Weld

Editor’s note: This Weld review is based on my hands-on test of Weld’s 14-day trial in October 2026. I weighed pricing and the active-row meter, connectors and sync modes, SQL modeling with Ed, reverse ETL and the API/MCP server, and security.

Quick verdict: I recommend Weld for teams of 1 to 5 people on Snowflake, BigQuery, or Databricks that want ingestion, SQL models, orchestration, and reverse ETL under one Premium plan instead of several vendors. Premium costs $389/month billed yearly, but caps connector types at 6 and excludes CDC.

Weld combines over 200 sources, 12 supported destinations plus Weld managed BigQuery, SQL models with lineage, and reverse ETL to about 25 tools. In this Weld review, I’ll take a closer look at Weld’s pricing, features, and setup, so you can see exactly whether Premium can replace separate ingestion, transformation, and activation subscriptions for your team.

Key Takeaways

• Plans start at $99/month billed yearly, followed by Premium at $389 and Business at $959; Enterprise is custom. Every plan has a 14-day trial, but no free plan.
• Active rows measure inserts and updates, not stored data. New-connector initial loads and first syncs of new tables are free.
• Connector caps count types: Basic includes 2, Premium 6, and Business 10. Extra types cost $49 or $99/month.
• Premium includes dbt and orchestration, plus custom connectors, REST API, and MCP. CDC requires a $479/month Business add-on or Enterprise.
• AWS Frankfurt hosting and SOC 2 Type II support EU-focused buyers. Weld has no self-hosted option.

Pros and Cons

How Much Does Weld Cost?

Weld sells four plans, with published prices of $99, $389, and $959/month billed yearly, plus custom Enterprise pricing. Annual billing saves 20 percent; Basic and Premium also offer monthly billing, while Business and Enterprise are annual-only.

Every plan includes a 14-day trial without a credit card. There is no free plan.

• Basic ($99/month): For a solo analyst with 2 connector types and daily refreshes; includes Ed, SSO, and 2FA.
• Premium ($389/month): My pick for a shared workflow; adds dbt, orchestration, custom connectors, REST API, and MCP.
• Business ($959/month): For multiple destinations, 15-minute refreshes, activity logs, and access tokens.
• Enterprise (custom): For custom limits, included CDC, cloud choice, a Customer Success Manager, and an SLA.

Plan detail	Basic	Premium	Business	Enterprise
Monthly price, billed yearly	$99	$389	$959	Custom
Active rows included	5M	10M	50M	100M to 10B+
Connector types	2	6	10	Custom
Users	1	3	5	Custom
Sync frequency	Daily	Hourly	15 minutes	5 minutes
Destinations	1	1	Multiple	Multiple
dbt + orchestration	No	Included	Included	Included
API + MCP server	No	Included	Included	Included
CDC	Unavailable	Unavailable	$479/month add-on	Included
Activity logs	No	No	Included	Included
Support	Basic	Premium	Priority	Priority, Customer Success Manager, SLA

Monthly add-on	Basic	Premium	Business
Extra user	$19	$19	$19
Extra connector type	$49; maximum 2 extra	$99; maximum 2 extra	$99
Extra active rows	$49 per 5M	$99 per 10M	$479 per 50M
Faster syncs	$99 for 6-hour syncs	$99 for 30-minute syncs	$99 for 5-minute syncs
CDC	Unavailable	Unavailable	$479

How Weld Counts Active Rows

An active row is inserted or updated in the destination during a month. Incremental updates count each time a row changes; unchanged rows count nothing, while full syncs count each row once monthly regardless of frequency.

Weld excludes the following activity from active-row billing:

• New connector types: Their first 14 days; custom connectors get this allowance per connection.
• New tables: Their first sync.
• Weld-triggered resyncs: Including schema-change reloads.
• Support backfills: Backfills requested through support.

Accounts created before October 19, 2026 switch to stricter counting on January 1, 2027. Accounts created on or after October 19 already follow these rules:

• Subtables: Rows count normally.
• History tables: Rows count at 50 percent.
• Additional ELT destinations: Each adds 50 percent; 1,000 active rows sent to one extra destination count as 1,500.

