Ed

Ed is Weld's AI agent. It works through the same tools as the Weld MCP server, so instead of answering from a fixed description of your schema, it looks things up in your workspace while it answers.

In practice that means you can describe what you want in plain language, and Ed finds the tables itself, writes the query, and checks that it is valid before it shows it to you.

By default Ed sees your workspace's metadata and not the values in your data. Data shared with AI is where that is changed.

You will find Ed in the sidebar of the model editor, on the Fix with Ed button after a query fails, and on an empty editor tab.

What Ed does

Finds the right data

Ed searches your synced tables and your existing models and reads their columns, so it writes queries against what is actually in your warehouse. You do not have to select the tables before you ask.

Checks its own SQL

Before a query reaches you, Ed validates it against your warehouse. If the check fails, Ed reads the error, corrects the query and checks again. A proposal that passed is marked Valid.

A proposal Ed could not make valid is still shown, marked as not valid, with the error and the line it came from.

Looks at the data, if you allow it

By default Ed works with your workspace's metadata only: schema, table and column names, column types, and the SQL of your models. It does not run queries and never sees the values in your data.

An admin can turn that off under Settings → AI. Ed can then run read-only queries against your warehouse, for example to see which values a column holds before it writes a filter, and a limited sample of each result is sent to the AI model. See Data shared with AI.

Applies to the editor

A valid proposal has four actions:

Apply writes the query into the open tab as a single undo step, so one Cmd+Z (Ctrl+Z on Windows) restores what was there before.
Run opens the query in the preview.
Show changes shows an inline diff against what is in the editor now.
Copy copies the SQL.

Fixes a failed query

A query that fails in the preview has a Fix with Ed button, which sends the error straight to Ed. The corrected query is checked the same way as any other proposal.

Starts from an empty editor

You do not need an open model to use Ed. A new editor tab asks What do you want to query?, and your answer opens a draft with Ed working on it. Next to each of your synced sources there is also an Explore with Ed button, which points Ed at that source and lets it suggest where to start.

Remembers the conversation

Chats are saved, so you can reopen one later. Each chat is tied to the editor tab it was about: if you move to a different tab, a proposal offers to take you back to the tab it belongs to rather than applying somewhere it does not fit. The bar above the message box lets you continue an earlier chat about the tab you are on, or start a new one.

What Ed will not do

Ed is read-only. It reads your schemas and, where AI settings allow it, runs SELECT queries. It does not write to your warehouse, create or change a model, and it cannot publish. Anything that writes still happens when you run or publish a model yourself.

Where Ed is allowed to run queries, it runs them against your warehouse, so the compute they use is billed to your warehouse like any other query you run from Weld. Each query is checked against the data access of the person asking before it runs.

Who can use Ed

Ed is available to members with permission to run queries. See Workspace settings for the roles a member can hold.

Ed and the MCP server

Ed uses the same tool definitions as the Weld MCP server: reference search, schema lookup and read-only queries are the same tools in both places. The difference is where they run. The MCP server exposes them to an agent outside Weld, such as Claude, Cursor or GitHub Copilot; Ed uses them inside Weld.