--- title: "Start here — Norcaster without an ML background" description: "Plain-language guide to quality checks, knowledge lookup, templates, and when Norcaster is worth adopting." canonical: https://www.norcaster.com/docs/start --- New here # Using Norcaster without an ML background Norcaster is the **AI Runtime Control Plane** for production AI—still easy to read as a control layer: safety rules, document lookup, execution timelines, and automated quality checks. You can be a strong application developer and still use it effectively—the product labels advanced terms like *evals* and *retrieval* beside plain English in the UI. ## Choose your quickstart path ### Try in 2 minutes Use the fast first run to ask a test question with demo defaults and see how enforcement appears in the portal. ### Deploy in 15 minutes Connect a provider, create a deployment, add a policy, run one request, then inspect the trace and enforcement events. ### Self-host in your VPC Run the API, portal, orchestrator, database, and supporting services in your own environment when data boundaries require it. ### Create your first workflow Start with input → MCP → redact\_text → model → output, or use retrieval → answer\_with\_citations for evidence-backed responses. Keep privacy steps before model inference. ## What “quality checks” (evals) are Saved test questions and scoring that tell you if answers got worse after you changed a prompt, model, or document set—similar to unit tests for software, applied to AI outputs. ## What “knowledge lookup” (retrieval) is Searching your own uploaded or synced documents to ground answers, often with source references (citations). It is how “chat over my PDFs” works without training a custom model. ## What templates are for Pre-built outcomes—safer chat, document Q&A, approval workflows—so you start from something that already runs instead of wiring every step from scratch. ## When Norcaster is worth it Helpful when you have real users or production traffic, need visibility into what the model did, or must enforce safety and budgets consistently. For a one-off throwaway experiment with no risk, it may be early—that is normal; the same product is there when your app grows. ## Common first decisions - Use [provider integrations](https://www.norcaster.com/docs/providers) when you want Norcaster to control and route Anthropic, OpenAI, Azure OpenAI, Bedrock, Voyage AI, or similar model calls. - Use [workflows](https://www.norcaster.com/docs/workflows) when a request needs data fetching, redaction, retrieval, approvals, or tool calls before the model. - Use [workflow examples](https://www.norcaster.com/docs/workflow-examples) when you want copyable node chains for common use cases. - Use [policies](https://www.norcaster.com/docs/policies) when you need consistent block, warn, budget, or audit behavior across requests. - Use [security and compliance](https://www.norcaster.com/docs/security-compliance) when you need audit exports, MFA, SSO/OIDC, GDPR rights, or signed evidence.