The questions we'd ask too.

You can, and many teams start there. It costs you twice. Every question pays for everything the agent pulls in, about eight times more per answer than asking Mithrandir on our own demo. And those tools only hold what is live today: the reasoning behind a decision made two years ago by someone who has left is in none of them. Mithrandir keeps it, ties it to the result it explains, and hands each person only what they may see.

No. Engineering was the first wedge because the pain is sharpest there and the value shows in a week. The same ontology holds a sales discount, a contract clause, a hiring decision and the code that implements them, each tied to the reason it exists and each visible only to the people allowed. One seat price covers the whole company.

Connect the sources where your decisions already live: Slack, tickets, code, docs, transcripts. Mithrandir builds the memory from your history, not from a blank page, so the first useful answers come back the same week. Your agents reach it over MCP, so nobody changes the tool they work in.

No, and that's the point. Every knowledge tool that depends on someone filing a doc eventually goes stale, because the filing is the first thing to get dropped. Mithrandir reads the places where decisions already happen (threads, tickets, reviews, meeting transcripts), so nobody writes the same thing down twice, and nobody has to remember to. The answers then come back inside the tools your team already works in.

Then nothing can recover it, and we won't pretend otherwise. But most reasoning is written down; it just lives in a thread, a ticket comment, a review, a transcript, and then gets archived. Mithrandir catches it where it already happens and keeps it, so the answer is there the next time someone asks.

The new one becomes the answer and the old one stays on record with its dates. An amended contract supersedes the clause it replaced, and the replaced terms still answer for the period they covered. When someone flags a reason as wrong, the correction becomes current and the old reason is retired: kept for the history, never served as current again. Nothing is silently overwritten, and nothing old comes back dressed as new.

Mithrandir mirrors each tool's own permissions rather than inventing new ones, so a person sees in their agent what they could already open in the source. When someone leaves a channel or is removed from a tool, access follows, within minutes. If permission data ever goes stale, the affected content goes dark rather than leaking.

We don't train models on customer data. Mithrandir ships self-hosted: it deploys inside your own VPC or on your own hardware, and it can run against a model you host yourself, so restricted content never has to leave your infrastructure. If your organization has hard requirements beyond that, bring them to the first call.

Thirty minutes with the founders, not a scripted sales demo. Bring one workflow where decisions and context keep getting lost. We show Mithrandir on a real codebase, talk through which of your sources it would connect to and how it would sit in your infrastructure, and answer your security questions directly. There is no form to fill in first.