Auto-indexing inference configuration reference

This document details how a site administrator can supply a Lua script to customize the way Sourcegraph detects precise code intelligence indexing jobs from repository contents.

By default, Sourcegraph will attempt to infer (or hint) index jobs for the following languages:

Inference logic can be disabled or altered in the case when the target repositories do not conform to a pattern that the Sourcegraph default inference logic recognizes. Inference logic is controlled by a Lua override script that can be supplied in the UI under Admin > Code graph > Inference.


The Lua override script ultimately must return an auto-indexing config object. A configuration that neither disables or adds new recognizers does not change the default inference behavior.

return require("sg.autoindex.config").new({
  -- Empty configuration (see below for usage)

To disable default behaviors, you can re-assign a recognizer value to false. Each of the built-in recognizers are prefixed with sg. (and are the only ones allowed to be).

return require("sg.autoindex.config").new({
  -- Disable default Python inference
  ["sg.python"] = false

To add additional behaviors, you can create and register a new recognizer. A recognizer is an interface that requests some set of files from a repository, and returns a set of auto-indexing job configurations that could produce a precise code intelligence index.

A path recognizer is a concrete recognizer that advertises a set of path globs it is interested in, then invokes its generate function with matching paths from a repository. In the following, all files matching Snek.module (Snek.module, proj/Snek.module, proj/sub/Snek.module, etc) are passed to a call to generate (if non-empty). The generate function will then return a list of indexing job descriptions. The guide for auto-indexing jobs configuration gives detailed descriptions on the fields of this object.

local path = require("path")
local pattern = require("sg.autoindex.patterns")
local recognizer = require("sg.autoindex.recognizer")

local snek_recognizer = recognizer.new_path_recognizer {
  patterns = {
    -- Look for Snek.module files 
    -- (would match Snek.module; proj/Snek.module, proj/sub/Snek.module, etc)

    -- Ignore any files in test or vendor directories
    pattern.new_path_exclude(pattern.new_path_combine {

  -- Called with list of matching Snek.module files
  generate = function(_, paths)
    local jobs = {}
    for i = 1, #paths do
      -- Create indexing job description for each matching file
      table.insert(jobs, {
        indexer = "acme/snek:latest",  -- Run this indexer...
        root = path.dirname(paths[i]), -- this directory
        steps = {},
        indexer_args = {},
        outfile = "",

    return jobs

return require("sg.autoindex.config").new({
  -- Register new recognizer
  ["acme.snek"] = snek_recognizer,

Available libraries

There are a number of specific and general-purpose Lua libraries made accessible via the built-in require.


This auto-indexing-specific library defines the following two functions.

  • new_path_recognizer creates a recognizer from a config object containing patterns and generate fields. See the example above for basic usage.
  • new_fallback_recognizer creates a recognizer from an ordered list of recognizers. Each recognizer is called sequentially, until one of them emits non-empty results.


This auto-indexing-specific library defines the following four path pattern constructors.

  • new_path_literal(pattern) creates a pattern that matches an exact filepath.
  • new_path_segment(pattern) creates a pattern that matches a directory name.
  • new_path_basename(pattern) creates a pattern that matches a basename exactly.
  • new_path_extension(ext_no_dot) creates a pattern that matches files with a given extension.

This library also defines the following two pattern collection constructors.

  • new_path_combine(patterns) creates a pattern collection object (to be used with recognizers) from the given set of path patterns.
  • new_path_exclude(patterns) creates a new inverted pattern collection object. Paths matching these patterns are filtered out from the set of matching filepaths given to a recognizer's generate function.


This library defines the following five path utility functions:

  • ancestors(path) returns a path's parent, grandparent, etc as a list.
  • basename(path) returns the basename of the given path.
  • dirname(path) returns the dirname of the given path.
  • join(paths) returns a filepath created by joining the given path segments via filepath separator.
  • split(path) split a path into an ordered sequence of path segments.


This library defines the following two JSON utility functions:

  • encode(val) returns a JSON-ified version of the given Lua object.
  • decode(json) returns a Lua table representation of the given JSON text.


Lua Functional is a high-performance functional programming library accessible via local fun = require("fun"). This library has a number of functional utilities to help make recognizer code a bit more expressive.