Check Projected Usage under Settings > Plan Details & Billing for the cycle forecast. Use Modify Plan to add rows or move up a plan; Weld does not charge for API calls.

Is Weld Good Value for Money?

• Premium earns its price when it replaces ingestion, transformation scheduling, and activation subscriptions.
• Fivetran favors small workloads: Its free plan covers 500,000 MAR; paid billing is per connection by MAR, without published plan prices.
• Lower entry prices exist: Airbyte Cloud Standard starts at $20/month and Plus at $189/month; Stitch Standard starts at $100/month.
• Separate tools deserve a comparison: Hevo Starter costs $299/month, or $265 billed yearly; Hightouch Reverse ETL is free for up to 2 active syncs.

Reviewer’s Notes: I recommend skipping Basic unless 2 connector types and daily syncs cover your workload. Start with Premium and count types, not accounts: five Shopify stores consume one type. I reserve Enterprise discussions for teams needing included CDC or a non-AWS cloud; Premium provides the bundled workflow without requiring that commitment.

My Experience With Weld

I started Weld’s 14-day trial without a credit card. I named the account before choosing storage, so the warehouse decision came before my first data connection.

Signing Up and Choosing a Warehouse

I reached the choice between my own warehouse and Weld-managed BigQuery. The managed option includes 1TB of storage and 5TB of queries monthly, with an EU or US region.

Bring your own warehouse if dashboards are the immediate goal: BI tools cannot connect to managed BigQuery. I favor the managed option for initial ingestion and SQL modeling, not as a permanent shortcut around warehouse setup for BI.

The migration path is Settings > Data Warehouse > Migrate Data Warehouse. Weld does not charge active rows for that move, so migration itself does not consume the ingestion allowance.

Adding My First Data Source

1. I opened Data Sources > Add Data Source, selected the tool, and created a connection.
2. I completed Connection Settings, where Limit historical data was an optional control.
3. I selected tables under Data to sync.
4. I configured frequency and destination schema under Configure sync, then clicked Start Initial Sync.

The initial load was excluded from active-row billing. Separate that free load from the recurring usage forecast: subsequent incremental updates count each time a row changes.

For tables without a reliable change signal, Weld provides Set up Incremental sync to select a primary key and cursor. Protect tables where the original application no longer retains the history already stored in your warehouse.

Writing My First Model With Ed

I kept the default Weld Transform Engine under Settings > Transform. GitHub and dbt were alternatives, but the built-in engine kept my SQL workflow inside Weld.

In Transform, I described the query and clicked Ask Ed. Ed wrote SQL using {{raw.source.table}} references, validated it against the warehouse, and marked it Valid.

I used Run before Publish; publishing offered a live view or a scheduled table. Ed also provides Apply, Show changes, and Copy, but publishing remained my action, not an autonomous decision by the agent.

A live view computes when queried; a scheduled table can join an orchestration. Choose the table option when downstream activation must follow a model rebuild.

Reviewer’s Notes: Decide sync modes before the initial load. Incremental syncs miss source deletes, Capture deletes is beta, and CDC requires Business plus an add-on or Enterprise. Mark tables Protected when the original application cannot resend historical records; otherwise, a schema change can trigger a full resync that loses warehouse-only history.

Connectors, Sync Modes and CDC

Weld advertises 300+ connectors; its directory lists about 217 ingestion sources, 14 destinations, and 25 reverse ETL targets. Fivetran’s 700+ managed connectors and Portable’s 1,500+ connectors offer broader catalogs.

Weld supports 12 destinations plus its managed BigQuery option:

• Snowflake
• Google BigQuery
• Databricks
• Redshift
• ClickHouse
• MotherDuck
• PostgreSQL
• MySQL
• SQL Server
• Azure Synapse
• Amazon S3
• Apache Iceberg
• Weld-managed BigQuery

Apache Iceberg support arrived in October 2026 and works with any Iceberg REST catalog. Business and Enterprise support multiple destinations per sync, subject to the additional-destination row counting explained above.

