Privacy

Cost metadata, not customer conversations.

SpendLens is designed to calculate usage and classify workloads without collecting full user or customer prompts.

What SpendLens stores

Operational fields include project, provider, model, endpoint, token counts, latency, cost, function name, file name, line number, prompt sampling mode, and permitted custom metadata.

What SpendLens does not store

The SDK does not send or store complete message arrays or full user/customer prompt content. Do not place personal, confidential, or prompt content in endpoint names or custom metadata.

Prompt sampling modes

template_only (default)

May send only system instructions or a reusable prompt template for workload classification.

off

Sends operational metadata only and stores no prompt text.

redacted_sample

Reserved for optional redacted sampling. Use only after reviewing your privacy requirements.

Classification and recommendations

Classification runs once per stable workload and is reused. Prompt Waste Signals are deterministic checks over metadata and approved template samples; they do not automatically rewrite prompts.

For personal-information handling, retention, access, and complaints, read the Privacy Policy.