Spendline is a proxy in the path of your AI traffic. That position comes with an obligation to be precise about data handling, so this page is written against the code, not against marketing copy. It states exactly what we retain, what we strip, what we forward, and what we encrypt.
The short version: Spendline stores call metadata and one-way fingerprints, and does not log prompt or completion text. For failed calls it keeps the provider's error message, which can quote part of the request. Your prompts and your model's responses pass through Spendline in memory. There is no setting that turns prompt logging on. What we keep is the accounting: tokens, cost, latency and attribution.
Request content exists inside Spendline only for the lifetime of the request. That is long enough to count tokens, apply your budget and policy rules, and forward it. When the response is delivered, the content is gone. Spendline does not keep a request log of message content, and there is no optional prompt-logging feature. The exception is a failed call: the provider's error message is kept, and some providers quote part of the request in it (see Outcome metadata below).
Each supported request routed through Spendline writes one append-only ledger row containing operational metadata:
The ledger is append-only: in normal operation rows are not edited, and corrections are recorded as new adjustment rows. Every row is isolated to your tenant. Records are kept while your account is active; you can ask us to delete your data, as described in our Privacy Policy.
Before your request is forwarded, Spendline removes its own credentials. The Spendline API key headers your application sends us never reach your model provider. Your provider receives your request authenticated with your provider key, nothing more.
Your request content is forwarded to the provider you chose. Spendline may rename or drop a request parameter a model does not accept (for example, the field that sets the maximum output tokens, or temperature on models that reject it); it does not change your messages. The other exception is a cross-provider reroute you have explicitly configured, where the request is translated format-to-format in memory. Forwarding happens over TLS. Your provider key is used solely to authenticate the request with your provider.
Spendline's job is financial: attribution, budgets and cost records. All of that is computable from metadata. Storing your content would add risk without adding a capability, so we don't.
Every claim on this page corresponds to a specific code path in the proxy. If your security team wants to validate any of it, whether that is the ledger schema, how keys are encrypted, or how headers are stripped before forwarding, contact us at fida@spendline.ai and we will walk them through it directly.