Custom connectors start on Premium and typically use Python with Weld’s connector framework. Each counts as one type regardless of reuse, with hosting on:

• Google Cloud Functions
• AWS Lambda
• Azure Functions
• Any HTTP endpoint

Data syncs and ad accounts are unlimited, but connector-type allowances still apply. Weld sets the sync mode per table, and incremental syncs do not see source deletes.

• Incremental: The default; MERGE matches primary keys, while APPEND adds rows. Ordinary incremental syncs miss source deletes.
• Full sync: Reloads tables without change tracking.
• Scheduled full or Always full sync: Provides recurring reloads instead of relying exclusively on incremental changes.
• Protected tables: Pause and notify rather than automatically reload after a schema change.
• History tables: Retain history, with rows counted at 50 percent under the newer billing rules.
• Capture deletes: Adds deletion handling, but remains beta.

Change Data Capture

Weld CDC reads PostgreSQL’s write-ahead log and the MySQL or MariaDB binlog. SQL Server and MongoDB remain in development, so I exclude them from a current CDC shortlist.

Tables need a primary key or unique index. Enabling CDC also enables Protected and removes the always-full and scheduled-full options.

CDC costs $479/month extra on Business and is included on Enterprise. I do not recommend Weld for a budget CDC deployment: Basic and Premium cannot add it.

SQL Models, Ed and Orchestration

Weld models use SQL with {{raw.source.table}} references and publish as live views or scheduled tables. Browser drafts archive after 24 hours without changes and remain restorable for 30 days.

Each model has version history. Lineage builds automatically from references, connecting syncs and source tables through models to downstream activation and webhooks.

With GitHub Sync, the repository becomes authoritative, and Weld’s model editor becomes read-only. Pushes create, update, or delete models; SQL files must live in subfolders.

Weld’s dbt integration starts on Premium, with dbt Cloud connected under Settings > Transform.

• dbt Cloud: On Premium and higher, Weld triggers jobs inside orchestration after their prerequisite syncs.
• dbt Core: The dbt workflow triggers a Weld orchestration through the API.
• Cube and SQLMesh: Webhooks connect jobs to orchestrated steps.

Weld orchestration starts on Premium; Basic presents an upgrade prompt.

• Dependency order: Syncs, table models, model tests, and reverse ETL run in lineage order.
• One schedule: The default is daily at 00:00 UTC.
• Custom cron: Scheduling remains in UTC and cannot exceed the plan’s frequency allowance.

Ed, the AI SQL Agent

Ed finds tables and writes SQL, then validates and retries on errors. Saved conversations preserve context, and Fix with Ed appears after a failed query.

I like Ed’s permission boundary: it cannot write to the warehouse, create models, or publish them. Metadata-only access is the default, covering schema details and model SQL rather than query results.

An admin can enable read-only queries under Settings > AI. A limited result sample then reaches the model, warehouse compute charges apply, and processing runs through Amazon Bedrock in EU regions without training on customer data.

Ed is included on every plan, including Basic.

Reviewer’s Notes: I recommend Weld’s engine for teams maintaining plain SQL models. For an existing dbt project, keep dbt Core or Cloud and use Weld for ingestion, orchestration triggers, and reverse ETL. Keep established dbt tests and macros rather than moving them simply to consolidate the subscription.

Reverse ETL, the REST API and the MCP Server

Weld reverse ETL sends a published model, synced table, or warehouse table/view into one destination object, such as a Salesforce Lead or HubSpot contact. I favor it for warehouse-driven CRM updates alongside ingestion.

I configured my reverse ETL sync for HubSpot contacts.

1. I chose the destination, HubSpot.
2. I set target details, selecting the contact object.
3. I mapped data, including an identifier.
4. I set the schedule, then started the sync.

Weld supports Insert, Update, Upsert, and Mirror on supported destinations, with an identifier mapping.

