Web Research — bring the open web into your second brain.

Most research tools chat. Knovya searches the open web, picks the strongest sources, and lands a structured, cited note in your knowledge base — which then becomes memory for every AI you've connected through MCP. Pro removes the cap and runs on priority models.

Pipeline stages
4
Source modes — web · news · papers
3
Closed loop into memory
1
Web Research
Experiment 01 · The Lab

Pick a question. Watch it become a note.

Three real-shaped questions. A four-stage pipeline. Click one and the open web becomes a structured, cited entry in your knowledge base — the same way it does in production.

Live pipeline idle

Pick a question on the left.
Watch it become a cited note.

parse → search → synthesize → save

That note is now part of your knowledge base. The next time Claude or Cursor asks about this topic through MCP, this is the source they'll see — automatically.
All twelve · what builds a research note

Four stages, twelve components.

Every research note is built from the same anatomy. Knowing it lets you trust it — and tune it when a question deserves more than the default run.

Stage I Parse
3 components
01
Question intake The verbatim question is kept as the note's anchor. No silent rewrites. If the model softens or expands a query before searching, you see the exact rewritten string in the saved note.
02
Entity extraction People, organizations, products, dates, technical terms — pulled out and used to widen and tighten the search. The pills you see in the Lab are the same pills that drive the actual fan-out.
03
Intent typing Factual, comparative, landscape, troubleshooting. Each shape pulls a different mix of source modes — a comparison weights peer reviews, a landscape weights recent news, a fact weights primary documentation.
Stage II Search
3 components
04
Multi-engine fan-out Web, news, and academic surfaces in parallel — not sequential, not single-engine. Each surface returns its own ranked list; the synthesis stage decides which ones survive into the note.
05
Allowlist & blocklist Per-query domain control — only peer-reviewed, only government, never these aggregators. The same workspace can run strict academic research and broad web scans without contaminating either.
06
Recency window Last 24 hours, last week, last quarter, all-time. A "what changed" question and a "what is" question deserve different windows — Knovya picks a default and lets you override.
Stage III Synthesize
3 components
07
Multi-source fusion Sources that agree get consolidated; sources that disagree get flagged. The note shows you the consensus claim and the dissent next to it — disagreement is information, not noise.
08
Citation tracking Every claim carries the URL it came from. A claim without a source does not survive into the note. This is the contract — and it is enforced before the note is allowed to save.
09
Confidence flagging Single-source claims are marked. Corroborated claims are not. You see a small chip next to anything that rests on one source alone, so you can decide whether to widen the search.
Stage IV Save
3 components
10
Structured headings Question, summary, findings, sources — the same skeleton every research note keeps. Predictable shape means future-you can scan a six-month-old run in three seconds.
11
Sources block URL, title, retrieval date, source mode (web · news · paper). Editable by you. If a source rotted or moved, you can replace the link without losing the claim it supported.
12
KB integration Auto-folder, auto-tag, link to neighbors that already cover the topic. The note arrives related to the rest of your knowledge — not as a standalone artifact you need to file by hand.

Research is still copy-paste-and-pray —
and your second brain pays the bill.

Most "research" today is fourteen tabs, three rabbit holes, a paragraph copied into a doc, the URLs lost an hour later. The thinking happened. The artifact did not.

Perplexity gave you a cited chat. ChatGPT Deep Research gave you an export. NotebookLM gave you a closed sandbox over documents you already have. None of them left a real note in a place you can find again.

The cost
The work you did to find an answer evaporates the moment you close the tab. Next month you re-research the same question — usually worse, because you forgot which sources were strongest the first time.
The fix
Every research run lands as a structured, cited note in your knowledge base. It joins the rest of your work — and becomes context for every AI you've connected.
The lineage

From the Memex to your second brain.

Web Research is not invented from nothing. Five ancestors taught the open web how to become a note worth keeping.

  1. 1945
    Vannevar Bush — As We May Think The Memex: a desk-sized machine that stored documents and let the reader build associative trails between them. The idea that research should leave a navigable artifact, not just a memory. The Atlantic · July 1945
  2. 1989
    Tim Berners-Lee — World Wide Web proposal An open hypertext system at CERN. For the first time, every document on the planet had a citable address — the precondition for any tool that wants to research the web honestly. CERN · proposal · March 1989
  3. 2020
    Lewis et al. — RAG paper Retrieval-Augmented Generation: pair a language model with a retrieval index instead of training the answer in. The architecture that lets an AI cite, update, and ground itself in evidence. NeurIPS · arXiv:2005.11401
  4. 2022
    Perplexity — consumer LLM-grounded search The first mainstream product to put cited, conversational web answers in front of millions of people. It proved a market — but the answer still lived inside Perplexity's own walls. Perplexity · founded August 2022
  5. 2026
    Knovya Web Research The first to close the loop. Research lands as a structured, cited note in your own knowledge base — and immediately becomes memory for every AI you've connected through MCP. Knovya · production
First of its kind

Nobody else closes the loop.

