How do you build a company knowledge base for ChatGPT and Claude?

Build one company memory from the tools where work already happens, and let ChatGPT, Claude and every other agent ask it. Each person's agent then answers from the company's own knowledge, limited to what that person can already open at work, with a source on every answer.

Last updated 11 November 2026 · Mert Uzun, Co-Founder

Why it matters

$37B

spent by enterprises on generative AI in 2025, 3.2× the year before. (Menlo Ventures, December 2025)

95%

of enterprise generative AI pilots show no measurable return. (MIT NANDA, The GenAI Divide, July 2025)

40%

of enterprise apps will include task-specific AI agents by the end of 2026, Gartner predicts. (Gartner, August 2025)

How it works in practice

Most teams start by pasting documents into one chat. That helps one person, in one tool, until the files go stale.

A shared knowledge base fixes that. It reads your chat, drive, tickets and email, and stays current as they change.

Both ChatGPT and Claude connect to it through MCP. Nobody has to choose one assistant for the company.

Access matters most. If the base shows everyone everything, an agent becomes a way around your rules. Each answer should follow the person who asks.

How Mithrandir does it

  • One memory for the whole company, in your own cloud, from 26 sources.

  • Works inside Claude, ChatGPT, Copilot, Cursor and 16 more agents.

  • Measured on 30,000+ entries: 8.3× lower cost, 2.2× faster, 100% sourced, 0 leaks.

Questions

Can ChatGPT and Claude use the same knowledge base?
Yes. Both connect to a company memory over MCP, so both answer from the same knowledge.
Do we need to write documentation first?
No. A company memory reads the tools your people already use, so nothing is filed twice.
Will every employee see everything?
No. Each person gets answers limited to what they can already access at work.

Remember.

Every company needs agents that know it. We are here for that.

Remember more. Spend less.

Bring one workflow. See it on your own data in 30 minutes.