Penny

Subprocessors

Version 1.0 · effective September 25, 2026

Mintry gives at least 30 days' notice of a new or replacement subprocessor by updating this list and emailing the members of every team (DPA, section 7.2).

Mintry uses the following subprocessors to provide Penny. "Customer Content" means the messages, mail, files, code, database results and derived memory of a team (Mintry acts as processor). "Account data" means members' account, billing and security data (Mintry acts as controller; these providers are Mintry's processors for it).

1. Infrastructure and service providers

SubprocessorServiceDataLocation of processingTransfer safeguard
Google Cloud (Google Cloud EMEA Ltd / Google LLC)Hosting: GKE (application), Cloud SQL (PostgreSQL database), Cloud Storage (files, uploaded exports), Secret Manager (connector credentials), Cloud Logging, Cloud NATAll Customer Content and Account dataeurope-west1 (St. Ghislain, Belgium) for data at restCloud Data Processing Addendum; SCCs; EU-US DPF
Temporal Technologies, Inc. (Temporal Cloud)Durable workflow orchestration for sync, backfill, extraction and digestsWorkflow inputs, results and historyTemporal Cloud namespace on GCP europe-west3 (Frankfurt); company in the US (support access)Temporal DPA; SCCs
OpenRouter, Inc.Routing of AI model requests; usage and cost reportingExcerpts of Customer Content sent in prompts, and the responses; request metadataUnited States (EU in-region routing is available on OpenRouter's Business/Enterprise plans)OpenRouter's data processing terms; SCCs
Stripe (Stripe Payments Europe Ltd / Stripe, Inc.)Payments, saved cards, invoices, Stripe Tax, Customer PortalAccount data only: billing email, address, tax id, card data (held by Stripe), paymentsEU and USStripe DPA and Data Transfer Addendum; SCCs; EU-US DPF
Resend (Plus Five Five, Inc.)Transactional email: magic links, invitations, low-balance and billing notices, digests by emailAccount data; Customer Content in digest emailsUnited States (sending may use eu-west-1, but account data and logs stay in the US)Resend DPA; SCCs; EU-US DPF

2. AI model providers (reached through OpenRouter)

Penny sends Customer Content to these models only through OpenRouter, with the settings in section 3. The provider that serves an open-weight model is chosen by OpenRouter's routing unless pinned; section 3 recommends pinning.

Model (as configured)Used forProviderLocationProvider's data terms (API)
anthropic/claude-sonnet-5Answers (Ask agent), extraction, digestsAnthropic, PBC (or, if routed, Google Vertex AI / Amazon Bedrock)USCommercial Terms: no training on customer content; inputs and outputs deleted within 30 days by default (flagged content up to 2 years); ZDR by agreement
google/gemini-3.1-flash-liteClassification and gating (secret check, mail gate, extraction gate)Google (Vertex AI or Gemini API paid tier)US / globalPaid services: prompts and responses not used to improve products; logs kept for a limited time for abuse detection. Free tier must never be used (human review, training; and not allowed for EEA/UK/CH users)
baai/bge-m3Embeddings (search index)An inference provider chosen by OpenRouter among zero-data-retention providersVaries by providerZero data retention through OpenRouter
cohere/rerank-v3.5Re-ranking search resultsCohereVaries by providerZero data retention through OpenRouter
qwen/qwen3-reranker-8bRe-ranking (alternative)An inference provider chosen by OpenRouter among zero-data-retention providersVaries by providerZero data retention through OpenRouter

3. How AI requests are configured

Mintry's OpenRouter account has prompt and response logging off and the use of inputs and outputs for training off, with zero data retention enforced account-wide. Every request asks for zero data retention, denies data collection, and is limited to named providers, so a fallback never leaves that list. Free (:free) model variants are never used.

4. Planned (not in use during pilots)

SubprocessorServiceWhen
UnipileGmail access for gmail.com users through an aggregator with its own CASAOnly if Google verification and CASA are not complete by public launch (umbrella spec). Its brand appears on the Google consent screen.

5. Not subprocessors: the sources you connect, and your own agents

These services are where the Customer's data comes from or where Penny posts on the Customer's instruction. The Customer chooses and contracts with them directly; they process data under their own terms as independent controllers (or as the Customer's processors), not on Mintry's behalf. Mintry does not list them as subprocessors, and the DPA's subprocessor-change notice does not apply to them.

ServiceWhy it is not a subprocessor
Slack (Slack Technologies)The Customer's workspace. Penny reads it with the Customer's authorisation and posts answers and digests into it. Slack holds the data under the Customer's own Slack agreement.
Telegram (Telegram Messenger Inc. / Telegram FZ-LLC)The chat platform. Penny's bot receives messages from groups and channels it is added to and posts into them. Telegram processes that under its own terms. Uploaded Telegram Desktop exports go straight to Penny (Google Cloud).
Google as Gmail providerThe member's mailbox. Distinct from Google Cloud and Google's AI models above, which are subprocessors.
GitHubThe Customer's repositories, read through the Penny Memory GitHub App.
The Customer's database hostThe Customer's own infrastructure.
AI assistants connected by MCP (Claude, ChatGPT, Cursor and others)Connected and configured by the Customer under its own agreement with that vendor.