Ephemeris privacy policy

Information we handle

• Account and sign-in information: your email address, account identifiers, authentication and authorization records, and information you provide through our sign-in service. API-key records include a hash, identifying prefix, optional name, and creation, use and revocation dates.
• Forecast data: the numeric observations, timestamps or frequency, model choices, forecast horizon, quantiles, covariates and other fields you submit, together with the resulting forecasts. Supplied labels and units may contain personal or commercially sensitive information.
• Usage and billing information: request identifiers, models used, request size and shape, response status, latency, credit balances, charges, purchases, refunds and payment-provider records. Blockchain payments can include wallet addresses and transaction identifiers.
• Technical and support information: connection and security logs, browser or device information, diagnostic events and information you send when requesting support. Hosting and authentication providers may process IP addresses and cookies needed to operate and secure the service.

Only submit information you are authorized to share. Avoid including names, contact details or other personal information in numeric data and labels unless they are needed for your task.

How we use information

We process information to authenticate you, authorize connected applications, generate forecasts, describe and plot saved results, administer credits and payments, recover retryable requests, prevent abuse, diagnose failures and respond to support requests.

The forecasting backend records derived series characteristics, model selections and timing information for operational analysis and routing evaluation. These records can include frequency, series lengths, sparsity and statistical summaries. A forecast uses existing forecasting models; generating a forecast does not itself fine-tune those models.

Inference and TEE protection

Ephemeris runs forecast inference in a Trusted Execution Environment (TEE), which provides hardware-backed isolation for the inference workload. This protection applies to the inference execution environment.

Your request still passes through the Ephemeris API gateway for authentication, validation, routing and billing before inference, and the result passes back through the gateway. TEE inference does not mean that account records, payment records, gateway response caches, application logs or files saved by your client are all inside a TEE.

A TEE is a security control, not a promise of zero retention, end-to-end encryption from your device into the enclave, or protection of copies held by another application. We do not provide a per-request attestation report through the public forecast API.

Service providers and connected applications

The service uses Clerk for authentication and OAuth, Vercel for website and API hosting, database infrastructure for account and operational records, Chutes for forecasting infrastructure, Stripe for card-payment processing, and Sentry for configured error and performance monitoring. Each provider receives information needed for the part of the service it performs. Blockchain payment networks and their infrastructure providers process transactions when you choose that payment method; public-chain records are visible independently of Ephemeris.

Information may be processed outside your country by these providers. TEE inference does not establish a particular data-residency location for the service as a whole.

When you connect through ChatGPT, Claude or another MCP client, that application processes the information you give it and the tool arguments and results it exchanges with Ephemeris under its own policies. Ephemeris receives the tool requests sent to it, rather than automatic access to your entire conversation. Hosted description and plotting tools receive the saved artifact, including any history and labels it contains.

We may disclose information when required by applicable law or when necessary to address fraud, abuse or security incidents.

Storage and retention

Account, usage, billing, payment and security records are stored to operate the service, reconcile transactions, handle disputes and meet applicable obligations. These records do not all expire when a forecast finishes or when an API key is revoked.

For forecasts submitted with an idempotency key, the gateway stores request and key hashes and the response body to support recovery. The current replay window is 24 hours after completion. Expiry ends eligibility for replay; it is not a guarantee that the database record, response or backups have been erased at that time.

The inference backend logs derived request characteristics and performance information. When its feedback facility is enabled, it can also hold a bounded set of recent inputs and model predictions in memory for outcome evaluation. This policy does not promise zero retention of forecast information.

CLI inspection, dry-run validation, descriptions and plots run locally. Local forecast artifacts and submission receipts can contain your full input history and remain until you remove them. CLI credentials are also stored locally when you use browser sign-in. Hosted MCP returns artifacts and plots to your client; it does not create a hosted file library. The separate gateway recovery records described above may still exist.

Our Sentry configuration disables default personal-information reporting and filters common credential fields and request bodies from error events. This reduces diagnostic exposure; it is not a guarantee that every log or provider record contains no personal information.

Your choices and requests

You can revoke API keys from the dashboard and disconnect Ephemeris in your MCP client. CLI logout removes the locally saved credential; it does not revoke the server-side key. Delete local artifacts and receipts separately. Disconnecting a client or revoking a key does not erase prior usage, payments or copies already received by that client.

Contact us to request access to, correction of or deletion of personal information, or to raise a privacy concern. We may need to verify your identity. Some records may need to be retained for legal obligations, transaction reconciliation, disputes or security, and public blockchain records cannot be removed by Ephemeris. Rights and available complaint mechanisms depend on the law that applies to you.

Contact and policy updates

Operator: TensorLink Pty Ltd. Privacy contact: chris@tensor-link.com.

Describe your request without sending passwords, API keys or unnecessary sensitive data. We will review privacy concerns and explain the response and any applicable limitations. You may also contact the privacy regulator available in your jurisdiction.

We may update this policy as the service changes. The date above identifies the current version.