Cycle on Jun 2, 2026

I will start by listing the active threads for this topic to understand the specific questions and priorities that need to be addressed in this research cycle.

I will read the existing notes for this topic to understand what has already been researched and written, so I can build on and refine them.

I will read the content of these existing notes to see exactly what they cover, starting with `market-map-positioning-hey-lefty` and others, to identify gaps or areas to refine. However, I don't have a direct tool to view the content of a note unless I fetch it or rewrite/read it. Oh, wait, I can search or read if I use other tools? No, `read_notes` only returned previews. Let's see if there is an option to read a specific note. Ah, `read_notes` lists them, but it doesn't have a specific `read_note` tool. I can search the web or check if I can fetch. Let's do some web searches and HN searches to see if there are fresh discussions or details on these specific platforms (Gemini Deep Research, Perplexity Deep Research, Google Scholar, Sacra, and how they fit into the market map). Wait, let's look at the HN stories list in the prompt. Is there any story related to research tools? Theme 4 mentions: - Florida sues OpenAI and Sam Altman over AI risks - OpenAI frontier models and Codex are now available on AWS - Can the stockmarket swallow Anthropic, SpaceX and OpenAI? - Alphabet announces $80B equity capital raise to expand AI infra and compute Theme 5 mentions: - The newest Instagram “exploit” is the goofyest I've seen - I made my phone slow on purpose - Age verification for social media, the beginning of the end for a free internet? - U.S. Midterms Have a Cyber Problem, but It's Not at the Ballot Box Theme 7 mentions: - Nvidia RTX Spark - Microsoft builds MacBook Pro rival with NVIDIA-powered Surface Laptop Ultra Let's do some web searches to find more details about these tools: 1. "Gemini Deep Research" 2. "Perplexity Deep Research" 3. "Sacra" platform and "MCP" (Model Context Protocol) or private market research data 4. "Google Scholar" AI features or "Google Scholar GPT" or similar academic tools (like Elicit, Consensus, Scite.ai) 5. Market map of autonomous AI research tools. Let's run a web search to gather concrete evidence on Gemini Deep Research first. Or let's search for "Gemini Deep Research" to understand its features, value prop, core technology, data sources, pricing, and business model.

I will fetch the Google Blog post `https://blog.google/innovation-and-ai/technology/developers-tools/deep-research-agent-gemini-api/` to get official details on Gemini Deep Research, its API (Interactions API), features, and core tech. Let's also search for Perplexity Deep Research to compare. I'll fetch the URL first.

Excellent! I have retrieved key details about Gemini Deep Research: - **Core Technology**: Powered by `Gemini 3 Pro` (Google's "most factual model yet"), trained with multi-step reinforcement learning for search to reduce hallucinations. It uses an iterative planning and searching loop (formulates queries, reads results, identifies knowledge gaps, searches again). - **Features**: Iterative web search, navigates deep into sites, analyzes uploaded documents (PDFs, CSVs, docs) via File Upload and File Search Tool, report steerability (prompt-defined structures, headers, data tables), detailed citations, structured JSON outputs. Soon to support native chart generation and Model Context Protocol (MCP) data integration. - **Data & Access**: Standard web search, uploaded files, and developer integrations. - **Distribution/Integrations**: Available in Google Search, NotebookLM, Google Finance, Gemini App (Gemini Advanced), and available to developers via the `Interactions API` (Google AI Studio) and coming to Vertex AI for enterprises. - **Value Proposition**: Automates labor-intensive initial research stages (due diligence, biotechnology/drug discovery, market research). "Unparalleled level of initial research depth and granularity," force multiplier for analysts. - **Pricing/Business Model**: Built into Gemini Advanced ($20/mo subscription) for consumers. For developers, priced per agent run/interaction via Gemini API (Interactions API), optimized to generate reports at much lower cost. Now let's search for "Perplexity Deep Research" to understand its details, features, technology, and pricing. I'll search for "Perplexity Deep Research" or "Perplexity Pro Deep Research" or "Perplexity reasoning". Let's run a web search.

Let's look up the official blog announcement of Perplexity Deep Research (`https://www.perplexity.ai/hub/blog/introducing-perplexity-deep-research`) to see what details it offers, and also let's check the Hacker News discussion for the launch of Perplexity Deep Research (HN item id `43061827`). I will fetch the Perplexity blog first.

