Hyperresearch provides an agent-driven deep research knowledge base, transforming AI models into persistent research assistants. It automates the collection, searching, and synthesis of web research into a searchable wiki, producing adversarially-audited reports with full source provenance. This tool is designed for researchers and power users needing robust, reproducible, and transparent deep-dive analysis.

How It Works

Hyperresearch employs a sophisticated 16-step pipeline, orchestrated via Claude Code's Skill tool, to manage complex research tasks. Each step is loaded contextually to prevent "context-rot" in long processes. The system operates on two tiers: 'light' for bounded queries and 'full' for deep argumentative analysis, with distinct step sets and typical execution times (30-40 min for light, 1.5-2.5 hours for full). A core design principle is "patch, never regenerate," where modifications post-synthesis are surgical edits, preventing wholesale rewrites. Research agents leverage Anthropic's Opus, Sonnet, and Haiku models for specialized roles, from fetching and analysis to synthesis and adversarial criticism.

Quick Start & Requirements

• Install: pip install hyperresearch && hyperresearch install followed by /hyperresearch <anything> within Claude Code. A global install option (hyperresearch install --global) is available.
• Python: 3.11–3.13 (3.14 is not yet supported).
• Environment: Requires Claude Code.
• Links: DeepResearch-Bench Leaderboard: https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard

Highlighted Details

• Agent-driven 16-step research pipeline with 'light' and 'full' depth modes.
• Persistent, SQLite-indexed research vault that compounds across sessions, storing fetched sources.
• Adversarial auditing and strict source provenance enforcement via "patch, never regenerate" modifications.
• Authenticated crawling capabilities for sites requiring user login (e.g., LinkedIn, Twitter).
• Prioritizes academic APIs (Semantic Scholar, arXiv, OpenAlex, PubMed) before general web searches.

Maintenance & Community

No specific details regarding maintainers, community channels (Discord/Slack), or roadmap were found in the provided README excerpt.

Licensing & Compatibility

• License: MIT.
• Compatibility: The MIT license generally permits commercial use and integration into closed-source projects, though specific terms should be reviewed.

Limitations & Caveats

The system relies exclusively on Anthropic models (Opus, Sonnet, Haiku); porting to other architectures like Codex is a community-driven effort. Costs are variable, scaling with the chosen tier and the size of the research corpus. The tool cannot access content behind paywalls for which the user is not logged in. While it enforces provenance and adversarial review, ultimate factual accuracy remains the user's responsibility. Python 3.14 is not yet supported.