Terms and Policies
This page describes how the platform treats a customer's data and credentials, as documented across the rest of this site. It is a technical summary, not a contract. The binding agreements are listed at the foot of the page.
What crosses the boundary
Nothing reaches the platform that is not sent to it by an explicit call.
| Sent to the platform | When |
|---|---|
| Policy text | Once, when a policy set is saved. Never with a check. |
| The content being checked | On each checkText or checkImage call. |
| An API key | On every call, to authenticate. |
A compiled policy configures the service ahead of time. It is not re-sent per request, so the per-request payload is the content under check and nothing else. See Deployment architecture.
What each side owns
The split is documented in full on Deployment Model.
The customer owns applications and user workflows, AI gateway logic, the model, business rules applied after a verdict, and human review queues. The model is included in that list: the platform is called by a customer's gateway and never sits in front of it.
The platform owns authentication with project-scoped API keys, policy storage and compilation, text and image inference, usage and credit accounting, and the SDK and API contracts.
Credentials
The credential is one API key. It belongs to one project, which scopes the policy sets it can reach, and it is the only secret involved. Only a hash of it is stored, so it is shown once, when it is created. See Projects.
Handling obligations: keep API keys in a secret manager, never expose one in browser code, make policy management calls server-side, use a separate project and key per environment, and revoke a key that has been logged, leaked, or shared with an unauthorized system. Revoking a key stops it working at once.
Audit and traceability
A check returns its verdict and, when asked for, a reason. It does not return a request identifier, so keep your own record of the content, the policy set name and the verdict if you need an audit trail. The platform keeps per-hour counts of checks, violations and tokens per project and policy set, which are shown in the dashboard.
On-premises and data residency
Deployment shape decides where data sits, and two of the three shapes keep it on the customer's premises:
- Industrial monitoring retains nothing in the cloud. Cameras, compute, storage and the operator console all stay on site, and the dashboard reads the local store. See Industrial Safety.
- Robotics runs inference on the device. Everything deciding whether an action may execute is on the robot; the cloud is reached asynchronously and never gates an actuator. See Robotics Safety.
Binding agreements
The documents that govern use of the platform — terms of service, privacy policy, and the data processing agreement — are provided as part of a commercial agreement rather than published here. Ask your Chidakashi contact for the current versions.
Where this page and a signed agreement disagree, the agreement governs.