Cookbook

Task-oriented recipes that compose DVARA's governance features into scenario-level workflows. Each recipe answers "how do I accomplish X" — not "what is X" (see Core Concepts), not "how do I call an OpenAI SDK" (see Integrations), not "what does feature Y do" (see Features or Governance & Safety).

Recipes are deliberately short: problem statement, prerequisites, the few steps that actually compose features together, verification, and links out to the feature pages for depth. The dvarahq/dvara-examples repository is the canonical source for full runnable code — recipes on this page link into it where relevant.

When to use the Cookbook

You want to …	Read this
Understand the platform	Core Concepts
Run a single curl against a provider	Provider Setup
Wire an SDK at the client side	Integrations
Understand what feature Y does in detail	Features or Governance & Safety
Solve a specific cross-feature problem	This cookbook

Available recipes

Onboarding

• Add DVARA to an existing OpenAI-SDK app — retrofit governance onto an existing codebase without rewriting client code

Rollouts

• Canary a new model version and roll back on error rate — test a new model on a small traffic slice, monitor, promote or revert

FinOps

• Attribute cost per tenant for chargeback — set up per-tenant API keys, tag usage, and generate a monthly chargeback report

Streaming

• Stream chat completions end-to-end — working SSE consumers (Python / Node.js / Java), partial-JSON accumulation for json_schema streams, and a reconnect strategy that survives idle disconnects

Need a recipe that isn't here? Open an issue on the docs repo — we'd rather write the recipe you actually have than the one we guessed at.