oh-rid

oh-rid / deep-research

Public

Claude Code plugin: deep research with 3-way LLM triangulation (Claude + Gemini via agy + GPT-5 via codex). Worktree-isolated, primary-source verified (mechanical URL + passage check), defends against shared-corpus consensus hallucination.

10
0
85% credibility
Found May 23, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
AI Summary

Deep Research is a plugin for Claude Code that transforms how you conduct web research. Instead of trusting a single AI's answer, it spawns three independent AI researchers—each using a different search engine—to investigate your question simultaneously. The tool then verifies every source by actually visiting webpages and confirming quoted passages exist. It protects against common AI failure modes like hallucinations (fabricated facts) and fake citations by requiring mechanical proof. You receive a structured report with clear confidence levels, primary-source links, and honest notes about what couldn't be verified. The goal is research you can actually trust for important decisions.

How It Works

1
🔍 You need trustworthy research

You've been burned before by AI making up facts. Now you need a tool that actually verifies its answers before you make important decisions.

2
📦 You install the research plugin

You drop the plugin into your AI assistant's folder and restart. Now a new command becomes available that changes how research works.

3
💬 You ask your research question

You type your question naturally, like 'what did the latest Fed meeting say about interest rates?' The research begins.

4
Three AI researchers get to work simultaneously
🔵
AI Researcher 1 searches the web

First AI uses its own search engine to find information

🟡
AI Researcher 2 searches the web

Second AI uses a different search engine to find the same information

🟢
AI Researcher 3 searches the web

Third AI uses yet another search engine to find the same information

5
✅ Every source gets verified

The tool doesn't just trust what the AIs found. It actually visits each webpage and confirms the quoted facts are really there—no made-up citations allowed.

6
📊 You receive a structured report

Instead of a wall of text, you get clear findings with links, confidence levels for each topic, and honest notes about what couldn't be verified.

🎯 You make your decision with confidence

You now have research you can actually trust—cross-checked by multiple systems, backed by real sources, with clear confidence levels. No more guessing if the AI made something up.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is deep-research?

Deep-research is a Claude Code plugin that performs rigorous web research using three independent LLM families as cross-checkers. When you run `/research `, it spawns parallel research streams from Claude's native WebSearch, Gemini via the agy CLI, and GPT-5 via OpenAI's codex CLI. Each uses a different search backend, and the results get triaged into agreements, conflicts, and minority dissents. Every retained claim gets verified mechanically: URLs must resolve with HTTP 2xx status, and quoted passages must actually appear in the fetched text. The output delivers a TL;DR, numbered findings with primary-source URLs, confidence ratings per topic, and a list of claims that failed verification.

Why is it gaining traction?

The hook is defense against known LLM failure modes. The project points to published research showing that LLM agreement without evidence is weaker than minority dissent with evidence (arXiv:2407.16604). When three different model families trained on overlapping corpora all agree on something false, that consensus feels confident but is actually unreliable. The plugin uses a simple rule: one cross-checker with a verified primary source beats a two-way consensus with no citation. The worktree isolation also protects your actual project files from researcher agents running untrusted web content.

Who should use this?

Developers who use Claude Code for technical research and need verifiable claims rather than confident confabulation. Researchers drafting documents where hallucinated citations could slip through peer review. Anyone working with CLI tools where a hallucinated flag could send them down a rabbit hole. If you are doing high-stakes research on academic consensus, policy documents, or technical specifications, this is worth evaluating against your current workflow.

Verdict

With only 10 stars and a credibility score of 0.85%, this is an early-stage project that has not been battle-tested by a large community. The documentation is thorough and the architecture is well-reasoned, but test coverage and community feedback remain unknowns. If you are evaluating alternatives, factor in that adoption is minimal. The core ideas are sound, but the low maturity signals caution for production reliance. Watch the repo for updates before committing.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.