AI assistant

The assistant is optional: QueryLane works the same without it. To use it, set up a model provider first, as described in Choose a model provider.

Open the assistant

1. Select a job in Jobs or a connection in Connections.
2. Click the AI Assistant button in the top right corner of the window.
3. Type your question in the message field at the bottom of the panel and press Enter. Shift+Enter starts a new line.

Each job and each connection has its own chat. When you select another job or connection, the panel switches to its chat. Chats are saved in your profile: QueryLane keeps the 50 most recent chats and the last 200 messages of each. To start over, click Clear chat in the panel header.

What the assistant knows

In a job chat, the assistant receives the job name, description and variables, every step with its code, dependencies and result checks, and the list of data sources. Values of secret variables are replaced with [REDACTED].

In a connection chat, it receives the connection name and type, the table or collection you have open, and the jobs that use this connection.

In both, it also receives database schemas when Include database schemas in context is on in Settings > AI. How much of your data goes with them depends on Data for AI, see Data privacy.

To add more context, type @ in the message field and pick an item: a run or all runs of the job, the schema or a sample of a table, the rows you currently see, or another job or database chat.

The ring in the message field shows how much of the model's context window the chat uses. When the chat no longer fits, the assistant summarizes older messages. You can also do it yourself with the Compact context button next to the ring.

What the assistant can do

With Tools in chat turned on in Settings > AI > Assistant tools, the assistant calls QueryLane tools. It is on by default. The assistant has the same tools as an agent connected over MCP, except seven that serve external clients, such as job export and import. The MCP tools reference lists them all.

With these tools the assistant can:

• read jobs, connections, schemas, runs and plugins;
• run a query against a database;
• create a job;
• propose a change to a job, which QueryLane validates and returns as a diff, then apply the change or revert it later;
• test a step and read the rows it returned;
• render a chart from a result and show it in the chat.

Replies can also contain code blocks. Under each block, Copy and Explain are always available. Apply replaces the code of an existing step, Add step creates a new one. In a connection chat, Run query opens the query in a new tab and runs it.

If the model does not accept tools, the assistant answers without them and tells you so. After 25 rounds of tool calls in one reply, the assistant stops. Send "continue" to go on.

Approve changes and runs

Tools that only read run at once. A tool that changes something or runs something waits for you: its block shows Needs approval and the message The assistant wants to change your data. Check the arguments below or The assistant wants to run this. Check the arguments below.

1. Expand the block and check Arguments.
2. Choose one of the buttons:
   • Allow once runs this call only.
   • Allow in this chat runs this call and later calls of the same tool in this chat without asking again, until you clear the chat or restart QueryLane.
   • Deny skips the call. The block shows Declined, and the assistant continues without it.

While the assistant replies, the send button turns into Stop AI response. Stopping ends the reply and cancels the step tests and runs the assistant started for it.

If the assistant changes a job while you have unsaved edits in it, QueryLane keeps your edits and lets you choose Keep my edits or Load the assistant's version.