I will also search HN for the launch of Perplexity Deep Research to read the practitioner discussion and get some real quotes about how it compares to other tools, especially in terms of usage, limitations, and pricing. Let's call `hn_comments` on `43061827` (the story mentioned in the search results: "Perplexity Deep Research"). Wait, let's first check if this item id exists and what comments it has, or use `hn_search` to verify. I'll search on HN for "Perplexity Deep Research".

Let's call `hn_comments` on `43061827` to get the top comments and quotes. This is a key source of practitioner discussion.

I have read the HN thread for Perplexity Deep Research launch. Some interesting points: - **Naming standardisation**: "Deep Research" has become a term of art. Standardized by Gemini Deep Research (Dec 2024), ChatGPT Deep Research (Feb 2025), and Perplexity Deep Research (Feb 2025). - **Comparison to ChatGPT Deep Research**: ChatGPT's Deep Research is noted as taking longer and being more expensive (part of the $200/mo Pro plan or similar), while Perplexity's is faster (completing tasks in under 3 minutes) and has a free tier with limited daily queries, making it much more accessible. - **Performance & Edge Cases**: - One user (`larsiusprime`) noted that Perplexity gave equally good results to ChatGPT Deep Research for a 50-state tax and legal obstacle table, and didn't bungle CSV downloads. - Another user (`CSMastermind`) noted that Perplexity choked on queries requiring multi-step list cross-referencing (e.g., "List the college majors of all of the Fortune 100 CEOs") where it rambled generally, whereas OpenAI and Gemini handled it better by producing structured tables. - Elicit AI also rolled out a similar feature for scientific papers. Let's do some research on other tools mentioned in the brief: - **Google Scholar**: How does it compare? What are its features? Is there an AI/synthesis layer on top of it or does it remain a traditional lexical search engine? - **Sacra**: What is Sacra? What does it do? What is its value prop? What is its data source? Who does it sell to? How does it fit into the market map? Let's search for Sacra. - Let's run a web search for "Sacra" market research platform.

Let's do a web search on Sacra to understand its core offerings, research database, and how it is being positioned as "agent-ready infrastructure" or "MCP". I'll search for "Sacra" "private market" "research" or "Sacra AI".

I will fetch the Sacra MCP launch announcement from LinkedIn or Sacra's own blog: `https://sacra.com/p/introducing-sacra-mcp-connector-chatgpt/` and check `https://sacra.com/pricing/` to see what their product, pricing, features, and core values are, and how they sell to investors/analysts. Let's do a fetch of `https://sacra.com/p/introducing-sacra-mcp-connector-chatgpt/`.

This is incredibly useful. Sacra's positioning is highly distinct: - **Core Value Prop & Positioning**: Instead of being a generic search tool or trying to compete with OpenAI/Google on the AI reasoning layer itself, Sacra acts as **premium private market data infrastructure** for AI agents. They provide high-quality, trusted, curated private market data (company profiles, datasets, interviews, research reports) that they have written and researched, and expose it to AI tools like ChatGPT, Claude, and Claude Code using the **Model Context Protocol (MCP)**. - **Target Audience / Customers**: Venture capitalists, angel investors, secondary market investors, institutional LPs, investment teams, and startup founders. - **Core Technology**: Sacra uses MCP (Model Context Protocol) to connect their database directly into the user's AI chat window (Claude, ChatGPT, etc.). This solves the "knowledge cutoff" and "hallucination" problems of generic LLMs by grounding them in curated, high-integrity financial and private market research. - **Pricing & Business Model**: Subscription-based. They have a "Pro" plan for individuals and enterprise pricing for investment firms. The MCP integrations (Claude, ChatGPT connectors) are locked behind their Pro tier. Let's do a search on Sacra's pricing to find the exact price points. Let's search for "Sacra pricing" and fetch `https://sacra.com/pricing/`.

Let's fetch `https://sacra.com/pricing/` to get the exact subscription tiers, pricing, and feature differences for Sacra. This is excellent primary evidence!

This is fantastic! Let's analyze Sacra's pricing: - **Standard**: $50/mo (1 seat, 500 tasks, reports, charts, datasets, expert interviews, MCP server access, API, revenue/valuation models, XLS/PDF export). - **Pro**: $350/mo (5 seats, 3500 tasks, same features). - **Platform**: $1500/mo (20 seats, 25000 tasks, same features, for user-facing platforms). - **Enterprise**: Custom pricing. This shows that Sacra's pricing targets professional/institutional researchers (e.g., $350/mo for a team) but is also accessible to individuals ($50/mo), offering a lot of premium datasets, expert interviews, and importantly, **MCP server access** and **API** at all tiers. Now let's look at Google Scholar. Is Google Scholar adding AI features? Let's search for "Google Scholar AI features" or "Google Scholar GPT" or "Google Scholar" vs "Elicit" or "Consensus" or "Scite". This will help us understand the academic research segment. Let's do a web search.

