Subprocessors
Every company that touches your data. Two are live and one is listed only because the code contains it.
1What this page is
A subprocessor is a company we use that may process data on our behalf. This is the current list, kept deliberately short.
We will update this page before a new subprocessor starts processing anything.
2Current list
As of the effective date at the top of this page.
- Cloudflare, Inc. Hosting, compute, database, object storage, key value storage, vector storage and AI inference. Processes everything the service stores. Global network.
- Discord, Inc. The platform the bot runs on and the identity provider for website sign-in. Processes your Discord account identifiers and the list of servers you administer. United States.
- Stripe, Inc. Payment processing. Not active. No data reaches Stripe today.
3Cloudflare
Cloudflare provides Workers, D1, R2, KV, Vectorize and Workers AI. Uploaded files, extracted passages, embeddings, usage counters, cost logs and website sessions all live in Cloudflare products.
Inference and embeddings run on Cloudflare Workers AI using @cf/meta/llama-3.1-8b-instruct-fp8-fast and @cf/baai/bge-base-en-v1.5.
Cloudflare processes this data as our processor under its own data processing addendum.
4Discord
Discord delivers the commands, the responses and the OAuth sign-in flow. Anything typed into a Discord channel is processed by Discord under its own terms and privacy policy before it ever reaches us.
We receive your Discord user id, display name, avatar hash, and the ids of servers where you hold Manage Server or Administrator.
5Stripe
Stripe is listed for transparency because an unused integration exists in the codebase. It is switched off, the webhook endpoint returns a not configured response, and no cardholder data passes through Mamori.
If card payments are enabled, this page and the Refund and Cancellation Policy will be updated before the first charge.
6What the bot actually uses
Everything below runs inside Cloudflare. Nothing in this list sends your documents or your questions to a third party model provider.
Models can change if a provider retires one. When that happens this page is updated before the change ships.
- Answers are written by @cf/meta/llama-3.1-8b-instruct-fp8-fast on Cloudflare Workers AI, with a hard ceiling of 400 output tokens per answer.
- Search is powered by @cf/baai/bge-base-en-v1.5, which turns each passage into a 768 number vector so the closest passages can be found.
- Text is pulled out of PDF and HTML uploads by Cloudflare's document conversion. That step uses no model at all, and it has no OCR, so a scanned image PDF indexes as empty.
- Vectors are stored in Cloudflare Vectorize, isolated per server and filtered on the server id on every single query.
- Uploaded files are stored in Cloudflare R2. Passage text, usage counters and logs are stored in Cloudflare D1. Short lived counters and the global pause switch are stored in Cloudflare KV.
- The optional gateway bot runs on Node.js on a separate server. It only relays messages. It never calls a model, never touches the database, and cannot spend anything on your behalf.
7Your content is never used for training
We do not train, fine tune or evaluate any model on your documents, your questions or your answers.
Retrieved passages are sent to the model only to answer the question that was just asked, and are not retained by the model afterwards.
8No AI subprocessors
There is no OpenAI, Anthropic, Google, Mistral, Cohere or other external model provider in this product. Every model call goes to Cloudflare Workers AI.
Your documents and your questions do not leave Cloudflare in order to be answered. This is a deliberate design choice, not an accident of the current setup.
9Changes to this list
Where we can, we will give at least 30 days notice before adding a subprocessor, so that a customer who objects has time to raise it.
Objections go through a ticket in the support server. If we cannot resolve an objection, you can cancel and we will refund any period paid for and not used.
10Contact
Questions about this list go through a ticket in the support server.