Hyperresearch

Hyperresearch is an open-source deep research Skill and command-line tool by Jordan Gibbs for Claude Code and Codex. The README says it turns Claude Code into a deep research agent and claims first place on the DRACO deep research benchmark; that is the author's own claim, and we haven't checked the leaderboard. Research runs through 16 steps, and the source texts it collects go into a Markdown archive indexed with SQLite so later research can reuse them. It critiques its sources, maps where claims conflict and checks citations one by one at the end, so each conclusion can be traced to where it came from.

Who it's for:Researchers and analysts who need traceable research material and use Claude Code or Codex.

Before you start: The local version is free and open source under the MIT licence. Claude, Codex and the optional search and Voyage or OpenAI embedding services may cost money; there's also a hosted version, which is separate from the free local one. Installing it writes Skill files, a Codex Stop hook and a short block in AGENTS.md into your project; the README shows a flag that skips the hook trust check, which the author says to use only after reading the hook. Your questions and the source content it collects go to the model you use, and the archive keeps the original texts on your computer. In Claude Code it can use your real Chrome with your existing sign-ins to open blocked pages; we don't recommend that.

Based on the project's public documentation. Not tested by AI Inspiration World.