Convert Claude Code or Codex into the most intelligent Deep Research Agent. Collect, search, and synthesize web research into a persistent, searchable wiki that builds on itself. Hosted API + MCP: hyperresearch.ai
Python agentsagentskillscitationsclaude-codecodexcodex-plugincodex-skilldeep-researchdeep-research-agentdeepresearchmcp-serverresearchresearch-agentresearch-tools
docs: README leads with #1 on DRACO, board chart, sample reports (#167)
Jordan Gibbs38f84cec9423
diff --git a/README.md b/README.mdindex ab23ba6..7f5476a 100644--- a/README.md+++ b/README.md@@ -16,7 +16,7 @@----**Hyperresearch turns Claude Code into a deep research agent: one that currently leads the DeepResearch-Bench RACE leaderboard (benchmarked internally). It runs in OpenAI Codex too.** A tier-adaptive 16-step pipeline takes one prompt and produces an adversarially-audited report with full source provenance. Every source it reads lands in a persistent, searchable vault, so each session starts smarter than the last.+**Hyperresearch turns Claude Code into a deep research agent, and it's #1 on the DRACO deep research benchmark, ahead of Claude Research, ChatGPT, Gemini and Perplexity. It runs in OpenAI Codex too.** A tier-adaptive 16-step pipeline takes one prompt and produces an adversarially-audited report with full source provenance. Every source it reads lands in a persistent, searchable vault, so each session starts smarter than the last.**Don't want to run it locally?** [Hyperresearch](https://hyperresearch.ai/?utm_source=github&utm_medium=readme) is the hosted version: the same pipeline, with no Claude Code or Codex install needed.@@ -31,10 +31,18 @@> Codex support ships in the next PyPI release; until then install from GitHub with `pip install git+https://github.com/jordan-gibbs/hyperresearch`. Details and what differs on Codex: [Codex](#codex).<p align="center">- <img src="assets/benchmark.png" alt="DeepResearch-Bench top-5 hyperresearch leads the chart ahead of Grep Deep Research, Cellcog Max, nvidia-aiq, Gemini Deep Research, and OpenAI Deep Research" width="780">+ <a href="https://hyperresearch.ai/benchmarks/draco?utm_source=github&utm_medium=readme"><img src="assets/draco-board.svg" alt="DRACO deep research benchmark, 10 tasks, same judge: Hyperresearch Deep 77.7 is #1, ahead of Claude Research (Opus) 73.0, Parallel Ultra 61.5, You.com Research 60.1, Perplexity Deep Research 57.8, Valyu DeepResearch Heavy 57.2, ChatGPT Deep research 55.2 and Gemini Deep Research 54.7" width="780"></a></p>-<p align="center"><sub>Forward-looking projection from a stratified pilot against the DeepResearch-Bench leaderboard snapshot (https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard). Third party validation is pending.</sub></p>+## See a sample report++Read a full report before you install anything. These are unedited Deep runs from the hosted version, and you don't need to log in:++- [Toast vs Square vs Clover](https://hyperresearch.ai/samples/restaurant-pos-competitor-analysis?utm_source=github&utm_medium=readme): a restaurant POS competitor analysis, built mostly from SEC filings (241 sources)+- [Buying a $2M HVAC business](https://hyperresearch.ai/samples/buying-an-hvac-business?utm_source=github&utm_medium=readme): multiples, SBA loans and diligence (213 sources)+- [Who is winning AI coding agents, October 2026](https://hyperresearch.ai/samples/ai-coding-agents?utm_source=github&utm_medium=readme) (259 sources)++These are the same kind of Deep runs that put Hyperresearch at #1 on DRACO. Method and per-task scores are on the [benchmark page](https://hyperresearch.ai/benchmarks/draco?utm_source=github&utm_medium=readme).## Why it wins@@ -445,7 +453,7 @@ Publishers block their own open-access PDFs often enough that one attempt isn't- It doesn't replace your judgment on which sources matter. The agent picks, you steer.- It can't fetch what's behind a paywall you haven't logged into. Open-access recovery finds a legal free copy when one exists — even when the publisher blocks the fetch outright — but when none exists you get the abstract, or nothing, and the note says so.-- On Claude Code it runs on Anthropic models via the subagent roster (per-agent assignments come from the profile's model map). On Codex, subagents use the session's model unless a Codex override is configured. Usage scales with tier, gear, and corpus size. The Codex port is new and has not been benchmarked yet; the Claude Code pipeline is the one the leaderboard numbers come from.