Frequently asked questions

Weld moves data from your apps, databases and files into your own warehouse and keeps it in sync. On top of that you can model it with SQL, run everything on a schedule, send results back to your tools with reverse ETL, and give AI agents access through the MCP server.

Start with the quickstart.

Weld is primarily an ELT tool: it extracts and loads data into your cloud warehouse, then transforms it using dbt or SQL. But Weld goes beyond basic ELT: it also includes reverse ETL, orchestration, data lineage, and a Connect API for AI agent workflows, making it an agent-native data platform.

The warehouse or storage Weld loads your data into, such as Snowflake, BigQuery, Databricks or Amazon S3. It is yours: Weld moves data into it and keeps it in sync, and your models, BI tools and agents read from it. Read more about data warehouses, or see all destinations.

Yes. Every plan starts with a 14-day free trial with full access and no credit card, and usage during the trial isn't billed. There is no free plan after the trial. Sign up to start.

Apps, ad platforms, databases, files and APIs. Browse the full list in the connector directory, with a setup guide for each in the connector docs. If a source is missing, see what if I can't find the connector I need.

Snowflake, Google BigQuery, Databricks, Redshift, ClickHouse, MotherDuck, PostgreSQL, MySQL, MS SQL Server, Azure Synapse Analytics, Amazon S3 and Apache Iceberg, plus a Weld managed BigQuery if you don't have a warehouse yet. See all destinations.

During sign-up, under Let's set up your first account, choose your own warehouse at Where should your data be stored? and follow the steps for it. Each destination has its own guide in the destination docs, and the quickstart walks through the whole setup.

A BigQuery warehouse that Weld sets up and runs for you, so you can start without a warehouse of your own. It is free up to 1TB of stored data and 5TB of queries a month, in the EU or US region you choose. Read more.

No. To connect Looker Studio, Tableau, Power BI or another BI tool, move to a warehouse of your own first. Migration is under Workspace Settings → Data Warehouse → Migrate Data Warehouse: Weld re-syncs your sources in full and rebuilds your models, and the rows it moves aren't charged. See the migration guide.

Yes. Weld loads data into your own warehouse, so any BI tool that connects to it works, such as Looker Studio, Tableau, Power BI or Mode.

No. Syncing data into your warehouse needs no code, and you can work with it in whatever tool you already use. If you want to model data inside Weld, Ed, the AI assistant, writes SQL with you, and our partners can help build your data stack.

A connector links Weld to an app, database, file or ad platform so data can move to and from it. Each connector supports as many syncs and accounts as you need. Learn more.

Yes. One connector can have as many data sources as you need, for example several Shopify stores or Facebook Ad accounts. Plans count unique connector types, so this doesn't use up extra connectors. Learn more.

Tell us in the chat in your workspace or on the support page. We build new connectors based on what customers ask for.

If you need the data now, you can build a custom connector for any API. It counts as one connector type on your plan however many times you use it.

A custom connector brings in a source that isn't on the standard Weld list, for example an internal API, a niche SaaS tool or a partner system. You define how to authenticate and which data to pull, and Weld treats it like any other connector.

It's typically written in Python with Weld's connector framework and runs on Google Cloud Functions, AWS Lambda, Azure Functions or any other environment that serves HTTP endpoints.

A custom connector counts as one connector type on your plan. Once it's built, you can reuse it for as many integrations as you need.

Learn more about custom connectors.

It depends on the plan: daily on Basic, hourly on Premium, every 15 minutes on Business and every 5 minutes on Enterprise, with faster schedules available as add-ons. Premium and up can also use custom cron schedules and run syncs on demand, and each table can have its own interval. For changes within seconds to minutes, use CDC. See pricing for the add-ons.

The first sync loads all the history, so it can take a while, and a table can't be queried until it has finished. When the table's status says Completed, its data is in your warehouse. Tables with more data take longer than others.

Yes. Pick the tables to include when you set up a sync, and expand a table to hash or leave out individual columns. You can change the selection after the first sync. Learn more.

Yes. Set a start date when you create the sync, from presets like 30 or 90 days ago or this year, or leave it clear to load everything. To reload history for a table you already sync, ask us in the support chat and we'll run the backfill free of charge. Learn more.

Weld picks up the change. On an incremental sync, a schema change it can't apply in place triggers a full sync of that table, which can lose data the source no longer keeps. To decide for yourself, turn on table protection: Weld then notifies you and waits for you to start the full sync. Resyncs Weld runs after a schema change don't count toward your active rows. Learn more.

Incremental syncs only pick up new and changed rows, so a row deleted in the source stays in your warehouse until the table is fully synced again. Run a full sync to remove them.

