Research that survives the chat

A research tool for gathering structured data with help from AI assistants and agents.

Build collections of the things you research, keep the evidence behind every figure, and work on them with your AI assistants.

Sign in How to start

AI assistants such as ChatGPT and Claude are good at quick desk research, but what they find depends on what they can access, how they search, and how they read the evidence. They can miss sources, misunderstand them, or state unsupported conclusions with confidence.

If accuracy and provenance matter in your work, as they do in mine, you probably use more than one assistant, and they often disagree. Answers and sources end up scattered across conversations, and a spreadsheet or document doesn’t track where each value came from or how it changed.

A simplified illustration of the problem and of what Infocrash does about it: asked how many people at a company hold PhDs, assistants take different routes to three different answers, and Infocrash keeps all three. 

Company A: PhD headcount

value    source                  found by   as of
-----    ------                  --------   -----
~18      team page               ChatGPT    Feb 2026
8–12     professional profiles   Claude     Jan 2026
30+      papers and patents      Claude     Feb 2026

Each comes from a different way of looking. In a chat you would see one, probably in the middle of a long conversation. Here you keep all three with their sources, and can settle on one or leave them side by side. How it works follows this example through.

Works with your AI assistants

Infocrash is used through your AI assistants. It runs as an MCP server, the standard way assistants connect to outside tools, so compatible assistants and agents can research, add evidence, and update your data while you work with them in conversation. Use the dashboard to review their work afterwards, inspect sources, and manage your data.

  • Every value keeps its sources, the assistant that found it, and the date it refers to. Nothing is overwritten.
  • Web sources are checked against the page they cite.
  • Import from a CSV file, and export to Excel.

I’m building this as I explore new ways to make my work more efficient, and it’s open to anyone who’d like to try it while I continue developing it. How to start takes about five minutes.

Use cases

  • Academic researchpapers and patents, literature reviews
  • Market and competitive researchcompetitor research, market maps, product comparisons
  • Consultingdue diligence, supplier comparison
  • Company and internal researchportfolio companies, hiring and team data
  • Journalism and policygrant and funding programs, public organizations
  • Personal projectsreading lists, trips, used cars

Working on something else? Get in touch to talk about your use case.