Tool

query

Answer questions about your code as data. query runs SQL over the index, so you can count, list, join and compare files and declarations without opening them one by one.

Start from the views

The index is exposed as tables you can query with DuckDB SQL. Most questions start from four views:

• Files: one row per indexed file, with its path, size, headline and token count.
• Functions: functions, methods, constructors and other members of types, with their signature and line range.
• Types: classes, interfaces and other types, with what they extend and implement.
• Annotations: diagnostics such as imported lint findings.

Run DESCRIBE Files to see a view’s columns. When you need the raw graph, the base tables node and edge hold every declaration and relationship.

Scope by address. The views cover everything indexed, including the built-in help:/// manual and any imported repositories. Filter with matches_glob(uri, 'src/**'), which understands addresses; a pattern without a scheme means your repository. Don’t use LIKE with a glob. It uses % and _, so a glob matches nothing and returns no rows without an error.

Join the index to more

Registered functions return tables too, so you can join them with the views:

• search_pipeline ranks files by words and meaning, like explore.
• glob_files, grep_matches and regex_matches select files and lines.
• git_status, git_diff, git_blame, git_hotspots and git_patches read your git history.
• parse turns CSV, JSON or YAML text into rows, and xlsx reads spreadsheets.
• ask sends rows to a language model and returns its answer.
• mcp_tools lists the tools on your connected MCP servers and whether SQL can call them. Read-only tools can be called; others need to be allowed first.

Most queries run locally. A query reaches the network only when it calls something that does, such as ask or an external MCP tool.

What comes back

Terminal

rql query "SELECT name, function_kind, file, start_line
FROM Functions
WHERE matches_glob(uri, 'src/auth/**')
ORDER BY file, start_line"

Output

name	function_kind	file	start_line
SignOut	constructor	file:///src/auth/SignOut.cs	4
Execute	method	file:///src/auth/SignOut.cs	7
TokenService	constructor	file:///src/auth/TokenService.cs	4
Issue	method	file:///src/auth/TokenService.cs	7
RevokeCliTokens	method	file:///src/auth/TokenService.cs	10
Refresh	method	file:///src/auth/TokenService.cs	13

• Every method and constructor in src/auth, one row each.
• The file and start line of each. Combine them into an address, such as src/auth/TokenService.cs#symbol=TokenService.RevokeCliTokens, to read the code.
• A stable order, because the query says ORDER BY.

Keep results complete

Always use ORDER BY when order matters. Without it, rows can come back in any order, and a LIMIT keeps an arbitrary set.

Fit the result to the budget. When the rows don’t fit the token budget, you get the rows that fit and a note such as “showing 6 of 272 rows”. Select fewer columns, add LIMIT, or aggregate. If the SQL contains a comment, RepoQL tries to summarise the result in place of the cut table. That summary may not arrive, so never rely on it when you need every row.

Parameters

Further reference

In the installed manual: help:///tools/query.md, help:///commands/query.md, help:///schema/core.md, help:///schema/views/ and help:///schema/functions/.