// Agent profile
HyperResearch by Jordan Gibbs turns Claude Code into a deep research agent that leads the DeepResearch-Bench RACE leaderboard. Its tier-adaptive 16-step pipeline takes a single prompt and produces an adversarially-audited report with full source provenance. Every source lands in a persistent markdown-plus-SQLite vault that future sessions reuse before fetching anything new, so each run starts smarter than the last. It handles 250+ sources in a single run, seeks open-access copies of paywalled papers, and includes a local web UI with search, tag pages, and an interactive link graph.
HyperResearch is an open-source Claude Code skill by Jordan Gibbs that transforms Claude Code into a deep research agent. It currently leads the DeepResearch-Bench RACE leaderboard for research quality. The tool takes a single research prompt and runs it through a tier-adaptive 16-step pipeline that searches the web, collects sources, synthesizes findings, and produces an adversarially-audited report with full source provenance. What sets it apart is persistence: every source it reads lands in a searchable vault, so each new research session builds on everything the agent has already learned.
The core pipeline is built for thoroughness and trust. HyperResearch can process 250+ sources in a single run, with every citation verified by a skeptical cite-checker that audits whether each cited source actually supports its sentence. For academic research, it automatically seeks legal open-access copies of paywalled papers from Unpaywall, Europe PMC, and CORE. The persistent knowledge vault uses markdown files backed by a SQLite index, creating a searchable wiki that grows with every research session. A local web UI (hyperresearch serve) provides note browsing, tag pages, full-text search, and an interactive link graph for exploring the accumulated knowledge base.
Researchers use HyperResearch for literature reviews, competitive analysis, technology evaluations, and any research task that benefits from comprehensive source coverage and verifiable citations. The persistent vault makes it particularly valuable for ongoing research programs where context accumulates over weeks or months. Teams use the web UI as a shared knowledge base. The crash-resilient design with manifest-based resume makes it practical for long-running research jobs that might be interrupted.
HyperResearch requires a Claude Code subscription and consumes significant tokens during large research runs due to the multi-step pipeline and source verification. The quality of output depends on web search access and the availability of open-access papers. The SQLite vault grows over time and benefits from occasional curation.
HyperResearch is built for researchers, analysts, and knowledge workers who need thorough, well-sourced research reports and want their research tool to remember what it has already learned. It appeals to anyone who has outgrown one-shot research prompts and wants a persistent, accumulating knowledge base with rigorous source verification.
AI agents that autonomously conduct ML research experiments on single-GPU training, running a keep-or-revert loop overnight.
Stealth headless browser for AI agents — wraps a Firefox fork with C++-level fingerprint spoofing to bypass Cloudflare, bot detection, and anti-scraping.
An AI-powered academic search engine that finds and synthesizes evidence-based answers from peer-reviewed scientific research.