SDK Bridges

Convert messages and supply lightweight generation functions to framework-neutral Core APIs.

Most applications should execute a Prompt through a full Crux adapter. Use the lower-level bridges on this page when migrating existing SDK messages or supplying generation to a framework-neutral Core utility.

Convert Messages

Each adapter converts between its SDK format and Crux's canonical Message type:

import { fromMessages, toMessages } from "@use-crux/ai";

const messages = toMessages(coreMessages);
const sdkMessages = fromMessages(messages);

import { fromMessages, toMessages } from "@use-crux/openai";

const messages = toMessages(chatMessages);
const sdkMessages = fromMessages(messages);

import { fromMessages, toMessages } from "@use-crux/google";

const messages = toMessages(contents);
const sdkMessages = fromMessages(messages);

import { fromMessages, toMessages } from "@use-crux/anthropic";

const messages = toMessages(anthropicMessages);
const sdkMessages = fromMessages(messages);

Use these helpers when integrating an existing chat UI or migrating from raw SDK calls. New Crux Prompt definitions normally use system, prompt, or messages and let the adapter handle conversion.

Supply Framework-Neutral Generation

Core utilities such as judge and retrieval recipes accept GenerateObjectFn or GenerateTextFn instead of depending on one SDK. Each adapter provides compatible wrappers.

import { generateObjectFn, generateTextFn } from "@use-crux/ai";
import { judge } from "@use-crux/core/scoring";

const evaluator = judge({
  id: "helpfulness",
  criteria: "Does the answer resolve the request?",
  scale: { min: 1, max: 5 },
  generate: generateObjectFn,
  model,
});

import { createGenerateObjectFn, createGenerateTextFn } from "@use-crux/openai";

const generateObject = createGenerateObjectFn(client);
const generateText = createGenerateTextFn(client, "gpt-4o");

await generateObject({ model: "gpt-4o", prompt, schema });

import { createGenerateObjectFn, createGenerateTextFn } from "@use-crux/google";

const generateObject = createGenerateObjectFn(client);
const generateText = createGenerateTextFn(client, "gemini-2.5-flash");

await generateObject({ model: "gemini-2.5-flash", prompt, schema });

import {
  createGenerateObjectFn,
  createGenerateTextFn,
} from "@use-crux/anthropic";

const generateObject = createGenerateObjectFn(client);
const generateText = createGenerateTextFn(client, "claude-haiku-4-5-20251001");

await generateObject({
  model: "claude-sonnet-4-5-20250929",
  prompt,
  schema,
});

These are lightweight provider-native wrappers. They do not run Prompt resolution, validation retry, Safety, tools, memory capture, or instrumentation. Use full adapter generate() for normal Prompt execution.

The Vercel AI SDK adapter exports pre-bound functions because the model is passed per call. OpenAI, Google, and Anthropic object factories bind the client; their text factories bind both client and model.

Managed generation results are finalized before hooks.onGenerate runs. The hook and caller see the same generation.call trace and span identity; provider response identity remains separate. See Operation Result Correlation.

SDK Bridges

Convert Messages

Supply Framework-Neutral Generation

Next Steps

Prompt Execution

Prompt Reference

Adapter Reference

On this page