Let's do a search on `Undermind` and `Elicit` and `Consensus` to see how they perform academic research synthesis. Wait, let's search for "Undermind Google Scholar" or "Elicit vs Google Scholar" to understand how these specialized academic AI tools position themselves compared to traditional search. This is a very interesting angle: Google Scholar is a traditional index, but these new tools are "synthesis engines" or "deep discovery engines". Let's run a web search.

Let's fetch `https://aarontay.substack.com/p/google-scholar-vs-other-ai-search-tools` to get Aaron Tay's excellent comparison of Google Scholar vs. other academic AI search tools like Undermind, Elicit, and SciSpace. This blog post is a goldmine for understanding academic research synthesis tools, their features, and their core technology. Let's fetch it.

This is a goldmine of information! Let's analyze the Academic AI Research segment based on Aaron Tay's excellent breakdown: 1. **Google Scholar (Traditional Lexical/Keyword Search)**: - **How it works**: Uses lexical/keyword matching (TF-IDF/BM25), heavily weighting paper citations and matching title keywords. - **Features/Coverage**: Unmatched index size (200M+ works) and unmatched full-text indexing. - **Weaknesses**: Keyword-bound, lacks semantic understanding, query character limits (256 chars), weak advanced search filters, non-deterministic behaviors are not present (it's deterministic), but weak for systematic reviews due to exporting caps (max 1000 results) and lack of nested boolean. - **Target Audience**: Default starting point for all researchers/academics worldwide. 2. **Undermind.ai (Agent-Based Iterative Research & LLM Reranking)**: - **How it works**: Uses a hybrid lexical/semantic search to find initial candidates, then sends them to an LLM (e.g., GPT-4) as a reranker to evaluate relevance. It runs an **iterative, multi-step agent-like search** (takes ~3 minutes). - **Features**: Natural language prompts (expects highly detailed, conversational prompts), iterative search loops (citation chasing, adapting search queries over multiple rounds), paper relevance summaries. - **Value Prop**: High recall (recall@10 and recall@50 beat almost all other search engines for specific, niche queries). Great for disproving hypotheses or finding very specific, obscure research papers. - **Weaknesses**: Slow (takes 3+ minutes per query), not suitable for broad exploratory queries, relies on Semantic Scholar corpus metadata (less full-text access unless open access), LLM interpretations can hallucinate. 3. **Elicit.com / SciSpace / Consensus.ai (Semantic Search & RAG Synthesis)**: - **How it works**: Uses dense embeddings or learned sparse embeddings (like SPLADE in Elicit) to match queries based on semantic meaning rather than exact keywords. - **Features**: Natural language querying, data extraction into synthesis matrices (e.g., extracting population size, methodology, outcomes), citation filtering, journal quality filters (Scimago Journal Rank). - **Value Prop**: Fast, semantic similarity matching (finds papers without exact keyword matches), extracts key data from papers instantly to build literature review tables. - **Weaknesses**: Favour new papers over older, seminal ones (due to lack of citation weighting in some algorithms), can surface low-quality/predatory journal papers if filters aren't used, non-deterministic search results. Let's look at the remaining gaps in the brief: - What does **Gemini Deep Research** do? (We got the details from the Google blog: uses Gemini 3 Pro, multi-step reinforcement learning, Interactions API, handles files + web data, steerable reports, low cost). - What does **Perplexity Deep Research** do? (We got the details: uses reasoning model, performs dozens of searches, reads hundreds of sources in 2-4 minutes, generates comprehensive report, exports to PDF/Pages, free tier with daily limits, Pro tier with high volume). - What does **Sacra** do? (Premium private market research platform, pre-IPO and growth-stage company profiles, datasets, interviews, revenue/valuation models. Serves VCs/investors. Offers an MCP server and API starting at $50/mo up to $1500/mo, integrating Sacra's private market data directly into ChatGPT/Claude). Let's do a search on how these deep research tools (Gemini, Perplexity, OpenAI Deep Research) actually execute under the hood, and how they compare in terms of features, pricing, and business model. Let's do a search.

Let's fetch `https://leonfurze.com/2025/02/15/hands-on-with-deep-research/` to get a direct hands-on comparison of these three deep research tools. This will provide great details on user experience, outputs, features, and pricing.

