AI product Open source · MIT
Hyperresearch is an MIT-licensed Python command-line research harness that extends Claude Code with an agent-driven, persistent research knowledge base. It classifies a query into light, full, or opt-in dissertation work and runs a tier-adaptive pipeline covering query decomposition, source search and fetching, contradiction and locus analysis, depth investigation, evidence digestion, multi-angle drafting, synthesis, adversarial criticism, targeted gap fetching, surgical patching, citation checks, and readability review. The pipeline stores fetched sources as Markdown notes with a rebuildable SQLite index, provenance links, lifecycle statuses, full-text or optional semantic search, source-quality and retraction metadata, and resumable per-run manifests. It can fetch PDFs, seek disclosed open-access replacements for thin or blocked academic sources through Unpaywall and Europe PMC, and apply structural checks for citation bindings, quote integrity, provenance, retractions, and patch-only editing. The vault can also be accessed through an MCP server or a dependency-free local web UI. It is installed with pip for Python 3.11–3.13 and is intended to run with Claude Code; model assignments and research scale can be configured per project.
2 uses taken from transcripts — each links to the moment in the video.
An open-source harness that turns Claude Code into a disciplined deep-research agent. Its multi-step pipeline gathers and cross-checks sources, verifies citations, produces audited reports, stores research persistently, and can resume interrupted runs.
An AI deep research agent that writes reports where every citation is checked against its source. It reads web pages in full, calls over 40 data APIs for official records, and saves every source in your vault so follow-up questions build on what it already read. Connects to Claude, Cursor or ChatGPT through MCP.
2 in the library.