Perplexity gives you a cited chat — and Spaces and Pages to organize it inside Perplexity. ChatGPT Deep Research gives you an export. NotebookLM gives you a sandbox over documents you already have. ScholarAI handles the academic side. There is no second product where research lands as a structured, cited note in your knowledge base — and immediately becomes memory for every AI you've connected.

  • Perplexity chat · spaces · pages · stays in perplexity
  • ChatGPT Deep Research chat · export · no kb
  • NotebookLM closed sandbox · your docs only
  • Gemini Deep Research document export · gemini-only
  • ScholarAI · Elicit academic only · paper-bound
  • Knovya open web → cited note → memory for every AI
The honesty contract

A research tool you can argue with — because every claim has a URL.

Research without provenance is a confident guess. These four rules are not settings; they are the contract Knovya keeps with the note it just saved.

  1. 01
    Every claim shows its source. Click any inline citation. The URL, the page title, and the date Knovya retrieved it are right there. No hidden synthesis.
  2. 02
    No URL, no claim. If the synthesis stage cannot anchor a sentence to at least one source, the sentence does not enter the note. Silence is preferable to fabrication.
  3. 03
    Allowlists are first-class. Per-query domain control. Restrict to peer-reviewed journals, government domains, your own published work, or any list you curate — without leaving the run.
  4. 04
    You own the note. The result is a normal Knovya note from the moment it lands. Edit it, link it, archive it, encrypt it, export it. Knovya does not hold your research hostage.
Surfaces

Web Research shows up where you already work.

Slash command, top-bar action, MCP tool, home feed. Same loop, four entry points — research never asks you to leave the place you were.

Editor slash menu inline · /research

Type /research mid-paragraph and a research note appears in line — folded by default, expandable to its full structure with all sources.

Note "Research" button on every note

Opening a note shows a Research action in the top bar. One click runs the open-web sweep on that note's topic, and the result lands as a linked sub-note.

MCP tool call knovya_research

Claude, Cursor, ChatGPT and any MCP-capable client can call knovya_research directly. The note saves; the agent's next answer reads it.

Bonded with

Web Research composes with the rest of Group I.

02
AI Memory
Research notes become memory the next AI session reads.
09
Conversation → Note
Sibling capture: chat→note pairs with research→note.
05
AI Skills
Compose research into multi-step skills you call by name.
01
MCP
Research available as a tool to Claude, Cursor, ChatGPT.
Frequently asked

A few honest answers.

What is the best AI research assistant?
The best AI research assistant is the one that does not strand your research in a chat window. Knovya searches the open web, synthesizes the strongest sources, and lands a structured, cited note in your knowledge base — which then becomes context for every AI you've connected through MCP. Most tools give you a chat. Knovya gives you a knowledge base that grows.
How is Knovya different from Perplexity?
Perplexity gives you a cited chat — and now Spaces and Pages to organize it inside Perplexity. Knovya runs the research and lands the result as a real note in your own knowledge base. That note is searchable alongside everything else you've written, surfaces in Experience Envelope, and becomes memory that Claude, Cursor, ChatGPT and any other MCP-capable AI can read. The research stops being trapped in one tool's walled garden.
How does Knovya compare to NotebookLM and ChatGPT Deep Research?
NotebookLM is closed RAG — you upload sources, it grounds responses in those documents. It does not search the open web. ChatGPT Deep Research runs a multi-step open-web search and exports a document, but the document does not become part of any reusable knowledge layer. Knovya searches the open web like Deep Research, lands a note like a real document, and then plugs that note into your AI memory layer via MCP. Different tools for different stages — Knovya is the one designed for keeping what you find.
Does Knovya cite every claim?
Yes. Every claim in a research note links back to the source URL it came from. If a claim cannot be sourced, it is not included. The sources block at the bottom of every research note lists each URL with its title and the date Knovya retrieved it. This is not a setting — it is a hard rule.
Can I scope research to trusted sources?
Yes. Each research run accepts an allowlist (only these domains) or a blocklist (never these domains). Common allowlists: peer-reviewed journals, government domains, your own published work, a competitor list, a curated news roster. The allowlist is per-query, so the same workspace can do strict academic research and broad web scans without contaminating either.
Is AI Web Research available on the Free plan?
Free includes a small monthly allowance of research runs — enough to feel the loop and decide whether the workflow earns a Pro slot. Pro removes the cap and runs research on priority models with longer query windows. Team adds shared research history and per-workspace allowlists. See pricing for current allowances.
Can Claude or Cursor trigger Knovya research?
Yes. Knovya exposes research as an MCP tool. From Claude Desktop, Cursor, ChatGPT, Goose, or any other MCP-capable client, the agent can call knovya_research with a question and receive back a structured note with citations — saved into your knowledge base in the same call. The agent's next answer is informed by what it just saved.

Bring the open web into your second brain.

Free includes a few research runs each month — enough to feel the loop. Pro removes the cap and runs on priority models.

Run your first research See Pro
element 06 · Group I — AI