---
title: "Architecture"
description: "How Norcaster fits into production AI systems: request lifecycle, managed vs self-hosted deployment, data flow, and storage boundaries."
canonical: https://www.norcaster.com/docs/architecture
---
<!-- Generated at build time from the rendered page by portal/scripts/generate-llms.mjs. The page is the source of truth; edit it, not this file. -->

Architecture

# Where Norcaster sits in production

Norcaster is the **AI Runtime Control Plane**: a runtime layer in the request path between your application and AI models, tools, and knowledge sources, plus the management surface where you configure deployments, policies, providers, evals, and observability. Run it managed or self-hosted when your data boundary requires it.

## Request lifecycle

1. Your application calls Norcaster over HTTPS with JSON.
2. Norcaster resolves the deployment, workflow, provider model, and policy configuration.
3. Contracts, policies, redaction, and budget gates run before provider inference.
4. Allowed requests are sent to the configured model, tool, retrieval source, or workflow node.
5. Norcaster stores execution metadata, traces, policy decisions, and audit evidence.

## Data flow

```
App
  -> Norcaster API / Workflow runtime
  -> Contract, policy, redaction, budget checks
  -> Provider model, retrieval source, or MCP tool
  -> Trace, audit event, model output
```

In workflow use cases, each node writes structured output to the execution payload. Downstream nodes read named fields such as `question`, `raw_tickets.text`, `retrieval`, or `model_request`.

## Managed vs self-hosted

### Managed

Norcaster hosts the control plane. You configure providers, deployments, workflows, policies, users, and audit exports through the portal and API.

### Self-hosted

You run the API, portal, orchestrator, database, and supporting infrastructure in your own environment. This is the right path when data residency and network boundaries are strict.

[Full self-hosted setup guide →](https://www.norcaster.com/docs/self-hosted-setup)

## What is stored

- Organizations, teams, projects, deployments, providers, and workflow definitions.
- Execution records, node executions, trace IDs, timing, status, and error payloads.
- Policy decisions, enforcement events, budget outcomes, and audit export metadata.
- Provider credentials are encrypted before storage.

[Learn how to debug with traces and run history →](https://www.norcaster.com/docs/observability)