This page is not a request to “trust us.” It exposes what can be verified: what we store, what we do not store, and how customers can check the boundary themselves.
We return the upstream model as it is. Key fields such as id and model keep upstream format instead of showing one model name in marketing and silently routing another one in the backend.
Successful text model prompts and responses are not stored as regular persistent records. Video generation keeps the job ID, status, required task parameters, result URLs, and billing fields needed for polling, download, and reconciliation.
Infwave handles gateway routing, authentication, billing, and visibility. Requests are forwarded to the selected upstream provider; training usage is governed by upstream terms, not a second Infwave training pipeline.
Review the public schema snapshot, export your own usage and billing records, and compare response model fields. Contract customers can request deeper read-only due diligence under NDA.
The point is to state what traces the routing and billing layer actually leaves, not to hide behind vague security language.
A commitment that customers cannot verify is closer to a slogan than a mechanism.
Pick a real call and compare the returned id, model, timestamps, and upstream format.
The public path is /transparency/schema.json. Search for prompt / response body fields and review the highlighted tables below.
After logging in, export records by model, date, and API key, then reconcile token usage, video task cost, rates, and timestamps.
The most important part to write down is often the exception.
The billing log (logs) retains each call's source IP, timestamp, model name and token counts for billing, abuse-prevention and troubleshooting. These are metadata and do not include prompt or response bodies; access is admin-limited.
Video generation is asynchronous. We need task IDs, status, required parameters, billing fields, and result URLs to support polling, download, and reconciliation. Result URL validity and availability follow the returned URL and upstream policy; deletion or privacy requests should be sent from your registered email.
The public paths are /transparency and /transparency/schema.json. This is not a real-time public database console; it is a durable public asset for customer-verifiable boundaries.
The complete snapshot is published as JSON. This page expands the tables customers most often ask about.
Loading public schema snapshot summary...
The current snapshot covers customer-reachable application tables. Below are the ones customers most often ask about.
logs — usage & billing log billing summary only, no bodiesThe content field stores only a billing summary (e.g. ratio / token recalculation notes), not prompt or response bodies; it also keeps token counts, model name, timestamp and source IP.
tokens — API key metadata key metadata onlyusers — account identity password hashedquota_data — usage aggregates numbers onlyuser_subscriptions — subscription & quota cycle quota numbers onlysubscription_orders — top-up orders no promptstasks — async tasks (video generation) includes task params / promptAsync tasks such as video generation retain task parameters (including the generation prompt), status and result URL for polling, download and reconciliation. This is required for async tasks and differs from text completion's successful-request body path.
If public claims and real behavior diverge, that divergence itself is a transparency issue that should be fixed first.
[SECURITY]When a public page is not enough, we can provide more detailed read-only verification materials under an NDA or commercial framework.