destinations page title, the ELT destinations section on a sync's page, and the Also load into headings in the new sync form. Nothing about how it works has changed. See Load into several destinations.

The Beta label is also gone from the descriptions of the /elt_destinations and /elt_sync_destinations Weld Connect API endpoints, the additional_destinations field when creating an ELT sync, and the matching MCP server tools, such as add_elt_destination and list_elt_sync_destinations. Availability is unchanged: it is still limited to accounts with the multi-destination ELT feature.

Update a Reverse ETL sync without pausing it

You no longer need to pause a Reverse ETL sync to change it. Update sync is always available on the Configuration tab and stays disabled until you change something.

  • If an active sync runs on its own schedule, the footer shows "Saving runs the sync now." while you have unsaved edits, because saving starts a run. Paused syncs and syncs started by an orchestrator show no hint.
  • Pause sync and Resume sync now sit at the top right of the page header, next to the schedule badge, so they stay in place when you switch tabs. Run now stays on the Overview tab.
  • Update sync no longer starts enabled for syncs without a primary key mapping, such as Google Sheets in insert mode, and it disables again right after a successful save.

ELT settings no longer time out for large accounts in the Weld Connect API and MCP server

Loading a connection's ELT settings can take a long time when a source has a lot to list, for example a Stripe key with over a thousand connected accounts. Those requests used to be cut off by the gateway before they finished. They now wait a few seconds and, if the settings aren't ready, hand back a job you can poll while it keeps running.

EndpointBehavior
POST /elt_settings_jobsNew. Body: { "connection_id": "...", "partial_settings": {...} } (partial_settings is optional). Waits up to 5 seconds and returns { id, status, settings?, error? }, where status is PROCESSING, DONE, FAILED, or NOT_FOUND.
GET /elt_settings_jobs/{id}New. Polls a job by its id. A DONE or FAILED result is returned once. A malformed id returns 400.
GET /connections/{id}/settingsDeprecated. Now waits at most 25 seconds. If the settings aren't ready, it returns 504 with error_code: ELT_SETTINGS_TIMEOUT and a job_id you can pass to GET /elt_settings_jobs/{job_id}. A failing job returns 502 ELT_SETTINGS_FAILED.

The MCP get_connection_elt_settings tool returns the settings schema exactly as before when it is ready within 5 seconds. Otherwise it returns { "status": "processing", "job_id": "..." }; call the same tool again with that job_id to get the schema once it is built. The job_id is the same as the id the endpoints above return, so a job started through one can be collected through the other.

The ELT sync create and update endpoints are unchanged. See Creating an ELT sync for the full flow.

In the Weld app, the settings form for a source now loads the same way, so the source configuration page no longer fails to open on very large accounts.

Account admins can now turn Ed and other AI features off themselves

A new Ed page under Settings > AI lets an account admin switch Ed and the other AI features off or on for their workspace, without contacting Weld support. Turning it off disables Ed and its sidebar in the model editor, AI-generated documentation, and AI explanations for failed syncs; MCP clients you connect are not affected either way.

The page is only visible to members who can manage the account; everyone else keeps the toggle hidden from the nav, and sees it disabled if they open the URL directly. The notice shown elsewhere in Settings when AI is off now points admins at this page instead of Weld support.

7-day default and custom date ranges for sync success-rate stats

The Period picker above the success-rate stats on the ELT and Reverse ETL sync lists now defaults to 7 Days (previously 30 Days) and adds 90 Days, This Month, Last Month, and a custom date range.

The custom range opens a one-month calendar with an Apply button. Future days are disabled, and so are days older than the 15-month window stats are kept for; picking a range that exactly matches a calendar period (for example the 1st through the last day of last month) shows it as that period instead of a custom range. Requesting a period older than 15 months shows a clear message instead of a silent partial total.

Command palette: find models and raw views by their dotted reference path

The command palette (Cmd+K) now finds a model or raw view when you type its dotted reference path, like foo.bar.baz, the same way you'd reference it in SQL with a weld tag ({{foo.bar.baz}}). Partial paths, like bar.baz, also match. Searching by name alone still works exactly as before.

No more 10-sync cap per connection in the Weld Connect API and MCP server

Creating an ELT sync through the Weld Connect API or the MCP server no longer fails once a connection already has 10 syncs. The limit only ever applied to POST /elt_syncs and the MCP create_elt_sync tool; syncs created from the Weld app or templates were never capped.

Enabling a sync now checks that it has at least one table added. POST /elt_syncs/:id/enable and the MCP enable_elt_sync tool return a 400 error instead of silently enabling an empty sync.

Every cursor-paginated MCP list tool now documents how has_more and next_cursor work, and a page is neither the total count nor a hard cap. MCP list tools also default to pages of 50 items instead of 10; the Weld Connect API's default of 10 is unchanged.

