Deployment Model
The platform should keep a clear boundary between customer-owned systems and Chidakashi platform services.
Customer-owned systems
Customers own:
- Applications and user workflows.
- AI gateway logic, when present.
- Business rules after receiving a verdict.
- Human review queues.
- Model routing if they use their own model provider.
Platform-owned services
The platform owns:
- Authentication with project-scoped API keys.
- Policy storage and compilation.
- Text safety inference.
- Image safety inference.
- Usage and credit accounting.
- SDK and API contracts.
Prompt and response checks
For LLM use cases, the safety layer can be applied before and after a model call:
- Check the user prompt.
- If allowed, send the prompt to the model.
- Check the model response.
- Return, block, redact, or escalate the response.
This keeps safety decisions close to the model boundary while allowing the customer application to remain the system of record for business workflow.