OpenAI Wrapping

wrapOpenAI() preserves the client shape you already use, then inserts Captar’s runtime control steps around each model request. The main benefit is adoption speed: you keep your existing OpenAI integration, but you get budget checks, policy enforcement, trace spans, and usage reconciliation without rebuilding your request layer.

const openai = captar.wrapOpenAI(new OpenAI({ apiKey }), { session });

await openai.responses.create({
  model: "gpt-4.1-mini",
  input: "Summarize this support thread.",
  max_output_tokens: 180,
});

What Captar adds

Model and retry validation against call policy
Estimated spend checks before execution
Timeout-aware provider execution
Request spans with provider, model, method, and request metadata
Spend reconciliation after the provider response returns

Good wrapping habits

Wrap the client once, near your app runtime.
Pass the active session explicitly instead of relying on hidden globals.
Keep request metadata descriptive enough to debug a bad trace later.
Use the same wrapper for jobs, API routes, and background tasks when possible.

Call policy example

policy: {
  call: {
    allowedModels: ["gpt-4.1-mini"],
    maxEstimatedCostUsd: 0.35,
    maxOutputTokens: 300,
    retriesCeiling: 1,
    timeoutMs: 15_000,
  },
}

The v1 docs and SDK surface are optimized for OpenAI and OpenAI-compatible request flows. Do not treat this as a general multi-provider abstraction yet.