This is a brilliant, grounded, and insightful review of **ChatGPT Deep Research**, **Gemini Deep Research**, and **Perplexity Deep Research**. Let's analyze the core components of these three tools: 1. **ChatGPT Deep Research (OpenAI)**: - **Cost**: $200 USD/month (part of the ChatGPT Pro tier). - **User Interface / Interaction**: Begins with clarifying questions to narrow down the scope (geography, age level, etc.). Displays a reasoning panel on the right showing search queries, document creation thoughts, and source comparisons. Takes about 8 minutes to run. - **Output**: Generates extremely long, comprehensive reports (often 12,000 to 16,000 words). Includes inline references, and can generate complete reference lists in standard academic styles (APA7, MLA). - **Sources / Data**: Accesses a wide range of web sources (blogs, news, legal websites, government pages). It does not have access to paywalled academic journals, which limits its academic utility. - **Value Prop**: Blows standard GPT-4o out of the water in terms of depth, accuracy, and source quality. - **Limitations**: Banal summaries, lacks deep original analysis, relies on bullet points, cannot connect the dots in a highly advanced academic context (still descriptive rather than analytical). Extremely long reports might be too dense for busy professionals. 2. **Google Gemini Deep Research (Google DeepMind)**: - **Cost**: Included in the Gemini Advanced subscription ($20/month Google One/Workspace upgrade). Also available to developers via the Interactions API (Google AI Studio) and coming to Vertex AI. - **Output / Sources**: Shorter than ChatGPT's reports, but very accurate. Tends to draw heavily on news stories and standard web search. In the deepfake comparison, it cited 17 sources. - **Core Tech**: Powered by Gemini 3 Pro (highly factual model), trained with multi-step reinforcement learning for search. - **Value Prop**: Great for quick, factual synthesis of complex web topics. Accessible directly in Google's ecosystem (Search, NotebookLM, Finance). 3. **Perplexity Deep Research (Perplexity)**: - **Cost**: Free tier with limited daily queries for logged-in users; unlimited queries for Pro subscribers ($20/month). - **Output / Sources**: Shorter, synthesized reports, but cites a massive number of sources (e.g., 57 sources in the deepfake comparison compared to ChatGPT's 21 and Gemini's 17). Draws heavily on official government and authoritative websites. - **Core Tech**: Iterative web search plus coding/reasoning capabilities. Completes tasks very fast (usually under 3 minutes). - **Value Prop**: Highly accessible, fast, high source volume, and excellent for rapid expert-level summaries across finance, marketing, technology, etc. Let's do a quick web search on "Hey, Lefty" or "Hey Lefty" to see if there is any existing web presence, or if it is a stealth/conceptual project we are helping to define and position. Let's search.

This is incredible! - **Hey, Lefty** is an **autonomous research agent for professionals** built by the **Standd** team. - **Features**: It tracks any topic daily, builds a knowledge graph, and delivers a fresh briefing every morning (an "always-on" or scheduled research agent that runs on a schedule, searches the web, writes findings, and refreshes a briefing). - **Core Value Prop & Positioning**: Unlike one-off deep research tools (which require manual, active prompting and wait times of 3–8 minutes for a single-pass report), Hey, Lefty is **always-on, scheduled, and cumulative**. It builds expertise over time on a specific topic, updating findings and delivering daily/scheduled briefings automatically. It transitions research from a "reactive, manual query" workflow to an "always-on, passive synthesis" workflow. Let's do a web search on Standd to see if we can find more about their background, their team, and what they do. This will help us ground the positioning of Hey, Lefty even more deeply.