+- On Claude Code it runs on Anthropic models via the subagent roster (per-agent assignments come from the profile's model map). On Codex, subagents use the session's model unless a Codex override is configured. Usage scales with tier, gear, and corpus size. The Codex port is new and has not been benchmarked yet; the Claude Code pipeline is the one that has been benchmarked.- The lint gate catches **structural** failures (missing scaffold, broken provenance, unresolved CRITICALs). It cannot guarantee factual accuracy, that's still your call.---diff --git a/assets/draco-board.svg b/assets/draco-board.svgnew file mode 100644index 0000000..eb8d4ce--- /dev/null+++ b/assets/draco-board.svg@@ -0,0 +1,41 @@+<svg xmlns="http://www.w3.org/2000/svg" width="880" height="504" viewBox="0 0 880 504" font-family="-apple-system,Segoe UI,Helvetica,Arial,sans-serif">+<rect width="880" height="504" rx="14" fill="#ffffff" stroke="#e5e7eb"/>+<text x="32" y="44" font-size="22" font-weight="700" fill="#111827">#1 on the DRACO deep research benchmark</text>+<text x="32" y="70" font-size="14" fill="#6b7280">Mean score across 10 tasks, same judge for every system (higher is better)</text>+<text x="276" y="117" font-size="15" text-anchor="end" fill="#111827" font-weight="700">Hyperresearch Deep</text>+<rect x="290" y="100" width="470" height="26" rx="5" fill="#f3f4f6"/>+<rect x="290" y="100" width="365.2" height="26" rx="5" fill="#4F46E5"/>+<text x="665.2" y="118" font-size="15" font-weight="700" fill="#4F46E5">77.7</text>+<text x="276" y="161" font-size="15" text-anchor="end" fill="#374151" font-weight="400">Claude Research (Opus)</text>+<rect x="290" y="144" width="470" height="26" rx="5" fill="#f3f4f6"/>+<rect x="290" y="144" width="343.1" height="26" rx="5" fill="#9ca3af"/>+<text x="643.1" y="162" font-size="15" font-weight="500" fill="#374151">73.0</text>+<text x="276" y="205" font-size="15" text-anchor="end" fill="#374151" font-weight="400">Parallel Ultra</text>+<rect x="290" y="188" width="470" height="26" rx="5" fill="#f3f4f6"/>+<rect x="290" y="188" width="289.1" height="26" rx="5" fill="#9ca3af"/>+<text x="589.0" y="206" font-size="15" font-weight="500" fill="#374151">61.5</text>+<text x="276" y="249" font-size="15" text-anchor="end" fill="#374151" font-weight="400">You.com Research</text>+<rect x="290" y="232" width="470" height="26" rx="5" fill="#f3f4f6"/>+<rect x="290" y="232" width="282.5" height="26" rx="5" fill="#9ca3af"/>+<text x="582.5" y="250" font-size="15" font-weight="500" fill="#374151">60.1</text>+<text x="848" y="250" font-size="12" text-anchor="end" fill="#9ca3af">8/10 tasks</text>+<text x="276" y="293" font-size="15" text-anchor="end" fill="#374151" font-weight="400">Perplexity Deep Research</text>+<rect x="290" y="276" width="470" height="26" rx="5" fill="#f3f4f6"/>+<rect x="290" y="276" width="271.7" height="26" rx="5" fill="#9ca3af"/>+<text x="571.7" y="294" font-size="15" font-weight="500" fill="#374151">57.8</text>+<text x="848" y="294" font-size="12" text-anchor="end" fill="#9ca3af">8/10 tasks</text>+<text x="276" y="337" font-size="15" text-anchor="end" fill="#374151" font-weight="400">Valyu DeepResearch Heavy</text>+<rect x="290" y="320" width="470" height="26" rx="5" fill="#f3f4f6"/>+<rect x="290" y="320" width="268.8" height="26" rx="5" fill="#9ca3af"/>+<text x="568.8" y="338" font-size="15" font-weight="500" fill="#374151">57.2</text>+<text x="848" y="338" font-size="12" text-anchor="end" fill="#9ca3af">8/10 tasks</text>+<text x="276" y="381" font-size="15" text-anchor="end" fill="#374151" font-weight="400">ChatGPT Deep research</text>+<rect x="290" y="364" width="470" height="26" rx="5" fill="#f3f4f6"/>+<rect x="290" y="364" width="259.4" height="26" rx="5" fill="#9ca3af"/>+<text x="559.4" y="382" font-size="15" font-weight="500" fill="#374151">55.2</text>+<text x="276" y="425" font-size="15" text-anchor="end" fill="#374151" font-weight="400">Gemini Deep Research</text>+<rect x="290" y="408" width="470" height="26" rx="5" fill="#f3f4f6"/>+<rect x="290" y="408" width="257.1" height="26" rx="5" fill="#9ca3af"/>+<text x="557.1" y="426" font-size="15" font-weight="500" fill="#374151">54.7</text>+<text x="32" y="480" font-size="12" fill="#6b7280">Hyperresearch Deep (hosted), Oct 2026. GPT-5.2 judge with the DRACO paper's grader. Method and per-task scores: hyperresearch.ai/benchmarks/draco</text>+</svg>\ No newline at end of file