wtf is hyperresearch?

Hyperresearch turns Claude Code into a deep research assistant. You give it one question or topic, and it runs a long, structured pipeline of up to sixteen steps that searches the web, reads many sources, checks its own work, and writes a full report with citations. Every source it reads gets saved into a persistent, searchable knowledge base stored as markdown files plus a small database, so if you research a related topic later, it can reuse what it already found instead of starting from zero. What makes this different from a simple search-and-summarize tool is the amount of checking built in. After a draft report is written, several separate reviewer roles read it looking for problems, then a dedicated fixer role is only allowed to make small, surgical edits rather than rewrite the whole thing from scratch. A separate citation checker verifies that each fact cited in the report is actually backed up by the source it points to, and catches cases where many copies of the same press release are being treated as several independent sources instead of just one. The pipeline adjusts its depth automatically based on the question. Simple factual questions take a short five-step path and finish in around 30 to 40 minutes. More complex questions needing deeper argument and review use the full sixteen-step path and take one and a half to two and a half hours. There is also an opt-in dissertation mode for very large research projects, which can take four to eight hours and produce a much longer chaptered report using hundreds of sources. If a run is interrupted partway through, it can resume from where it left off rather than starting over. To use it, you install it as a Python package inside a project that uses Claude Code, then run a single command inside Claude Code to start a research task. It requires Python version 3.11 through 3.13. The project is released under the MIT license.