This is incredible! - **Standd**: Founded by **Julie Saltman** (CEO & Co-Founder). Standd is an AI-native due diligence, deal intelligence, and knowledge discovery platform for lawyers, VCs, and professional services. It provides secure virtual data rooms and AI-driven search solutions tailored for legal and investment professionals. - **Hey, Lefty**: Built by the Standd team, it is an autonomous research agent for professionals that continuously tracks topics on a schedule, builds a knowledge graph, and delivers daily/scheduled briefings. - **Unique Edge / World Model**: There are mentions of "Standd's World Model" outperforming other agents, indicating a highly advanced, domain-specific reasoning architecture. Let's organize our findings. We need to write notes representing our discoveries and refine or write new notes. First, let's look at the existing notes in our system: - `market-map-positioning-hey-lefty` - `perplexity-deep-research-consumer-agent` - `academic-ai-research-scholar-vs-synthesis` - `sacra-premium-private-market-mcp-data` - `gemini-deep-research-agent` Let's update/refine these notes. To keep our findings grounded and extremely robust, we should: 1. Revise `perplexity-deep-research-consumer-agent` to reflect its launch details (Feb 2025), its free tier vs Pro tier, its performance on Humanity's Last Exam (21.1%) and SimpleQA (93.9%), and user feedback on its high speed (under 3 mins) but tendency to struggle with complex multi-step list cross-referencing. 2. Revise `gemini-deep-research-agent` to reflect its Dec 2025 reimagining via the `Interactions API`, its use of `Gemini 3 Pro` (Google's most factual model), its training on multi-step reinforcement learning for search, its features (file upload, report steerability, detailed citations, JSON outputs, upcoming MCP support), and its target audience. 3. Revise `sacra-premium-private-market-mcp-data` to include its pricing structure ($50/mo Standard, $350/mo Pro, $1500/mo Platform) and its highly strategic **Model Context Protocol (MCP)** integration for Claude and ChatGPT, positioning private market data as "agent-ready infrastructure." 4. Revise `academic-ai-research-scholar-vs-synthesis` to detail the comparison between Google Scholar (traditional lexical index with unmatched full-text coverage but keyword limitations) vs. Undermind.ai (agent-based iterative search with LLM reranking, high recall, but slow) vs. Elicit.com/SciSpace (semantic search with SPLADE/embeddings and RAG synthesis). 5. Revise `market-map-positioning-hey-lefty` to construct a brilliant, comprehensive market map. We will define the four quadrants: - **Quadrant 1: Consumer & Generalist Deep Research** (OpenAI Deep Research, Perplexity Deep Research, Gemini Deep Research - consumer side). Fast, broad web coverage, but descriptive, non-persistent (single-pass), and limited by paywalls. - **Quadrant 2: Academic & Scientific Synthesis** (Google Scholar, Undermind.ai, Elicit.com, SciSpace, Consensus). Deep scholarly indexes (200M+ works), semantic search, and literature RAG, but highly scholastic and focused on papers. - **Quadrant 3: Proprietary Vertical Data Infrastructure** (Sacra, Daloopa, expert networks). Curated, high-integrity financial and private market data, made "agent-ready" via APIs and MCP servers rather than trying to build the reasoning agents themselves. - **Quadrant 4: Always-On, Scheduled Autonomous Intelligence (Hey, Lefty)**. The white space! Instead of manual, one-off queries that require waiting 3–8 minutes, Hey, Lefty runs autonomously on a schedule (daily/weekly), continuously tracks topics, builds a persistent knowledge graph, and delivers synthesized daily briefings. It represents a paradigm shift from **reactive query-based search** to **proactive, persistent topic intelligence**. Let's write these notes one by one, ensuring we ground them in concrete quotes and links from our research. Let's write the note for `perplexity-deep-research-consumer-agent` first. I'll make sure to use exact quotes and links.

Now let's write or update the note for `gemini-deep-research-agent`. I will use the information from the Google Blog post and the hands-on review.

Now let's write or update the note for `sacra-premium-private-market-mcp-data`. I will use the pricing info and the MCP ChatGPT/Claude integration details.

Now let's write or update the note for `academic-ai-research-scholar-vs-synthesis`. I will use Aaron Tay's excellent Substack post comparing Google Scholar, Undermind.ai, and Elicit.com.

Now let's write or update the note for `market-map-positioning-hey-lefty`. This will contain our market map. I'll make sure to use wikilinks (`[[slug]]`) to link the other four notes, and explain exactly where Hey, Lefty fits and its unique positioning. This is the culmination of the research cycle! Let's write it.

I have successfully written and refined all five research notes for this topic: 1. `perplexity-deep-research-consumer-agent` (updated with launch details, benchmarks, and practitioner feedback) 2. `gemini-deep-research-agent` (updated with Google's Interactions API, Gemini 3 Pro reasoning core, and benchmark data) 3. `sacra-premium-private-market-mcp-data` (updated with detailed pricing, Model Context Protocol support, and its strategic positioning as agent-ready data infrastructure) 4. `academic-ai-research-scholar-vs-synthesis` (updated with the three distinct paradigms of academic search) 5. `market-map-positioning-hey-lefty` (updated with a complete market map of all 4 quadrants, identifying Hey, Lefty's distinct positioning in the white space of "Always-On Scheduled Intelligence") Let's call `finish_cycle` to complete this research cycle. Since I have written and updated several high-quality findings, I do not need to pass a `no_findings_reason`.

