Introduction
About Inferada
Inferada is a managed AI gateway from Servada. Your application uses one base URL and one bearer token; the gateway routes requests and applies the controls configured for your organisation.
Use Inferada to give applications a consistent interface to models and supporting AI services. Teams can change providers, add locally hosted models, and apply shared policy without rebuilding those concerns in every client.
What Inferada handles
- Model and service routing. Configure locally hosted and external upstreams, then address the resulting models and services through the same API. Clients do not need provider-specific credentials or request code.
- PII detection and redaction. Detect personal data directly, or replace detected values with placeholders before an external model call and restore matching placeholders in the response. Detection is best-effort; see the PII documentation for its limits.
- Request policy. Processors can reject a request when they detect PII, a suspected prompt-injection attempt, or a route that is not classified as local. They provide defence in depth, not a substitute for application-level security controls.
- Tool execution. Attach gateway-managed tools to a model and let Inferada execute the calls, or send your own function definitions and execute the returned calls in your client.
- Usage controls and audit data. Attribute requests to organisations, users, and tokens; apply rate, concurrency, and spending limits; and inspect request history in the portal.
- Managed applications and agents. Provision supported applications or run scheduled and webhook-triggered TypeScript agents with their own credentials and usage attribution.
- Organisation boundaries. Models, upstream credentials, tools, limits, applications, and request history are configured in an organisation context.
Request flow
When it helps
- You need one application interface for models from several providers, including local deployments.
- Several teams need the same authentication, policy, usage, and audit controls.
- Some workloads must use locally classified upstreams, while others may use external providers.
- Your application needs configurable PII handling, prompt-injection checks, or tool execution around a model call.
- You want to provision supported chat or workflow applications without operating each application stack yourself.