AI documentation you can talk to, for every repo

Original Article ↗ Hacker News Discussion ↗

Perceived value & success cases

  • Several users report DeepWiki “just working” for certain GitHub repos, especially medium-sized, well-structured ones.
  • Positive examples: plugin-based apps, large OCaml projects with good comments, long-lived Go projects, and some personal repos where it helped contributors understand structure and extension points.
  • As a chat/search tool over an indexed repo, it’s seen as faster than cloning dependencies and using a generic code assistant.
  • Some maintainers plan to link DeepWiki for contributors, valuing its overviews and “how to add X” style guides.

Accuracy problems & hallucinations

  • Many maintainers tested it on their own projects and found serious factual errors: non-existent features treated as primary, outdated APIs, invented performance claims, wrong mutability guarantees, and misleading installation paths.
  • Outputs are described as “broken clock”: good where it can lean heavily on existing docs, poor where it must infer from code or fill gaps.
  • Concern that AI docs elevate unfinished/WIP code or internal experiments to end-user instructions.

Diagrams & information hierarchy

  • Common criticism that diagrams are arbitrary, incorrect, or superficial and emphasize implementation trivia over what users need.
  • For low-abstraction libraries, DeepWiki allegedly invents architecture layers just to satisfy a template.

Impact on OSS maintainers & ecosystem

  • Strong resentment toward unrequested, public, AI-generated docs for open-source projects, with no clear removal mechanism.
  • Fears: users get confused and never reach official docs; maintainers face extra support burden; LLMs train on LLM-written docs, amplifying errors.
  • Some label the service “parasitic” SEO slop and block it in search engines.

Scope, hosting & technical limitations

  • Despite marketing implying “any repo,” current behavior appears GitHub-centric; non-GitHub URLs often fail.
  • Indexing is on-demand and slow; some users hit “No repositories found” or CAPTCHA timeouts.

UX & accessibility complaints

  • Persistent, unhideable floating chatboxes are heavily disliked and described as anxiety-inducing.
  • Layout is criticized on small screens; accessibility of CAPTCHA flow is poor for some users.

Broader views on AI documentation

  • Split between people who find AI docs genuinely useful and those who see them as dangerous distractions.
  • Several argue AI docs work only when existing human docs are already strong; others report reasonable output even with minimal docs.
  • Some suggest using LLMs as a rough first draft for humans to correct; others see alignment/hallucination limits as a fundamental blocker.
  • Comparisons made with generic code agents (Claude Code, Gemini CLI, etc.), with some calling DeepWiki redundant unless it clearly surpasses them.