Trust & AI data use

Updated November 3, 2026

NetConverter is operated by ValeronLabs LLC. This page describes our AI data-use policy, the service providers identified in our implementation, and the verification limits that matter when you assess the service.

AI data-use statement

NetConverter does not use proprietary customer configurations to train third-party AI models. Service improvement uses abstracted translation patterns, rather than full customer configurations. Model inference to process a job is separate from model training.

Our policy prohibits sending full raw customer configurations to third-party models by default. It requires summaries and counts by default; tokenized repair excerpts require zero-retention and no-training terms, and full-fidelity assistance requires a separate tenant opt-in. Credentials and secrets must not be sent to models. Moonshot (Kimi) is restricted to research and challenge work with no customer-derived content, including tokenized content.

Verification limit: that policy is not a certification of every backend path. The backend payload audit remains open. Some legacy summaries contain customer object names, and some advisory calls contain redacted output excerpts. Redaction is not the same as complete tokenization or anonymization. We have not yet published verified account-specific zero-retention or regional-processing guarantees. Do not infer those guarantees from this page or from a provider’s general API terms.

What AI calls receive

Structural guidance: typed structural summaries on the paths that implement that contract.
Legacy feature-mapping review (CP2): feature counts, mapping context, knowledge-base references and customer-derived identifiers such as object and zone names.
Legacy final advisory (CP4c): statistics, findings, knowledge-base context and sampled output with address and secret redaction. Names may remain.
Other backend calls: classification is incomplete. An unclassified evidence value means the payload has not been classified; it does not mean that no customer data was sent.
Your connected assistant: MCP results are sent to the assistant provider you choose, under your own account’s terms. Hosted view settings and terminal exposure levels determine the returned content. These conversation-retention settings are separate from NetConverter’s backend provider accounts. See privacy and data flow.

Service-provider inventory

This implementation inventory is not a completed contractual subprocessor register. Account-specific processing regions, retention settings and applicable agreements still require verification. Anthropic, OpenAI and Gemini credentials are configured in the application runtime, verified November 3, 2026 without reading their values. Configuration does not establish that a provider handled any particular job.

Amazon Web Services: hosted compute, application database, storage and delivery; may process uploaded configurations, job results and operational records. The application runtime reports us-east-1 (US East, Northern Virginia), verified November 3, 2026. This is not an end-to-end residency guarantee: delivery, backups and each other provider require their own scope verification.
Neo4j Aura: managed per-job analysis graphs; customer-derived network relationships. Instance region and contractual retention remain to be confirmed.
Clerk: account authentication and sessions; account and authentication data. See the authentication disclosure in our Privacy Policy. Account region and retention terms remain to be confirmed.
Anthropic: backend inference; customer-derived summaries and excerpts on applicable paths. Published retention policy describes standard retention and exceptions; it does not prove NetConverter has a zero-retention agreement. Account, model and feature eligibility remain unverified.
OpenAI: backend fallback inference and embeddings; content depends on the calling path. Published API data controls distinguish training, abuse-monitoring retention and endpoint storage. Account approval and enabled controls remain unverified.
Google Gemini: backend fallback inference where enabled; content depends on the calling path. Paid-service status, applicable API terms, retention exceptions and processing location remain to be confirmed.
Moonshot (Kimi): research and challenge capability. Our policy excludes all customer-derived content; this entry is not authorization to send customer data.
Website services: Google Tag Manager, Google Analytics, advertising tags and font delivery are described separately in the Privacy Policy. They handle browser and website-event data; they are separate from backend inference.

Reading job evidence

New instrumented conversion evidence records the actual accounted provider and model, purpose and payload class without storing prompts or responses in those rows. Deterministic checkpoint timing is not an AI call. In MCP, use get_evidence with section="ai_calls" and follow next_cursor to page through these metadata rows. The coverage field states the boundary: failed request attempts and separate provider gateways are not exhaustively represented. Historical evidence is not retroactively relabeled. An empty list alone is not proof that a job sent no data to AI.

Agreements and changes

For account-specific terms, a DPA, residency requirements or provider-change notification requirements, contact NetConverter. A contractual notice period and subscription mechanism have not yet been published; this page does not promise either. We date revisions here so changes can be reviewed.