Adapters

Adapters intercept a provider client and record every call automatically — prompts, streamed tokens, tool calls, and completions. No manual fr.record() calls needed.

All adapters support both streaming and non-streaming requests.

OpenAI

import OpenAI from "openai";
import { FlightRecorder, wrapOpenAI } from "@ai-flight-recorder/sdk";

const fr = new FlightRecorder();
const openai = wrapOpenAI(new OpenAI(), fr.recorder);

fr.startSession({ label: "chat" });

const response = await openai.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Hello" }],
});

fr.endSession();

wrapOpenAI returns a proxy that behaves identically to the real OpenAI client. Swap it in without changing any other call sites.

Anthropic

import Anthropic from "@anthropic-ai/sdk";
import {
  FlightRecorder,
  wrapAnthropic,
  type AnthropicClientLike,
} from "@ai-flight-recorder/sdk";

const fr = new FlightRecorder();
const client = wrapAnthropic(
  new Anthropic() as unknown as AnthropicClientLike,
  fr.recorder,
);

fr.startSession({ label: "claude-chat" });

const message = await client.messages.create({
  model: "claude-sonnet-4-5",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello" }],
});

fr.endSession();

Note

The as unknown as AnthropicClientLike cast is required because the real Anthropic SDK uses strict overloads that TypeScript won’t structurally assign to the adapter interface. This is safe — it only affects the type assignment.

Google Gemini

import { GoogleGenerativeAI } from "@google/generative-ai";
import {
  FlightRecorder,
  wrapGeminiModel,
  type GeminiModelLike,
} from "@ai-flight-recorder/sdk";

const fr = new FlightRecorder();
const genAI = new GoogleGenerativeAI(process.env.GOOGLE_API_KEY!);
const model = wrapGeminiModel(
  genAI.getGenerativeModel({ model: "gemini-1.5-pro" }) as unknown as GeminiModelLike,
  fr.recorder,
);

fr.startSession({ label: "gemini-chat" });
const result = await model.generateContent("Hello");
fr.endSession();

The same as unknown as GeminiModelLike cast applies here for the same reason.

Streaming

All three adapters handle streaming automatically. Tokens are recorded individually as token events as they arrive, and a completion event is emitted when the stream closes.

const stream = await openai.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Tell me a story" }],
  stream: true,
});

for await (const chunk of stream) {
  // consume as normal — tokens are recorded behind the scenes
}

Exported types

The following types are exported from @ai-flight-recorder/sdk for use with the cast pattern:

• AnthropicClientLike
• AnthropicMessageResponse
• GeminiModelLike