Methodology
AgentHub is a catalog and research surface. We expose the underlying signals so that a score is a starting point for investigation, not a claim that one project is universally best.
Sources
Catalog records are collected from public project sources such as GitHub, Hugging Face and papers. Source type, repository identity, language, license and update timestamps are stored as structured fields. A source is shown on a project profile so you can verify the original material.
Normalization
Project types and capabilities use normalized values. Tags are stored separately from display text, allowing Explore, category pages, search and API consumers to apply the same filters. Categories with multiple types use OR semantics: a project matching any selected type can appear.
Metrics and history
Stars, forks and activity are captured by daily snapshots. Seven- and thirty-day growth is calculated from the nearest available historical snapshots. Until enough history exists, cards and rankings show “not enough history” instead of a synthetic chart or growth percentage. See the status page for current service availability.
AgentHub score
The score combines persisted catalog signals including community traction, recent activity, documentation, maturity and production/compliance indicators. It is intended for comparison within the catalog and can change when source data changes. Individual dimensions remain visible on project and comparison pages.
Corrections
If a project is missing, misclassified or linked to the wrong source, please use the project submission page with the canonical URL and a short explanation.