To keep a record of deleted rows instead, turn on capture deletes: rows removed in the source are kept and stamped with _weld_deleted_at, so filter on _weld_deleted_at IS NULL for live rows. For databases, CDC picks up deletes as they happen.

Nothing changes in your warehouse. Weld loads the new data into a staging area first and only replaces the destination table once that load has fully succeeded, so you never end up with a half-overwritten table. Read more.

Open the sync under Data Sources, find the table, click the three dots next to it and choose Run sync now. To run the whole sync, stop and restart it.

Go to Settings → Connectors, open the connection, then Configuration → Authorize, and sign in with the account you want Weld to use.

If someone else owns the account, open the connection's Share Authorization Link tab and send them a link: they sign in without a Weld login. Authorization links are on Business and up. Authorization links.

Open the sync to see the error. Weld explains it and suggests a fix, and says whether you need to act: a Delayed sync retries on its own, while a Failed one usually needs something from you, such as re-authorizing the connector. Where a lock is something you can resolve, you can clear it and retry from the same screen. If it's still stuck, send us the error in the chat. Sync statuses explained.

Yes. Workspace admins can set up alerts under Settings → Notifications, by email, Slack or Microsoft Teams. Alerts cover failed syncs, connectors that need re-authorizing, failed model materializations and orchestration errors.

No. Frequent syncs against a busy production database can add load, so connect Weld to a read replica instead.

Allow Weld's IP addresses: 3.64.84.139, 3.65.119.169 and 35.156.133.78. We email you before this list changes. For a private network or a host without a fixed address, connect through an SSH tunnel to a server you control. Each database guide has the steps, for example PostgreSQL.

No. The name is used for the data in your warehouse, so it can't change after the source is created. To rename one, delete it and create it again.

Beta connectors are live and ready to use, but still in final testing, so some features or stability may not be complete yet. Tell us in the chat if you try one; feedback helps us finish it.

Soon connectors are being built now. We announce them in the release notes when they ship.

CDC tracks every insert, update and delete in a source database and sends only those changes downstream. It reads the database's transaction log or change stream instead of scanning tables, so changes arrive within seconds to minutes. CDC in Weld.

PostgreSQL, MySQL, Microsoft SQL Server and MongoDB, each with its own setup guide:

• PostgreSQL CDC
• MySQL CDC
• Microsoft SQL Server CDC
• MongoDB CDC

CDC is included on Enterprise and available as an add-on on Business. See pricing.

No, the two work together:

• Use CDC when changes need to arrive within minutes.
• Use scheduled syncs for the initial load, backfills, and tables where hourly or daily is enough.

A common pattern is an initial snapshot, then CDC to keep tables current.

Not always. CDC adds moving parts. If your tables are small, change slowly, or a few hours of delay is fine, scheduled incremental syncs are simpler and cheaper.

Yes. CDC captures inserts, updates and deletes. How a delete appears in your warehouse depends on the destination's apply strategy, so check the CDC guide for your database.

Yes, for relational databases. Each table needs a primary key, or another stable unique key, so updates and deletes land on the right rows. In MongoDB, _id does this.

Change events are staged, then applied to the destination table with MERGE or equivalent upsert and delete logic, on a schedule that matches how fresh you need the data.

Very little, since it reads logs instead of scanning tables. But the database has to keep its logs until Weld has read them, so if a CDC sync is paused or abandoned without cleanup, the logs grow and use disk.

At least:

• Replication lag
• Errors and retries
• Log retention (WAL, binlog or oplog growth)
• How long merges take in the destination

Yes. Write SQL models in Weld's editor, with Ed to help, and publish them to your warehouse as views or as tables on a schedule. Learn more.

Yes. Connect dbt Cloud under Settings → Transform and Weld can trigger your dbt jobs as part of an orchestration, right after the syncs they depend on. With dbt Core, trigger a Weld orchestration from your dbt workflow through the Weld API. Cube and SQLMesh jobs run as orchestrated steps too, triggered by webhook, with a direct webhook for Cube. You can also keep Weld's own SQL models in GitHub with GitHub Sync. The dbt integration is on Premium and up. dbt Cloud · dbt Core.

Yes. An orchestration runs ELT syncs, models and reverse ETL together, in order, on one schedule. Orchestration is on Premium and up. Learn more.

Yes, on every plan. Build a SQL model, then send its rows to tools like HubSpot, Salesforce, Klaviyo, Google Sheets, Google Ads or the Meta Conversions API, with column mapping and a schedule. Reverse ETL rows count toward your active rows. Learn more.

Weld's own models are SQL. Because your data lands in your own warehouse, you can connect any tool to it and work in Python or any other language there. Custom connectors are typically written in Python. Build a custom connector.

Published models are stored in your warehouse. Drafts stay in Weld when you close the app, and are archived automatically after 24 hours without use.

