Privacy Policy

1. Scope and roles

This policy applies to Elektric's websites, dashboard, APIs, support, and business operations. Elektric acts as a business/controller for account administration, billing, security, support, website operation, and its own records. For Customer Content submitted by a business customer, Elektric generally acts as processor/service provider under that customer's instructions; the customer determines what it submits and is responsible for notices and legal basis. The DPA governs where applicable.

2. Information and sources

• Account and identity: name, email, Google OAuth identifiers/tokens, verification, and session data.
• Credentials and access: ownership, access roles, API-key name/prefix/hash/status, and customer-supplied opaque user/conversation IDs.
• Billing: account membership, payer email, amounts, status, Stripe references, ledger entries, costs, credits, limits, and disputes. Stripe, not Elektric, collects full card details.
• Customer Content: prompts, messages, tool data/results, Outputs, Conversations, Memories, files, generated assets, and multimodal/realtime content.
• Operational data: request IDs, timestamps, IP/network and request data available to the edge service, tokens, latency, status, provider/model and routing data, context flags, cost, project/key identifiers, limits, job/asset events, and security records.
• Communications: feedback, support, legal requests, and correspondence.

Sources include Customer, Authorized Users, Customer Applications/end users, identity and payment providers, model providers, Cloudflare infrastructure, and service operation.

3. Uses

Elektric uses information to provide and route requests; authenticate and authorize; operate enabled Conversation, Memory; process Files, Assets and jobs; calculate usage and billing; prevent fraud and abuse; enforce limits; troubleshoot, monitor and secure; communicate; support customers; comply with law; and maintain records. Operational metadata and submitted feedback may be analyzed to evaluate reliability and routing.

No training on customer content

Elektric does not use customer content to train models. This includes prompts, responses, uploaded documents, Conversation history, Memory and other customer-provided content.

Relevant content is still processed to provide the service. Upstream providers operate under their own applicable terms; Elektric’s policy does not establish a no-training or zero-retention guarantee for those providers.

4. Feature-specific retention

Default request processing

Requests with user and conversation IDs can use Conversation and Memory automatically. Request content is processed to fulfill the request and may be transmitted to model and supporting processing providers. Operational records are stored. Uploaded files and generated Assets have separate lifecycles.

Conversation

For requests using Conversation, non-streaming requests with a conversationId store the latest submitted user message and successful assistant text. A supplied userId associates the thread with that user within the key’s account scope. User content is saved before inference, so it may remain after an unsuccessful request. Submitted attachment blocks can be part of stored messages. A bounded subset of history is used in later requests. Conversation history is retained until you delete it using the Conversation deletion controls; it does not expire automatically.

Memory

When Memory is enabled and a stable userId is supplied, Elektric reads selected user information and can update it after successful non-streaming responses. Summaries, categories, confidence, provenance and timestamps may persist across Conversations. Relevant excerpts from the user's previously stored Conversations may also be retrieved. Deleting a Memory card marks it inactive; its summary and provenance remain stored. Turning off Conversation does not stop Memory from reading existing history when Memory remains enabled.

Document Understanding, Files and Assets

Document Understanding analyzes request attachments and does not automatically add them to Knowledge. Uploads are stored privately with a one-hour access expiry and Project-scoped metadata. Inline chat attachments do not create temporary-upload records. PDFs are forwarded natively to models; other supported formats are decoded or converted to text. Physical deletion is asynchronous, active-job Asset pins can delay scheduled cleanup, and metadata can remain. Expiring or deleting an upload does not erase content already stored in Conversation, derived Memory, results or provider records.

Project Logs and saved results

Operational records include Request IDs, supplied user/conversation IDs, timestamps, provider/model, status, usage, latency, costs, fallback and Context activity. Ordinary request/attempt tables do not store full chat bodies. Knowledge diagnostics can store effective retrieval query text and source names, locations and IDs. Saved Context Playground runs separately retain questions, answers, diagnostics and feedback. The routing Playground also stores recent turns in browser local storage during a session and attempts to clear them and temporary session data when you leave. Browser close/cleanup delivery is not guaranteed; this is separate from saved Context Playground runs. Covered raw telemetry older than 90 days is scheduled for deletion; that schedule does not cover every aggregate, saved result, infrastructure log or provider copy.

Deletion and remaining records

Conversation deletion removes one thread and its messages, not derived Memory, uploads or saved results. Project deletion blocks access and revokes keys, then permanently removes Conversation, all Memory cards, managed Knowledge indexes and metadata, project files, saved results and scoped telemetry. Account deletion runs that cleanup across all owned projects and removes authentication, profile and account records. Projects owned by other users are preserved. Cleanup failures remain pending and retry automatically; completion is recorded only after the managed cleanup succeeds.

Account deletion retains selected payment and paid/refund ledger records under a pseudonymous HMAC for seven years (implemented as 7 × 365 days), followed by scheduled deletion. Transaction references remain; these records are not anonymous. Deletion receipts have no implemented expiry. No general automatic expiry is implemented for current Memory cards or saved Context Playground runs. Backup and provider-copy deletion periods are not established here.

See Context and the Privacy Policy for technical details.

5. Recipients

Answering providers receive the messages, supported attachments and enabled Context needed for a request; eligible fallback may contact another provider. Routing classifiers, document conversion and Memory updates can also transmit relevant customer content to supporting services.

Elektric discloses relevant information to infrastructure, AI/model, identity, payment, and communication providers on the Subprocessor List; authorized account members; professional advisers and transaction counterparties under restrictions; and authorities or others where required by law or needed to protect rights, safety, and the Service. Routing/fallback may expose a request to more than one model provider.

6. Cookies and technologies

Elektric uses cookies or similar storage required for authentication and sessions. The repository shows no general third-party advertising or website analytics SDK.

7. Legal bases and transfers

Where GDPR/UK GDPR applies, anticipated bases include contract performance, legitimate interests in operating/securing the Service, legal obligations, and consent where required. Customer determines the basis for Customer Content it controls. Information may be processed in the United States and other places where subprocessors operate.

8. U.S. state disclosures

Subject to applicable law, individuals may have rights to access/know, correct, delete, or obtain a copy of personal information; opt out of certain sale, sharing, targeted advertising, or profiling; appeal; and receive nondiscriminatory treatment. Customer Content requests should normally go to the relevant business customer.

Potential California categories include identifiers; customer records; commercial information; internet/network activity; professional information supplied for business accounts; approximate geolocation inferred from network data; and sensitive information if Customer submits it. Recipients are the provider categories above. Elektric does not use sensitive information to infer characteristics for its own purposes.

9. Rights, children, and changes

Individuals may use account controls or contact Elektric to exercise applicable rights. Elektric may verify identity/authority, apply lawful limits, and direct processor-data requests to the controlling customer. Authorized agents need sufficient authorization; appeals use the confirmed privacy channel. The Service is business-oriented and not directed to children. Elektric may update this policy with notice as required.