Subprocessors
Last updated · 21 October 2026
Mithrandir runs in the customer’s own cloud account (AWS, Azure or GCP), set up with our Terraform. Nearly all processing happens there, under the customer’s control, and involves no one else. This page lists the parties that can receive customer data anyway. It forms part of the Data Processing Addendum.
1. Mithrandir's subprocessors
These parties process customer data on Mithrandir’s behalf, under our contract with them. Nothing reaches us from a paying customer’s Deployment except the opt-in Telemetry below.
| Entity | Purpose | Data | Region | When it applies |
|---|---|---|---|---|
| GitHub, Inc. (tier 1 usage inbox: a private repository we operate for that customer alone, with its own token) | Receives the weekly usage report, if the customer has turned tier 1 on | Metadata only: call and error counts, latencies, sizes, the Software's own LLM token usage, configuration values, version identifiers, salted seat pseudonyms (pseudonymised personal data) and the license id. Never content and never a question | United States | Off unless agreed in writing. Off switch on the Deployment: TELEMETRY_PUSH_ENABLED=false |
| GitHub, Inc. (tier 2 crash inbox: a second, separate private repository for that customer alone, with its own token) | Receives crash reports so we can diagnose failures | The exception, stack trace and failing request after on-host masking of secrets and personal data; the free text depends on the agreed content class (full, question or metadata). The seat is a salted pseudonym | United States | Nothing is sent until we issue the crash inbox credentials. Off switch: TELEMETRY_CRASH_PUSH_ENABLED=false |
| Google Workspace (support email at help@trymithrandir.com, and Google Meet for screen-share) | Support | Only what the customer chooses to send or show in a support case. Never sent automatically | As stated in Google's live documentation | Only when the customer writes to us or opens a call. Support data is deleted when the case closes |
| Slack (a shared channel, only where one is agreed) | Support by chat | Only what the customer chooses to post in that channel. Never sent automatically | As stated in Slack's live documentation | Only where a shared Slack channel is agreed |
We keep usage reports for 13 months and crash reports for 90 days, and delete every Telemetry document from a customer’s Deployment within 30 days after its agreement ends. GitHub is used here only as the host of the inbox repositories. When a customer connects its own GitHub organization as a source, that is the customer’s own vendor.
A free pilot that chose to run on our OpenAI API account also sends content to OpenAI as our subprocessor. That option is never offered to paying customers.
2. Providers you choose yourself
Paying customers bring their own LLM keys. The provider works under the customer’s own contract with it and is the customer’s processor, never ours. The customer sets exactly one provider, and there is no failover between them. The decision model runs self-hosted in the customer’s own account. Embeddings are computed locally and never reach a provider, and the evaluation (“mirror”) calls are off for paying customers.
| Entity | Purpose | Data | Region | When it applies |
|---|---|---|---|---|
| OpenAI | Bridging, extraction and answer synthesis | Fragments of the customer's code, tickets, messages, documents and meeting text, and the asking person's question | Set by the customer's own account and contract | Only with LLM_PROVIDER=openai |
| Anthropic | As for OpenAI | As for OpenAI. Never image contents | Set by the customer's own account and contract | Only with LLM_PROVIDER=anthropic |
| Google, Gemini API | As for OpenAI | As for OpenAI. Never image contents | Set by the customer's own account and contract | Only with LLM_PROVIDER=gemini |
| Amazon Web Services, Amazon Bedrock | As for OpenAI, on the Bedrock model the customer configures | As for OpenAI, signed with the customer's own AWS role | The customer's own AWS account and region | Only with LLM_PROVIDER=bedrock |
| None: a self-hosted model | The same calls, to a model server the customer runs | The same content, which reaches no third party if the server runs on infrastructure the customer controls | The customer's own infrastructure | Only with LLM_PROVIDER=local |
3. Recipients that are not subprocessors
- The AI client’s own model vendor. An agent (MCP) tool result goes back to the AI client that asked for it, and that client sends it to its own model vendor. The customer picks the client.
- The customer’s source systems (for example GitHub, Atlassian, Linear, Google Workspace, Slack), connected under the customer’s own contracts.
- Destinations the customer configures: its identity provider, its SIEM, its secrets manager.
- Neo4j’s usage report, a default of the Neo4j database (metadata only), turned off in the compose file shipped from v0.1.17.
- Docker Hub, for the public
neo4jimage, at install and upgrade only.
4. Providers behind this website and billing
For this website, call bookings and online subscriptions, Mithrandir is the controller of the information collected, as the Privacy Policy describes. These providers process it for us:
| Entity | Purpose | Data | Region | When it applies |
|---|---|---|---|---|
| Stripe | Checkout, subscription billing, receipts and the billing portal | Billing name, email, payment details, the plan and number of Seats | As stated in Stripe's live documentation | When you subscribe |
| Supabase | Database for booking records, subscriptions and issued licenses | Name and email, subscription status and Seats, the licenses issued | As stated in Supabase's live documentation | When you book a call or subscribe |
| Resend | Sends the email that delivers each License File | Your email address and the license email | As stated in Resend's live documentation | When a License File is issued |
| Vercel | Hosts and serves this website, including the endpoints that renew License Files and sign you in to the image registry | Requests to the website, including IP addresses | As stated in Vercel's live documentation | When you visit the website |
| Cal.com | The calendar you book a call on | Your name and email, and what you enter when booking | As stated in Cal.com's live documentation | When you book a call |
5. Changes
We notify each customer in writing at least 30 days before we add or replace a subprocessor in section 1, and update this page in the same change. A customer that objects can turn off both Telemetry tiers on its Deployment, which removes both GitHub inboxes with no action from us. A new LLM provider supported by the Software changes nothing for a customer until it chooses that provider.