Yes. Deleted drafts and published models can be restored from View Archived Models, the archive icon at the top right of the Drafts pane. A restored model goes back to drafts, so publish it again to put it back in your warehouse.

Check that the SQL runs on its own first, and that no two columns have the same name. Ask Ed to find the problem; it checks SQL against your warehouse before suggesting a fix.

A Model Context Protocol server that lets AI clients like Claude, ChatGPT, Cursor and Copilot work in your Weld workspace: list and debug syncs, manage connections, write and publish models, run reverse ETL and orchestrations, and query your warehouse. Connect with OAuth and the agent acts with your own permissions. Set it up.

Yes. Weld's MCP server connects Claude, ChatGPT, Cursor and other MCP clients to your workspace, so agents can debug syncs, build models, run reverse ETL and query your warehouse with your own permissions. It is included from Premium and available as an add-on on Basic.

The MCP server is included on the Premium plan and above. On the Basic plan you can try it free for two weeks, and after that add it for €99 a month.

Any client that supports MCP over Streamable HTTP. Claude Desktop, claude.ai, Claude Code, Codex, ChatGPT Desktop and VS Code with GitHub Copilot log in with OAuth. Cursor, Windsurf and your own scripts use an API key in the x-api-key header.

There are 66 tools covering connections, ELT syncs and streams, custom reports, transforms, reverse ETL, orchestrations and warehouse queries. The full list, with what each tool does, is in the MCP server docs.

OAuth wherever your client supports it. There is no key to store, and the agent acts with exactly the permissions your Weld user has. An API key is scoped to one workspace but currently has full admin rights whoever created it, so keep it in an environment variable or a secrets manager. Role-scoped keys are on the way.

Yes, the tools include creating, updating and deleting syncs, models and reverse ETL syncs, within the permissions it has. Most MCP clients ask you to approve a tool call before it runs, and you can turn off the tools you do not want an agent to use.

Your data stays in your warehouse. When an agent runs a query, up to 1,000 result rows go back to your AI client, so they reach whichever model provider that client uses. Syncs, models and credentials stay in Weld.

Yes, 60 requests a minute per account and 60 per credential, plus 600 a minute per IP address. Every JSON-RPC call counts, including the initial handshake, and a warehouse query takes at least two calls. Past the limit the server returns 429 until the minute is up.

With OAuth, connect with the standard URL and tell the agent which workspace to use. It calls list_accounts to see the workspaces you can reach and works in the one you pick. An API key already belongs to one workspace.

Ed is the AI assistant in Weld's model editor. It finds the right tables, writes SQL and checks it against your warehouse before suggesting it. Ed can't write to your warehouse or publish models. Learn more.

Ed runs on models hosted by AWS in Amazon Bedrock, in EU regions only. Analysis of failed syncs runs on Gemini models in Google Cloud's Vertex AI, in the EU.

Not by default. Ed and the MCP server work with metadata, like table and column names, until an admin allows queries under Settings → AI. Then queries run as the user who asks, within that user's data access. AI data sharing.

By default Ed works with metadata only and runs no queries. If an admin lets it read data under Settings → AI, Ed acts as the user who asks it. Each query is checked against that user's data access before it runs and is refused if it reads restricted data or if the check can't tell what it reads. Restrictions are set per connector, apply to everyone except admins, and are available on BigQuery warehouses. Ed can still see the names and columns of restricted tables.

No. We don't use your data, or your inputs to and outputs from the AI features, to train AI models, and we only use AI providers whose terms don't allow it. Security.

Active rows are the rows inserted or updated in your destination in a month, and they are what plans are measured in.

• On incremental syncs, each new or updated row counts. Rows that don't change don't count.
• On a full sync, each row in the table counts once a month, however often the table is resynced.
• Rows in subtables count like any other table. Rows in history tables count at 50%.
• Each extra ELT destination a sync loads into adds 50% of its active rows.

The first 14 days of a connector type you haven't used before (custom connectors get 14 days per connection), the first sync of a new table, resyncs Weld runs itself and backfills you ask support for are free. See a worked example.

Accounts created before October 19, 2026 move to these rules on January 1, 2027. What's changing.

Yes, Weld prices by monthly active rows (MAR): the rows inserted or updated in your destination each month. Every plan includes a fixed number of them, and as long as you stay within it your price stays the same. If your volume grows, add rows or move to a higher plan directly in the app.

There are no extra costs for API calls, and a full sync counts each row once a month however often it runs. On incremental syncs, a row counts each time it changes and is synced, so syncing more often only adds usage for rows that change between syncs.

You don't have to guess. Start a free trial or connect a type of source you haven't used before, or a new custom connector, and let it run for its first 14 days: Weld counts your monthly active rows without charging for them.