Usage in the Weld Connect API and MCP server

You can now read your account's usage outside the Weld app, through the Weld Connect API and the MCP server. Each endpoint has a matching MCP tool that returns the same data:

EndpointMCP toolReturns
GET /usageget_usageA summary of the period: synced rows and active rows in total and split into ELT and Reverse ETL, the ELT split into main table, history table, subtables, and extra destinations, the active rows included in your plan (active_rows_limit), and rows per integration.
GET /usage/elt_streamslist_elt_stream_usageUsage per table of each ELT sync, with the same split. Filter by elt_sync_id or integration_id.
GET /usage/reverse_etl_objectslist_reverse_etl_object_usageUsage per destination object of each Reverse ETL sync. Filter by reverse_etl_sync_id or integration_id.

The two lists are sorted by active rows, most first, so the first page shows what drives your usage. They page with limit (default 10, max 100) and starting_after.

All three cover the current usage period by default. To look at another one, set start_date and end_date (exclusive) together, starting within the last 6 months.

The MCP tools are read-only and available to every role that can see usage in the app, including viewers.

Connector updates

Salesforce: Connect with the packaged Weld app

Salesforce is moving integrations from Connected Apps to External Client Apps. New Salesforce connections can now use the Weld app, an External Client App that a Salesforce admin installs into your org as a package. It is the default for new connections, and the setup form links to the install page. Switching on Use sandbox environment shows the install link for sandboxes.

If the package isn't installed in your org, Weld shows a message asking a Salesforce admin to install the Weld app. See Install the Weld app for the steps.

Databricks: Groundwork for data warehouse support

We've started adding support for using Databricks (Unity Catalog) as a data warehouse: running and validating SQL queries, and materializing models as views or tables in the account's model catalog.

PostgreSQL: Capture deletes on incremental syncs, without the WAL

You can now turn on Capture deletes for a PostgreSQL table and have its deleted rows marked with _weld_deleted_at on incremental runs, without enabling logical WAL for Change Data Capture. On each incremental run Weld reads the primary keys still present in one slice of the table and marks the rows of that slice that are no longer there.

  • The Capture deletes toggle is offered on tables with a primary key or a whole-table unique index. Tables without one don't show it.
  • Weld uses PostgreSQL's own delete counter to skip the check when nothing has been deleted, and reads keys in slices of about 5M rows, so each run stays cheap on your database. TRUNCATE and detached partitions are still caught.
  • Because deletes are found on every run, turning on Capture deletes no longer opens the full sync schedule prompt, and loading a sync with it on and no schedule no longer switches it off.

Other sources are unchanged: their tables still need a full sync schedule to reconcile deletes.

DynamoDB: Deleted items are captured as soft deletes

Items deleted in DynamoDB used to stay live in your warehouse until a full sync replaced the table. With soft deletes enabled on a DynamoDB table, incremental syncs now pick up deletes from the stream and set _weld_deleted_at on the matching row, which keeps its last values. This works with any stream view type, including New image. An item that is deleted and then written again in the same run stays live. Full syncs reconcile deletes as before. With soft deletes off, nothing changes.

A run that finds only deletes keeps its previous position and reads them again next time. If no other change arrives before the position expires after about 23 hours, the next run is a full sync, which reconciles the deletes. See the DynamoDB setup guide for details.

A new click_stats table syncs per-click data from click_view: gclid, customer_id, campaign_id, ad_group_id, date, clicks, plus the click's network, click type, device, slot, keyword, page number, and area-of-interest / location-of-presence geo fields. It's keyed on resource_name and date.

click_view rejects any query that isn't filtered to a single day, so click_stats and custom reports on click_view are now queried one day at a time instead of in 30-day windows. Because click_view only serves the last 90 days of data, click_stats starts no earlier than 88 days ago.

1 total change

Table/columnChange typeNotes
click_statsNew tablePer-click data from click_view.

Shopify: IP allowlist notice on custom-app connections

Shopify now enforces an IP allowlist on apps for bulk operations starting November 12. The Shopify with API key, Shopify Plus, and Shopify with access token connection forms now show an information box listing Weld's egress IPs to add to that custom app's allowlist. The IPs are also listed in the setup guides for Shopify Custom App, Shopify Plus, and Shopify with Access Token.

The OAuth connection form is unchanged: it uses Weld's own public app, whose allowlist is set in the Shopify Dev Dashboard rather than by the customer.

Stripe: Subscription-level default tax rates now sync

A Stripe subscription can set its tax rates at the subscription level instead of per item, stored in default_tax_rates.

A new subscription_default_tax_rate subtable now carries one row per default tax rate, with subscription_id and tax_rate_id linking back to subscription and the existing tax_rate table, the same way invoice_tax_rate and subscription_tax_rate do.

