Show HN: Auto Wiki – Turn your codebase into a Wiki

Original Article ↗ Hacker News Discussion ↗

An AI tool that turns GitHub codebases into Wikipedia-style documentation is drawing both enthusiasm and skepticism. Users are impressed by its ability to generate overviews for large and obscure projects in many languages, but repeatedly point out factual errors, overconfident hallucinations, confusing “wiki” branding for what is currently static docs, and the risk of misleading developers who treat it as authoritative. The creators emphasize that it works purely from code, plan a more accurate v2 with verification and edit features, and are actively stress-testing it on complex open source repositories and private deployments.

Product concept & scope

  • Tool auto-generates a Wikipedia-style documentation site (“Auto Wiki”) from a codebase.
  • Works on multiple languages, including less common ones; users report success with Verilog and LaTeX, and a question is raised about Julia support.
  • Vendor claims it scales to very large repos (e.g., Linux kernel) and offers free processing for open source and private deployments for closed code.

Demo quality & coverage

  • Many users request wikis for specific repos (games, engines, wasm runtimes, Nix flakes, mixed hardware/software, monorepos, etc.) to test coverage and robustness.
  • Several users want examples on small, obscure, or poorly documented projects to ensure it’s not just reproducing existing public docs.
  • Some niche/obscure repos are showcased and generally described as impressive.

Accuracy, hallucinations & verification

  • Multiple comments highlight serious technical inaccuracies, especially in the CPython wiki (e.g., mischaracterizing the VM, inventing “tiered interpreter” features, misdescribing modules, overstating capabilities).
  • Concern that plausible but wrong content would mislead non-experts and create support burden for maintainers (“But the wiki said…”).
  • Creator states:
    • Generation is based purely on code, not external blogs.
    • System forces citations to code to reduce hallucinations.
    • A v2 is planned with chain-of-verification and fine-tuning to improve reliability.
  • Debate on whether auto-generated, partially wrong docs have net positive, zero, or negative value, especially for evolving codebases.

“Wiki” vs documentation & editability

  • Multiple participants argue this is not a “wiki” but static docs with a Wikipedia-like theme; a wiki must be easily editable.
  • Tool author acknowledges this and considers adding editing, feedback, and repo-owner overrides, potentially moving toward a true wiki model.
  • Some suggest multi-agent, wiki-style editing workflows and transparent verification processes.

UX, infra & trust issues

  • Early launch issues: “Wiki not found” errors, first-load problems, and broken links; reportedly fixed after patches.
  • GitHub app initially requested write access; this is explained as legacy from PR functionality, with intent to move to read-only permissions.
  • Users request clearer repo search, better categorization, and obvious indicators when text is AI-generated.