Skip to main content

lmprobe

lmprobe is the companion code-search CLI for AI agents. It gives agents structured evidence for codebase questions: file discovery, regex search, symbol definitions/references, graph relationships, and GraphQL-composed search results.

Install the public npm package:

npm install -g @lmctl-ai/lmprobe

Or run it without a global install:

npx @lmctl-ai/lmprobe --help

The detailed public manual lives at lmctl.com/lmprobe. The LLM-facing quick skill is hosted at lmctl.com/skills/lmprobe-skill.md.

Why agents use it​

lmprobe replaces fragile shell pipelines with structured evidence. Prefer it when an agent needs to answer questions like:

  • Which files define or reference this symbol?
  • Where are routes, configs, package markers, or provider integrations?
  • Which imports, callers, callees, or related code anchors support this claim?
  • Can several code-search facts be collected in one GraphQL request?

Quick commands​

lmprobe --format json find '(^|[\\/])package\.json$' .
lmprobe --format json grep 'oauth|copilot|openai' .
lmprobe --format json def LoginService .
lmprobe --format json ref LoginService .

Use GraphQL for composed evidence:

lmprobe query --format json '{
  packages: find(pattern: "(^|[\\\\/])package\\.json$", path: ".") { paths }
  providers: grep(pattern: "oauth|copilot|openai", path: ".") {
    hits { filePath lineNumber lineContent }
    exitStatus
  }
}'

Current Linux note​

@lmctl-ai/lmprobe@0.42.1 was first published as a public npm package on 2026-07-05. The Linux x64 binary in that release requires glibc 2.39, so older Linux hosts can install the package but fail at runtime with GLIBC_2.39 not found. That is a binary compatibility bug to fix in the package pipeline; it does not change the public install channel.