← back to projects

contextsmith

active
RustCLITokenizationLLM

contextsmith is a CLI that builds the context bundle you feed to a language model. Think of it as grep rewritten for the question “what should I paste into Claude.”

The problem

Working with an LLM on a real codebase is a packaging problem.

Too little contextToo much context
Model invents APIsBlow the window
Hallucinated function namesPay for tokens you didn’t need
Miss the actual answerSlow round-trips

Doing this by hand, copying files, trimming, counting, is the worst part of the workflow.

What it does

You point contextsmith at a repo with a query or a path glob. It selects files, trims them deterministically, counts tokens against the model you’re targeting, and emits a clean bundle ready to paste or pipe. Same input, same output, every time. No network calls, no model in the loop, nothing to wait on.

It is built in Rust because the only thing worse than packaging context by hand is waiting for a slow tool to do it for you. Large repos, many files, sub-second runs.