AI Tools I use: Nuanced

What is it?

Nuanced is a local code analysis tool that builds call graphs to give engineers and LLMs an understanding of how code actually behaves. It's built by engineers who worked on GitHub's code intelligence.

What's a call graph?

A call graph maps which functions call other functions. By modelling real control flow, and function relationships, call graphs helps reduce hallucinations, surface broken dependencies, and improve AI-generated code reviews, test cases, and refactors.

Why it helps

If you've any experience with using LLMs to write and review code, you'll know that they can hallucinate function calls and miss dependencies - probably because they guess relationships. How Nuanced helps here is that it maps the actual execution path so LLM responses about your codebase becomes grounded in reality instead of guesswork.

How I use it

This isn't something I use all the time - more like a fallback when I know the issue or feature I'm working on covers a lot of the codebase. I have it set as an MCP in both Codex and Cursor and call it with use nuanced. You can also use it from the CLI like this (from the docs)

# Index a package
cd my_repo/my_package
nuanced index .

# Get context for a specific function
nuanced analyze-function src/utils/helpers.ts myFunction

Check it out at nuanced.dev