Academic Search Engine + Agentic AI 📚 Consensus & Zotero Deep Research
The video demonstrates Consensus AI, an academic search platform that searches over 220 million peer-reviewed papers using an agentic 'Scholar AI' system. The presenter walks through deep research workflows, library management, Zotero integration, and a citation graph feature that visually maps relationships between papers across time.
Summary
The video opens by framing the core problem with existing research tools: traditional AI hallucinates, while standard search engines return too many unfiltered results. The presenter, Callum, argues that the solution is a research management workflow where trustworthy sources are found first, and AI is then used to analyze only those vetted sources.
Callum introduces Consensus AI, a platform that indexes over 220 million peer-reviewed papers and uses an agentic 'Scholar AI' to search and synthesize before generating any response. He demonstrates a deep research query about how AI and LLMs impact learning, cognition, and flow states when paired with knowledge management systems. The deep research mode scans over a thousand papers and filters to the top 50, using a multi-agent architecture: a planning agent structures the search strategy, sub-agents execute independent searches, a reading agent processes relevant papers, and an analysis generation agent synthesizes the findings. The process took roughly 2 minutes and scanned over 101 million papers before deduplicating and selecting the top 50.
The resulting report includes a 'Consensus Meter' showing 60% of papers say AI positively impacts learning, 30% say possibly, and 0% say it negatively impacts it — a result Callum attributes to the framing of AI being paired with knowledge management systems rather than used in isolation. The report also surfaces a claims-and-evidence table, a results timeline, research gaps, and inline citations that link directly to the source papers.
Callum then explores the Library feature, which lets users save papers into collections within Consensus and chat with curated subsets of those papers using the same Scholar agent. The agent not only cites specific papers but pinpoints exact paragraphs within them, improving both credibility and precision.
A major highlight is the Zotero integration. By generating an API key in Zotero, users can import entire Zotero libraries — potentially thousands of papers built up over years — directly into Consensus. Callum notes that Consensus does not train AI models on user data or share it with third parties. Once imported, Zotero collections appear alongside native Consensus collections, and users can chat with their existing research using the Scholar agent. Papers found in Consensus can also be pushed back into Zotero, creating a two-way sync. Callum also describes a broader ecosystem connecting Consensus, Zotero, and Obsidian into a feedback loop for note-taking, paper discovery, and knowledge management.
The final major feature shown is the Citation Graph, which allows users to seed a visual map from one or more papers and see how they reference and relate to other works across time. Callum demonstrates seeding from both a 2024 paper on AI in learning management systems and the original 1998 'extended mind' paper by Clark and Chalmers, visually tracing how subsequent research has built upon that foundational concept. Users can chat with the papers within the graph, identify contradictions, find research gaps, and run additional deep searches anchored to the graph's contents.
The video concludes by briefly mentioning MCP (Model Context Protocol) support, which allows the Consensus Scholar agent to be accessed from external AI tools, effectively adding an academic research layer to existing AI workflows.
Key Insights
- Callum explains that Consensus uses a multi-agent architecture where a planning agent first structures the search strategy, sub-agents then execute independent searches in parallel, a reading agent processes the most relevant papers, and an analysis generation agent synthesizes the findings — all before generating a response.
- The Consensus Meter showed that zero of the 50 papers found said AI negatively impacts learning, which Callum attributes specifically to the framing of AI being integrated with a knowledge management system rather than being used as a standalone tool.
- Callum argues that LLMs can reduce cognitive load during information gathering but risk compromising depth of reasoning if users become passive consumers, framing his own use of AI as a tool for processing higher volume of papers rather than replacing critical reading.
- The Zotero integration allows users to import libraries of potentially thousands of papers built over years into Consensus, and Callum highlights that Consensus does not train AI models on user data, does not share it with third parties, and anonymizes everything.
- Callum demonstrates seeding a citation graph from both a 2024 paper on AI in learning systems and the original 1998 extended mind paper by Clark and Chalmers, showing how subsequent research visually traces back to that foundational concept and identifying the gap between the two eras as a potential area for further investigation.
Topics
Transcript
[0:00] Everyday millions of people search for answers. Some people use AI to save them time, but it often makes things up. It hallucinates. Others use traditional search engines, but this gives you thousands and thousands of results that you don't necessarily have time to sort through, let alone read. The fix isn't a better search engine, it's a better workflow, a research management system. Whether you're a student writing a literature review, a teacher building a curriculum, a researcher with years of papers already saved, or just someone who's looking to understand what the science actually says, the solution is the same. We need to find the right [0:31] information first and then use AI to analyze it, grounding…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Wanderloots
The Future Of AI? 🤯 Hermes Agent Intro & Benefits #ai #agenticai #hermes
Hermes is an AI agent that improves over time by learning from user workflows, unlike most static AI tools. It automatically identifies recurring patterns, creates reusable skills, and applies them to similar future tasks without explicit training.
One Agent, Many Devices? 🧠 Full Hermes Agentic AI + Tailscale Workflow
This tutorial demonstrates how to access a single Hermes AI agent from multiple devices using Tailscale VPN and a remote dashboard setup, enabling unified session history, centralized file access, and continuous memory across laptop and desktop without fragmentation.
Safer Agentic AI? 🧠 Full Docker + Hermes Agent Tutorial (Desktop)
This tutorial demonstrates how to safely run AI agents like Hermes inside Docker containers to sandbox code execution and prevent unauthorized access to sensitive files on your computer. By configuring Docker volumes and mounting specific folders, users can give agents access to only the files they choose while keeping code execution isolated from the main system.
Full Hermes Agent Tutorial (Desktop) 🧠 A Useful Agentic AI Workflow
This tutorial demonstrates how to set up and use Hermes, an open-source AI agent by Nous Research that learns and evolves over time through persistent memory and automated skill generation. The video covers local model setup with Ollama, cloud model integration with OpenAI, messaging gateway configuration via Telegram, and creating a self-improving daily AI briefing automation.
100% Free & Private AI 🦙 Build & Run Local AI Agents #ai #agenticai #ollama #localllm
The video introduces local AI models as a privacy-preserving alternative to cloud-based AI services. By running AI assistants locally on your computer, users can maintain complete control over their data while leveraging AI capabilities for personal tasks like analyzing notes and answering work-related questions.