About ErrLookup
The problem: the error on your screen is not in the source
You hit an error, copy the message, and search the library's repository. Often nothing
matches. The string on your screen was assembled at run time. In the source it is a
template — Cannot find module '${request}', %s: invalid
option, {0} is out of range — or it is concatenated across
several lines, or it lives in a resource file keyed by an error code while the throw site
holds only the code. Frameworks then wrap it with their own prefixes before it reaches
your terminal.
Cloning the repo does not close that gap, and neither does handing the repo to a language model at debug time. To connect one concrete runtime message back to the expression that produced it, something has to know every placeholder syntax the library uses, follow the string through builders and resource lookups, and do it against the version you are actually running. That is slow, error-prone work to repeat on every debugging session.
What ErrLookup does about it
We do that work once per library, ahead of time. We read the source of popular
open-source libraries, find every user-facing error they can raise with its template
intact, and document each one — what it means, what triggers it, the most likely fixes,
and how to write code that avoids it. Each templated message is also compiled into a
match pattern: placeholders like ${name}, %s,
{0}, and %(name)s become wildcards, so the concrete
message you saw — with its real path, port, or module name filled in — matches the
source template it came from.
Who it's for
| AI coding agents | An MCP server answers "what is this error" at runtime from a locally cached dataset — offline-capable, no API keys. |
| Answer engines | Every page has a JSON twin and the whole dataset is published as versioned static files described in llms.txt. |
| Humans | These pages — one per error, with the exact message text you just pasted into a search box. |
How records are made
A deterministic scanner extracts every error-raising site from a pinned commit of the repository, preserving template placeholders exactly as written. Language models then document each site — meaning, trigger scenarios, ordered solutions, defensive patterns — and flag variable literals so the derived match pattern treats them as wildcards. A second, independent model pass verifies the assembled records. Every page cites the analyzed commit SHA and links the exact source line on GitHub.
Honest provenance: records are AI-assisted analysis, clearly labeled, pinned to the
commit we analyzed (440276 errors across 1963 repos
as of dataset 2026-09-16T19:17:19.609Z). If something is wrong,
tell us — the fix ships
in the next dataset version.
Who builds it
ErrLookup is a project of Standard Beagle Studio, an Austin-based, woman-owned AI UX agency for bold product teams. We build AI-enhanced developer tooling and product experiences — ErrLookup is our answer to every debugging session that started with a cryptic stack trace.