Skip to content

Company

About BioWise

A structured database of repositories and agent skills, built so a developer can tell what is real.

Most developer directories are link lists. They tell you a project exists; they do not tell you whether it is maintained, what it is licensed under, whether it still works, or how to put it to use. BioWise is built around the opposite priority: every entry carries metadata that was actually retrieved from a source, with the date it was read.

The second half of the product is skills. A repository is raw material; an agent skill is a task-shaped set of instructions an AI coding agent can follow. BioWise turns one into the other with a reviewed pipeline, and publishes the result as a versioned, immutable package.

How an entry gets here

  1. 1. Resolved. A GitHub URL is parsed, normalised and looked up through the GitHub API. Duplicate detection uses GitHub's canonical numeric id, so a renamed repository is recognised rather than imported twice.
  2. 2. Classified. An editor assigns categories, tags and use cases. Where an AI provider is configured it proposes these, but they land in a draft that a person edits.
  3. 3. Published. Nothing is visible until an editor publishes it. Drafts and private entries are excluded from the sitemap, the API and the MCP server by construction.
  4. 4. Re-synced. A scheduled job refreshes stars, commits, releases and archive status, and records a failure on the entry if GitHub stops serving it.

How AI is used, and how it is labelled

  • Generated prose carries the provider, the model and the date, and says whether an editor has reviewed it.
  • Installation steps extracted by a model are marked unverified until someone has actually run them.
  • The skill generator is allowed to return nothing. Most repositories do not contain a reusable workflow, and a summary dressed up as a skill is a failure, not a result.
  • Ranking is deterministic and published: field matches, freshness and a capped popularity bonus. No model decides what you see first.

What this product does not do

  • It does not invent counts, ratings, testimonials or download totals.
  • It does not present a detected licence as legal advice. Detection is reported as detection; redistribution guidance only appears when an editor has read the licence.
  • It does not fetch URLs found inside uploaded screenshots. Extracted text is evidence for a catalog or GitHub search, never a fetch target.
  • It does not sell your browsing data. See the privacy policy.

Contact

Corrections and takedown requests are handled through the report button on any entry, which opens a moderation ticket. For partnership, enterprise or press enquiries, use the same route and mark the note accordingly, a moderator will route it.

Operated by Kode Foundry. Read the terms and privacy policy.