A knowledge base that refuses to guess
Upload markdown, text, HTML or PDF. Mamori retrieves the closest passages and answers from those alone. When nothing matches closely enough it says so and files the question for you to read later.
Mamori守 knowledge, guarded
Answers straight from your documents.
Nothing invented.
Mamori reads the files you give it and answers your members inside Discord. When the answer is not in your knowledge base it says so, logs the question, and moves on. No guessing, and no bill you did not agree to.
- Free plan, no card, no AI spend
- Hard monthly cap per server
- Answers cite their source
RI
rina11:04
What is the refund window on the annual plan?
Mamori11:04
Fourteen days from the renewal date. After that the plan runs to the end of the term and does not auto-renew again.
KA
kaito11:06
Can I transfer my licence to another company?
Mamori11:06
That is not in this server's knowledge base, so I will not answer it. I have saved the question for your admins to look at.
What it does
Three things, done properly.
Mamori is deliberately small. It answers from your documents, it sets up a server, and it never spends more than the plan you picked.
One command to shape a server
Run setup, pick a template, and Mamori shows the exact channels, roles and permissions it would create. Nothing is applied until you confirm the preview.
Spending you can actually cap
Every plan buys a fixed number of AI replies per month. There is no overage billing, no metered surprise, and one switch that stops every AI call at once.
How it works
Live in about ten minutes.
Invite and choose channels
Add Mamori, then name the channels it may answer in. It stays quiet everywhere else.
Feed it your documents
Upload files into collections. Mamori splits them into passages and indexes them for retrieval.
Members ask, Mamori cites
Answers arrive with the document they came from, so anyone can check the source in a second.
Each arc is one limit. A request has to clear all four before a single token is spent.
Cost safety
Four limits, all of them hard.
Runaway model bills are why bots like this quietly disappear. Mamori treats spend as a product constraint rather than something to watch nervously in a dashboard.
Per-server monthly quota
Each plan buys a fixed number of AI replies. At the limit Mamori stops answering until the 1st. It cannot spend past the number you picked.
Per-user daily cap
Twenty five AI messages per person per day, plus a short cooldown between requests, so one member cannot drain a whole month in an afternoon.
Semantic cache
Repeat questions are matched against answers already given. A close enough match is served from storage and costs nothing to run.
Global kill switch
One switch pauses every AI call across every server. The knowledge base, setup tools and the rest of the commands keep working.
Everything that is not AI stays available whatever happens to the budget, including on the free plan.
Put it in your server and see what it does not know.
Start on the free plan. Add documents when you are ready, and turn on AI replies only if you want them.