How ModelGenius Handles Your Data
When you use ModelGenius, your data passes through several steps — uploading your traces, running validations, sending prompts to model providers, and viewing your results. This page explains, in plain language, what we store, how we use it, and the choices you have at each step.
This page is provided for transparency and is updated as the Service evolves. It is an overview, not a contract: our Privacy Policy and Terms of Service govern.
The short version
- We store your workload data by default so we can power our validation infrastructure and your workspace, keep your past reports available, and let you re-run validations without re-uploading. You can switch this off with a 30-day retention setting.
- We never train a foundational model on your data, and we don't let third-party providers do so either — unless you explicitly opt in.
- You control which models your prompts are validated against — nothing is replayed to a candidate provider you haven't approved. To operate the platform itself (like analyzing and categorizing your traces), we use the infrastructure providers listed in our documentation, under no-train terms.
- By default, validation prompts are routed through OpenRouter only to model providers with a zero-data-retention policy. For three providers — OpenAI, Anthropic, and Google — we hold direct accounts and route to them directly instead, which avoids intermediary routing fees and speeds up your validations; for those three, the provider's standard no-train retention policy applies.
- We store de-identified categorizations and request metadata to power our reporting and validation infrastructure. These never contain the content of your prompts or responses.
What data we collect
Within the ModelGenius product, we collect and store the following types of data:
User profile data. The information tied to your account — like name, email, organization, billing details, and your settings.
Prompts and responses. The traces you upload for validation and the model outputs generated during a run.
Conversational data. Your conversations in the Service's conversational interface.
Input categorizations. ModelGenius creates categorizations of your prompts. You can think of them as higher-level groupings that help us improve our discovery, validation, and monitoring infrastructure over time. Your data is de-identified before categorization, and the categorization is performed by a model operating under no-train terms — it is never used to train that provider's models. A categorization does not include the content of your prompt and is designed so your prompt cannot be read from it.
Metadata. We store metadata for each request — for example, the number of prompt and completion tokens, latency, and which model was used. This metadata does not include the content of your prompts or responses, only information about the request itself.
How we use your data
We do not train a foundational model on your data, and we do not allow a third-party provider to do so — unless you explicitly opt in.
Your user profile data, prompts and responses, and conversational data are used only to provide the services to you, like running your validations, powering your workspace, and keeping your reports available.
Input categorizations and metadata may be used to deliver and improve the services, and to power our validation infrastructure and reporting.
How we store your data
We treat your workload data (prompts, responses, reports, and conversations) and your account data (profile and billing information) differently, because they play different roles:
- Workload data is encrypted per customer, covered by the retention controls below, and crypto-shredded when you delete your account.
- Account data is encrypted at rest and deleted when you delete your account; we retain minimal records of the deletion itself, plus records we're required to keep (like billing and tax records).
We store your prompts and responses, and your conversational data, so we can provide the service to you — this is what lets us power your workspace, give you access to your previous validation reports, and let you re-run a validation without re-uploading your traces, which saves you the cost of redundant replays.
Your retention choice
The 30-day retention setting applies to your workload data. Account data stays for as long as your account exists.
| Default | 30-day retention (opt-in, in Settings) | |
|---|---|---|
| Prompts & responses | Stored until you delete them | Deleted after 30 days |
| Conversational data | Stored until you delete them | Deleted after 30 days |
| Input categorizations | Always retained | Always retained |
| Metadata | Always retained | Always retained |
By default, we retain your workload data so your workspace and history stay available. If you'd prefer a tighter policy, you can turn on the 30-day retention setting under Settings. With it enabled, your prompts, responses, and conversational data are deleted from our servers 30 days after they're created.
Turning this on may affect parts of your experience — for example, runs older than 30 days will expire and their reports will no longer be available. Input categorizations and metadata are always retained, even under the 30-day setting.
Need your data deleted?
You can delete your account at any time from your Settings page, which removes your account data, prompts and responses, and conversational data. Deletion of your workload data is backed by crypto-shredding: the encryption keys protecting it are destroyed, so even a stray copy is unreadable. Deleted and expired data is also purged from our backups, currently within 30 days.
De-identified input categorizations and metadata are retained; when you delete your account, the link between them and your account is severed, so they are no longer connected to you in a way that identifies you.
How your data reaches model providers
ModelGenius uses AI model providers in two ways, and we're careful about which providers see your data in each.
To operate the platform. To analyze and categorize your traces and build your validation plan, ModelGenius processes your de-identified data using the infrastructure providers we use for those functions (a current list is available on request), under no-train terms.
To validate your prompts. You control which candidate models and host providers are in scope for your validations — nothing is replayed to a candidate provider you haven't approved.
Zero-data-retention by default for validation routing. Your prompts are sent to candidate model providers (via OpenRouter) that operate under a zero-data-retention policy — meaning they don't store your prompts or responses. We will only route to a provider with a different retention policy if you explicitly opt in. In all cases, we do not allow providers to train on your data unless you explicitly opt in.
Three direct connections, for cost and validation speed. For a small set of established, trusted providers — OpenAI, Anthropic, and Google (Gemini via Vertex) — we hold our own accounts and connect directly rather than through OpenRouter. That avoids the routing intermediary's service fee (a saving we pass on to you) and improves validation speed. For these direct connections, each provider's standard no-train policy applies: OpenAI and Anthropic may retain prompts for up to 30 days; Google (Gemini via Vertex) does not retain your prompts, except content flagged for abuse review, which may be retained for up to 90 days. All three operate under enterprise data terms that prohibit training on your data.
Security
Your data is protected in transit and at rest:
- Encryption in transit. All data moving between you and ModelGenius, and between ModelGenius and model providers, is encrypted using HTTPS/TLS. This covers uploading your traces, streaming results, and viewing your reports.
- Encryption at rest, with per-customer keys. Your stored workload data is encrypted using AES-256, with a separate encryption key per customer, managed in AWS KMS.
- Account isolation. Your data is isolated to your account and is not accessible to other customers.
More information on how we handle security is available on request.
Questions
If you have questions about how your data is handled, contact us at team@modelgenius.ai or see our Privacy Policy and Terms of Service.