1 total change

Table/columnChange typeNotes
subscription
subscription_default_tax_rateNew subtableOne row per subscription-level default tax rate.

Amazon Selling Partner: Skip inactive listings and low-traffic ASINs in Search Query Performance

Search Query Performance makes one report request per batch of ASINs, per marketplace, per week, and Amazon limits sellers to about one report a minute. On a large catalog that dominates the sync, since its ASINs come from the merchant listings report, which includes inactive listings.

Two new settings on Seller Central syncs cut that down:

  • Exclude inactive listings leaves out any listing whose status isn't Active before looking up its ASIN. New syncs start with this ticked; existing syncs keep including inactive listings until you turn it on.
  • Only request ASINs with recent traffic, off by default, goes further: each week of the report only asks for the ASINs that had sessions, page views, or orders in that week or a configurable lookback (default 4 weeks, up to 52).

Gmail: Sync email attachments into tables

Gmail can now load recurring report attachments, like a daily CSV export, into tables, the same way the FTPS connector loads files from a file server.

Add an entry per table to the Gmail sync's settings, with a table name and any combination of a subject regex, a sender regex, and a label to match the emails it should read (a table with none of them collects every attachment of its file type). Optionally narrow it down with an attachment filename regex, then set the file type (CSV or XLSX) and, for CSV, the delimiter and lines to skip. Each table reads its emails oldest first. See Syncing email attachments for every setting.

Every row carries _message_id, _part_id, _filename, and _received_at, plus _sheet_name for Excel attachments. A table's columns come from its newest matching attachment, so a table can only be added once a matching email has arrived, and the form will say so if it can't find one yet.

While you edit the settings, Weld shows a preview of the newest matching emails and attachments for each table, and flags an invalid regex as you type it.

Facebook Ads: Every custom report field now reaches the warehouse

The custom report builder's field picker offered 23 fields that weren't mapped anywhere.

  • 15 single-value fields are now columns on facebook_ads_custom_schema, including quality_ranking, engagement_rate_ranking, conversion_rate_ranking, attribution_setting, estimated_ad_recallers, full_view_impressions, full_view_reach, and social_spend.
  • 8 list fields, including outbound_clicks, cost_per_conversion, and cost_per_thruplay, now land in their own <report>_<field> table, keyed on action_type the same way as the existing unique_outbound_clicks table.

13 total changes

Table/columnChange typeNotes
attribution_settingNew ColumnString.
quality_rankingNew ColumnString.
engagement_rate_rankingNew ColumnString.
conversion_rate_rankingNew ColumnString.
canvas_avg_view_percent, canvas_avg_view_time, cost_per_estimated_ad_recallers, cost_per_unique_click, cost_per_unique_inline_link_click, estimated_ad_recall_rate, estimated_ad_recallers, full_view_impressions, full_view_reach, social_spend, unique_inline_link_click_ctrNew Column11 more single-value fields, all numeric.
<report>_outbound_clicksNew subtable
<report>_cost_per_outbound_clickNew subtable
<report>_cost_per_unique_outbound_clickNew subtable
<report>_cost_per_unique_action_typeNew subtable
<report>_cost_per_conversionNew subtable
<report>_cost_per_thruplayNew subtable
<report>_catalog_segment_valueNew subtable
<report>_converted_product_quantityNew subtable

TikTok Shop: Refundable sample, subscription, and authentication flags on orders

The order table has three new columns that TikTok Shop now returns on each order: is_refundable_sample, is_subscription_order, and authentication_required. They are null on rows synced before this change and on orders where TikTok Shop doesn't return them.

3 total changes

Table/columnChange typeNew nameNotes
order
columnNew Columnis_refundable_sampleWhether the order is a refundable sample.
columnNew Columnis_subscription_orderWhether the order is a subscription order.
columnNew Columnauthentication_requiredWhether the order requires authentication.

Bug fixes

NetSuite: Lines deleted from an edited transaction are now soft-deleted on incremental syncs

When a NetSuite transaction is edited, NetSuite can delete a line and recreate it under a new id. Incremental syncs of transactionline and transactionaccountingline added the new line but kept the old one, so general ledger amounts could be counted twice. With soft deletes on, an incremental sync now reads every line of each transaction it picks up and sets _weld_deleted_at on any stored line of that transaction that NetSuite no longer returns. No full sync is needed.

Filter out rows where _weld_deleted_at is set in your reports. Lines of transactions deleted outright in NetSuite aren't covered by this change.

Shopify: Day/hour custom reports no longer miss data Shopify publishes late

Some ShopifyQL datasets, like search analytics, aren't fully available until Shopify finishes publishing them a day or two later. An incremental custom report grouped by day or hour only re-read yesterday and today, so a day could be read while still empty or partial and never read again. The first run of each day now looks back 5 days, picking up data that has since settled; the catch-up read doesn't count against your usage.

Was this page helpful?