I have completed the research cycle for topic `019e8498-f497-7eb3-9d41-64bb48fe1e5d`. ### Summary of Findings & Market Map I have thoroughly investigated the landscape of autonomous and AI research tools, analyzed their core technologies, features, data sources, pricing, and target markets, and constructed a clear **four-quadrant market map** to position **Hey, Lefty**. All findings have been persisted as structured notes: 1. **[[market-map-positioning-hey-lefty]] (Market Map & Positioning: Where Does Hey, Lefty Fit?)** * **The Market Map**: * **Quadrant 1: Consumer & Generalist Deep Research** (*OpenAI Deep Research*, *Perplexity Deep Research*, *Gemini Deep Research*). Fast, broad web coverage, but reactive and "single-pass" (non-persistent). * **Quadrant 2: Academic & Scientific Synthesis** (*Google Scholar*, *Undermind.ai*, *Elicit.com*, *SciSpace*). Deep scholarly indexes (200M+ works) and literature RAG, but scholastic and bound to academic papers. * **Quadrant 3: Proprietary Vertical Data Infrastructure** (*Sacra*, *Daloopa*). Curated, high-integrity financial and private market data made "agent-ready" via APIs and Model Context Protocol (MCP) servers. * **Quadrant 4 (The White Space): Always-On, Scheduled Autonomous Intelligence (*Hey, Lefty*)**. * **Hey, Lefty's Positioning**: Hey, Lefty shifts the paradigm from **reactive, one-off query-based search** (where a user must wait 3–15 minutes for a static report) to **proactive, persistent topic tracking**. Running on a schedule (e.g., daily), it continuously tracks a topic, builds a persistent knowledge graph, and delivers a concise, high-signal briefing, providing compounding knowledge over time. 2. **[[perplexity-deep-research-consumer-agent]] (Perplexity Deep Research: Consumer-Centric High-Speed Autonomous Synthesis)** * **Features & Pricing**: Performs dozens of searches and parses hundreds of documents in under 3 minutes. Offers a limited free tier alongside a $20/month Pro plan. * **Benchmarks**: Achieves a 21.1% accuracy on *Humanity's Last Exam* and 93.9% on *SimpleQA*. * **Feedback**: Excellent for rapid synthesis of standard web content and structured tables, but struggles with multi-step structured data cross-referencing (e.g., matching Fortune 100 CEOs to their college majors). 3. **[[gemini-deep-research-agent]] (Gemini Deep Research: Multi-Step Reasoning via the Interactions API)** * **Core Tech**: Powered by **Gemini 3 Pro** (Google's most factual model) and trained using multi-step reinforcement learning for search. * **Features**: Iterative search planning, unified synthesis of web and uploaded files, and highly steerable reports with detailed citations. * **Distribution & Pricing**: Included in Gemini Advanced ($20/month). Crucially, Google has launched the **Interactions API** (via Google AI Studio) to allow developers to programmatically trigger and manage deep research runs at a lower cost. 4. **[[sacra-premium-private-market-mcp-data]] (Sacra: Premium Private Market Research as Agent-Ready Infrastructure)** * **Core Tech & GTM**: Rather than building reasoning agents, Sacra exposes its high-integrity, hand-curated private market datasets (pre-IPO profiles, expert interviews, models) directly to AI tools (Claude and ChatGPT) via **Model Context Protocol (MCP)**. * **Pricing**: Standard subscription ($50/month), Pro ($350/month for teams), and Platform ($1,500/month for user-facing platforms). All plans include MCP and API access. 5. **[[academic-ai-research-scholar-vs-synthesis]] (Academic AI Research: Google Scholar vs. Specialized Synthesis Tools)** * **The Three Paradigms**: * *Google Scholar*: Lexical/keyword-based index. Unmatched full-text coverage of major publishers, but lacks semantic understanding and is poorly suited for systematic reviews due to query limits (256 chars) and export caps (max 1000 results). * *Undermind.ai*: Agent-based iterative search. Uses hybrid search and GPT-4 as a re-ranker, chasing citations over multiple rounds. Extremely high recall for niche, obscure queries, but slow (3+ minutes) and expensive. * *Elicit / SciSpace*: Semantic search using dense/learned-sparse embeddings (SPLADE). Excellent for rapid mapping and extracting research data into structured synthesis matrices, but can favor new, un-cited, or predatory journal articles if not filtered.