In the app you can follow your usage and growth, and Projected Usage on Plan Details & Billing estimates where the current billing cycle will end, so you can see what you need before you commit to a plan. How the forecast works.

Under Settings → Usage. It shows active and total rows for the billing cycle, broken down by connector and by table, and every member of the workspace can see it. Monitor and optimize usage.

No. Everything a connector type you haven't used before syncs in its first 14 days is free, and so is the first sync of each table you add later.

The 14 days start the first time you connect that type of connector. A second connection of the same type, for example a second Shopify store, doesn't start a new 14 days, but the first sync of each of its tables is still free. Custom connectors get 14 days per connection.

To reload history for a table you already sync, ask us in the support chat and we'll run the backfill free of charge. Full syncs you start yourself count once a month, like any full sync.

Basic starts at $99 a month, Premium at $389 and Business at $959, billed yearly. Plans are priced by the monthly active rows you sync, with connectors added as you need them, and only unique connector types count. Enterprise is priced on request, and every plan starts with a 14-day free trial. Compare the plans on the pricing page.

A connector links Weld to your apps, databases or ad platforms so you can move data to and from them. Each connector supports as many syncs and accounts as you need.

We only count unique connector types. For example, you can connect several Shopify stores, Postgres databases or Facebook Ad accounts and they count as one connector each.

Learn more about connectors.

In Settings → Plan Details & Billing, choose Modify Plan to move to another plan or add active rows. You can upgrade at any time, including during the year on an annual plan. Billing docs.

In Settings → Plan Details & Billing: billing details, payment method and billing history, with your invoices, are all there. Only admins can change them. Billing docs.

Subscriptions renew automatically, so cancel before the current period ends; annual plans need notice about 30 days before renewal. You keep access until the end of the period you've paid for. Subscription docs.

Monthly or annual subscriptions, paid by credit card or invoice depending on the plan. Paying annually saves 20%. Basic and Premium can be paid monthly or annually; Business and Enterprise are billed annually. Subscription docs.

Admins invite people by email under Account Settings → Members and choose a role:

• Admin: full access, including billing and users.
• Editor: can change Data Sources, Reverse ETL and Transform, or only the areas you pick.
• Viewer: read only.

Each plan includes a number of users, and you can add more. Workspace settings.

Yes. Every user can turn on two-factor authentication in their user settings. Single sign-on is listed per plan on the pricing page.

Click Forgot password? on the sign-in screen and we'll email you a link to set a new one.

In your own warehouse. Weld runs on AWS in Frankfurt, Germany, and only stages rows during a sync, for at most 7 days. If you use a Weld managed BigQuery warehouse, your data is stored in the EU or US multi-region you choose.

Your synced data is stored in your warehouse, not by Weld. Rows are staged in Weld's EU storage during a sync and deleted when it succeeds, or within 7 days if it fails. Weld also keeps run history and error messages, which can contain values from rows that failed, and Ed conversations, as part of your account.

Yes. Weld is audited for SOC 2 Type II by an independent auditor. The report is shared under NDA; email us to request it. The SOC 3 report is available on request too. Security.

Email hello@weld.app. The SOC 2 Type II report is shared under an NDA, and the SOC 3 report on request.

Yes. Our Data Processing Agreement is public, with the list of sub-processors, and we give 30 days' notice before adding or replacing one.

We delete credentials when the agreement ends and everything else no later than 90 days after. Copies in backups expire within a further 7 days. You can ask for deletion sooner, free of charge.

Usually because of API limits, processing delays, time zones, or the source aggregating differently in its dashboard than in its API. The questions below cover the most common sources. If yours isn't here, tell us.

Most likely the daily limits of the Google Search Console API. Read more on our blog.

Pending orders, time zone shifts and attribution windows are the usual causes. Many Amazon reports also only keep data for a limited time, so older data can't be fetched; the v2 sponsored_brand_campaign_report, for example, only goes back 60 days. Read more on our blog.

GA4 samples data and estimates counts with the HyperLogLog algorithm, so its interface and its API can report different numbers. Read more on our blog.

Chat with us in your workspace or on the support page, or email hello@weld.app. To talk it through, book a call.

Monday to Friday, 9:00 to 17:00 Copenhagen time. How fast we respond depends on the problem's severity and the support level on your plan: Basic on every plan, Premium on Premium, and Priority on Business and Enterprise, where a critical issue gets a first response within 15 minutes. Service Level Agreement.

Your workspace name, the connector, sync or model and the table it concerns, what you expected and what happened and roughly when, and the error message or a screenshot. With those we can usually start without a back and forth.

Check status.weld.app for live status, incidents and their updates, and the incident history. Planned maintenance is announced there at least 48 hours ahead.