Weld supports about 25 activation targets, but reverse ETL consumes active rows. Hightouch’s free allowance of up to 2 active syncs deserves a comparison when activation is your only requirement.

Run now requires Premium, and editing requires pausing the sync. Its source and destination object cannot change after creation, so settle those choices before configuring field mappings.

Weld Connect REST API

Weld Connect starts on Premium and covers connections, ELT syncs, and scheduling. Its Transforms API can list, create, update, publish, and version models, with bulk creation of up to 50 models.

API-created ELT syncs currently exclude:

• Business Central
• Microsoft OneDrive
• SFTP
• Custom reports such as Facebook Ads and Google Ads

API keys carry full admin access. Treat them as workspace-wide credentials, not narrowly scoped integration tokens.

MCP Server

The Premium-included MCP server runs at connect.weld.app/mcp, using Streamable HTTP with OAuth or an API key. It works with Claude, Codex, ChatGPT Desktop, and VS Code.

Agents can list failing syncs, inspect connections, and create transforms. Unlike Ed’s read-only SQL assistance, MCP gives agents model-creation capabilities.

I favor Ed for query assistance and MCP for agents that need to operate Weld. An API key grants the same full-admin access whether used through an integration or an MCP client.

Security and Data Location

Weld runs on AWS in Frankfurt, but its service data spans other EU infrastructure. Credentials and SQL drafts reside on Google Cloud in the EU; job and activity-log data reside in MongoDB Atlas in Belgium.

Synced rows are staged temporarily, then deleted after success or within 7 days after failure.

• SOC 2 Type II certified: Report under NDA; SOC 3 available on request.
• GDPR provisions: Public DPA and sub-processor list.
• Transport protection: TLS 1.2+ and yearly penetration testing.
• Authentication: 2FA for every user and SSO.
• Permissions: Admin, Editor, and Viewer roles; activity logs start on Business.
• Retention: Deletion within 90 days after the agreement ends.
• Incident notification: Breach notification within 48 hours.
• Backups: Daily backups retained for 7 days.

Weld does not claim its own ISO 27001 certification; its data centers hold it. I do not recommend it for HIPAA or PCI DSS data without written agreement.

There is no self-hosting. Enterprise permits AWS, GCP, or Azure.

How Does Weld Compare to Competitors?

I favor Weld for small teams combining ingestion, SQL transformation, and activation, rather than buyers optimizing only connector breadth or the lowest entry price.

• Fivetran: Best for connector breadth. Its 700+ managed connectors exceed Weld’s ingestion catalog, and Standard includes 15-minute syncs with unlimited users. The free plan covers 500,000 MAR; paid usage is billed per connection by MAR, without published plan prices. I favor it when coverage outranks consolidation.
• Airbyte: Best for lower entry pricing or self-hosting. Cloud Standard starts at $20/month, while Plus starts at $189/month with SSO and 15-minute syncs. Airbyte offers a 14-day trial and an open-source core that you can self-host, an option Weld does not offer.
• Hevo Data: My pipeline-first alternative with a free tier. Its free plan includes 1M events/month, and its catalog has 170+ connectors. Starter costs $299/month, or $265 billed yearly; Professional costs $849/month, or $750 billed yearly. Hevo also has dbt integration.
• Stitch: My pick for a row-based ELT comparison. Standard starts at $100/month for 5M rows, 10 standard sources, and 1 destination. Compare that allowance with Weld Basic’s 2 connector types rather than treating their similar entry prices as equivalent.
• Portable: My catalog-first alternative for missing applications. Its 1,500+ connectors make it the alternative I prioritize when a required application is missing from Weld. Standard costs $1,800/month with 6 data flows; choose it for coverage needs rather than a lower subscription price.
• Hightouch: My activation-focused alternative. Reverse ETL is free for up to 2 active syncs, while paid plans are quote-only. Compare that allowance before paying for Weld primarily to send warehouse data back into business tools.

For Weld vs Fivetran, my deciding question is whether Premium replaces several jobs or merely ingestion. Weld alternatives also separate cleanly by requirement: Airbyte offers self-hosting, while Hightouch supplies a free starting point for activation.

