Hyperresearch

Hyperresearch installs into a project with pip and adds a /hyperresearch command to Claude Code. One prompt then runs a tier-adaptive 16-step pipeline: decompose the query, sweep sources in parallel waves, map contradictions, draft three versions, have four critics attack them, and let a tool-locked patcher apply only surgical edits. Every source it reads goes into a markdown-plus-SQLite vault that later sessions reuse before fetching anything new, and a cite-checker verifies that each citation actually supports its sentence before the report ships.

Paywalled papers are swapped for legal open-access copies from Unpaywall, Europe PMC and CORE when available, with each substitution disclosed. The maintainers say it currently leads the DeepResearch-Bench RACE leaderboard in an internal projection, while noting that third-party validation is still pending. Python 3.11 to 3.13 is required; 3.14 is not yet supported.