@agentrewind/codec-openai

OpenAI-compatible Chat Completions codec for AgentRewind.

Use this package when your agent calls the OpenAI Node SDK, including OpenAI-compatible providers configured with a custom baseURL.

Install

Most users get this codec through the SDK:

npm install @agentrewind/sdk

Use this package directly only when a library needs granular dependency boundaries.

Supported Client Shape

The codec wraps:

• client.chat.completions.create(request)
• client.chat.completions.stream(request)

That is the method shape provided by the OpenAI Node SDK. For an OpenAI-compatible provider, keep using the OpenAI SDK and set baseURL:

import { OpenAI } from "@agentrewind/sdk";

const client = new OpenAI({
  apiKey: process.env.COMPATIBLE_API_KEY,
  baseURL: "https://your-provider.example/v1"
});

Recording

import { AgentRewind, OpenAI, assertProviderClient, defineHarness, openaiChatCodec } from "@agentrewind/sdk";
import type { ChatCompletion, ChatCompletionChunk } from "@agentrewind/sdk";

const client = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY
});

const codec = openaiChatCodec();
assertProviderClient(client, codec);

const harness = defineHarness(async (ctx) => {
  const completion = await ctx.model.create<ChatCompletion>(
    {
      model: "gpt-5.5",
      messages: [
        { role: "system", content: "Answer tersely." },
        { role: "user", content: `Request ${ctx.uuid()}` }
      ],
      temperature: 0
    },
    { site: "answer-question" }
  );

  return completion.choices[0]?.message.content ?? "";
});

const recorded = await AgentRewind.recordRun(
  {
    id: "openai-demo",
    store: ".rewind",
    model: client,
    codec
  },
  harness
);

Replay

Strict replay needs no live model client:

const replayed = await AgentRewind.replayRun(
  recorded.path,
  { codec: openaiChatCodec() },
  harness
);

Streaming

ctx.model.stream() calls client.chat.completions.stream() under the hood:

assertProviderClient(client, codec, ["stream"]);

await session.run(async (ctx) => {
  for await (const chunk of ctx.model.stream<ChatCompletionChunk>(
    {
      model: "gpt-5.5",
      messages: [{ role: "user", content: "Stream one sentence." }]
    },
    { site: "stream-answer" }
  )) {
    process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
  }
});

Notes

• metadata is treated as volatile and stripped from strict request fingerprints.
• Token usage maps prompt_tokens to inputTokens and completion_tokens to outputTokens.
• The codec preserves raw responses in the normalized response for local replay fidelity.
• This package targets Chat Completions. It does not wrap the Responses API, Assistants API, or transport-level fetch.