Comparison	Weld	Fivetran	Airbyte Cloud	Hevo
Entry paid price	$99/month billed yearly	Not published	From $20/month	$299/month; $265 billed yearly
Free plan	No	Up to 500,000 MAR	Not published; 14-day trial	1M events/month
Billing unit	Active rows, with type and seat limits	Per connection by MAR	Not published	Events
Built-in SQL modeling	Yes	Not published	Not published	Not published
Reverse ETL	Yes	Not published	Not published	Not published
Self-hosting	No	Not published	Open-source core can be self-hosted	Not published

How I Reviewed Weld

I used Weld’s 14-day trial to follow the workflow from a data connection to a SQL model with Ed. I also set up orchestration and reverse ETL, focusing on the handoff between ingestion and activation.

• Pricing: I weighed active-row counting, connector-type caps, seats, and add-ons.
• Coverage: I evaluated supported destinations, sync modes, deletion handling, and CDC.
• Transformation: I examined SQL models, Ed permissions, lineage, and orchestration.
• Activation: I considered reverse ETL restrictions, Weld Connect, and MCP access.
• Security: I weighed EU processing locations, authentication, retention, and contractual restrictions.

I compared Weld against Fivetran, Airbyte, Hevo, Stitch, Portable, and Hightouch on their published plans. My recommendation favors workload fit over the cheapest entry price, particularly where Premium replaces separate subscriptions.

Prices current as of October 2026.

Should You Use Weld for Your Data Stack?

I recommend Weld Premium for small data teams that need ingestion, SQL models, orchestration, and reverse ETL together. Its $389/month price billed yearly is easier to justify when those features replace separate subscriptions, not when the task stops at loading a warehouse.

The connector-type model suits multiple accounts of the same application. However, Premium’s 6 types and 3 included users remain purchasing constraints, even with a 10M active-row allowance.

Skip Weld if cheap CDC, self-hosting, or a permanent free plan is essential. Fivetran offers 700+ managed connectors; prioritize Portable’s 1,500+ catalog for missing applications, or Airbyte for a self-hostable core.

Test deletion handling during the trial: ordinary incremental syncs miss source deletes. Mark tables Protected when a reload could lose historical records that the original application no longer retains.

I favor Premium over Basic for teams using orchestration and API access, but neither plan can add CDC.

List your connector types first, then start Weld’s 14-day trial on Premium and follow one workload through ingestion, a published model, and reverse ETL. Use Projected Usage to evaluate recurring active rows before choosing your allowance.

FAQs

How much does Weld cost?

Weld costs $99/month for Basic, $389/month for Premium, and $959/month for Business when billed yearly; Enterprise pricing is custom. Annual billing saves 20 percent, and only Basic and Premium also offer monthly billing. Budget separately for additional connector types, users, active rows, and any required CDC add-on.

Does Weld have a free plan?

Weld has no free plan, but every plan includes a 14-day trial without a credit card. After the trial, Basic is the lowest published option at $99/month billed yearly. Choose Premium for orchestration or API access; Basic excludes both, although it includes Ed’s AI SQL editor and SSO.

What counts as a monthly active row in Weld?

A monthly active row is a row inserted or updated in the destination. Incremental changes count each time; unchanged rows do not count, and full syncs count each row once monthly. A new table’s first sync is free. See How Weld Counts Active Rows above for history-table and additional-destination rules.

Does Weld work with dbt?

Weld works with dbt on Premium and higher plans. Weld orchestration can trigger dbt Cloud jobs after their dependent syncs, while dbt Core workflows can trigger Weld orchestration through the API. I favor retaining an established dbt project rather than migrating its tests and macros solely to consolidate tools.

Where does Weld store my data?

Weld runs on AWS in Frankfurt, with credentials and SQL drafts on Google Cloud in the EU and job/activity data in MongoDB Atlas in Belgium. Synced rows are staged until success or deleted within 7 days after failure. Destination data lives in your chosen warehouse; managed BigQuery offers EU or US regions.