pi-ai 统一多模型 API

import {
  createModels,
  fauxAssistantMessage,
  fauxProvider,
  fauxText,
  fauxThinking,
  fauxToolCall,
  Type,
  type Context,
  type Tool,
} from "@earendil-works/pi-ai";

const faux = fauxProvider();
const models = createModels();
models.setProvider(faux.provider);
const model = faux.getModel();

const tools: Tool[] = [
  {
    name: "get_time",
    description: "获取当前时间",
    parameters: Type.Object({ timezone: Type.Optional(Type.String({ description: "时区，如 Asia/Shanghai" })) }),
  },
];

// Context 是纯数据：系统提示词、消息、工具，可以直接用 JSON 序列化
const context: Context = {
  systemPrompt: "你是一个简洁的助手。",
  messages: [{ role: "user", content: "现在几点？", timestamp: Date.now() }],
  tools,
};

faux.setResponses([
  fauxAssistantMessage([fauxThinking("需要调用 get_time。"), fauxToolCall("get_time", { timezone: "Asia/Shanghai" })], {
    stopReason: "toolUse",
  }),
  fauxAssistantMessage([fauxText("现在是北京时间下午 3 点。")]),
]);

// 第 1 次请求：流式消费统一事件
const s = models.stream(model, context);
for await (const event of s) {
  if (event.type === "thinking_delta") process.stdout.write(`[thinking] ${event.delta}\n`);
  if (event.type === "toolcall_end") console.log("toolcall_end:", event.toolCall.name, event.toolCall.arguments);
  if (event.type === "done") console.log("done:", event.reason);
}
const first = await s.result();
context.messages.push(first);

// 执行工具，把结果作为 toolResult 消息追加
for (const call of first.content.filter((b) => b.type === "toolCall")) {
  context.messages.push({
    role: "toolResult",
    toolCallId: call.id,
    toolName: call.name,
    content: [{ type: "text", text: "15:00" }],
    isError: false,
    timestamp: Date.now(),
  });
}

// 第 2 次请求：非流式
const second = await models.complete(model, context);
context.messages.push(second);
console.log("回答：", second.content.filter((b) => b.type === "text").map((b) => b.text).join(""));
console.log("usage:", second.usage.input, "in /", second.usage.output, "out");

// Context 可以序列化保存，之后换任意模型继续
const restored: Context = JSON.parse(JSON.stringify(context));
console.log("序列化后消息数：", restored.messages.length);

[thinking] 需要调用 get_time。
toolcall_end: get_time { timezone: 'Asia/Shanghai' }
done: toolUse
回答： 现在是北京时间下午 3 点。
usage: 70 in / 4 out
序列化后消息数： 4

import { createModels, type Context } from "@earendil-works/pi-ai";
import { anthropicProvider } from "@earendil-works/pi-ai/providers/anthropic";
import { openaiProvider } from "@earendil-works/pi-ai/providers/openai";

// 按需注册提供方：每个提供方都是一个独立的子路径，打包时只会引入用到的那些
const models = createModels();
models.setProvider(anthropicProvider()); // 鉴权：环境变量 ANTHROPIC_API_KEY、已存储的凭据，或 OAuth
models.setProvider(openaiProvider()); // 鉴权：OPENAI_API_KEY 等

const claude = models.getModel("anthropic", "claude-opus-5-5");
if (!claude) throw new Error("模型不在目录里");

const context: Context = {
  systemPrompt: "你是一个资深后端工程师。",
  messages: [{ role: "user", content: "解释一下什么是幂等性。", timestamp: Date.now() }],
};

// completeSimple 加上统一的 reasoning 档位，各家的推理参数由 pi-ai 负责翻译
const reply = await models.completeSimple(claude, context, { reasoning: "medium" });
for (const block of reply.content) {
  if (block.type === "thinking") console.log("[思考]", block.thinking);
  if (block.type === "text") console.log(block.text);
}
console.log(`费用：$${reply.usage.cost.total.toFixed(4)}`);

import { StringEnum, Type, type Tool } from "@earendil-works/pi-ai";

const weatherTool: Tool = {
  name: "get_weather",
  description: "查询某地当前天气",
  parameters: Type.Object({
    location: Type.String({ description: "城市名或坐标" }),
    units: StringEnum(["celsius", "fahrenheit"], { default: "celsius" }), // 兼容 Google，见下方说明
  }),
};

// 可选：让提供方在服务端强制参数符合 schema。prefer 表示尽量强制，不支持时退回普通工具调用
const strictTool: Tool = {
  name: "edit_file",
  description: "编辑文件",
  parameters: Type.Object({ path: Type.String(), content: Type.String() }, { additionalProperties: false }),
  constrainedSampling: { type: "json_schema", strict: "prefer" },
};

console.log(weatherTool.name, strictTool.name);

import { createModels, type Context } from "@earendil-works/pi-ai";
import { anthropicProvider } from "@earendil-works/pi-ai/providers/anthropic";
import { deepseekProvider } from "@earendil-works/pi-ai/providers/deepseek";
import { openaiProvider } from "@earendil-works/pi-ai/providers/openai";

const models = createModels();
models.setProvider(anthropicProvider());
models.setProvider(openaiProvider());
models.setProvider(deepseekProvider());

const context: Context = { messages: [] };

// 先用 Claude 推理
const claude = models.getModel("anthropic", "claude-opus-5-5")!;
context.messages.push({ role: "user", content: "25 * 18 等于多少？", timestamp: Date.now() });
context.messages.push(await models.completeSimple(claude, context, { reasoning: "medium" }));

// 换成 GPT 来复核：它会看到 Claude 的 thinking，被转换成带 <thinking> 标签的文本
const gpt = models.getModel("openai", "gpt-5.6-sol")!;
context.messages.push({ role: "user", content: "这个计算对吗？", timestamp: Date.now() });
context.messages.push(await models.complete(gpt, context));

// 再换成 DeepSeek 总结
const ds = models.getModel("deepseek", "deepseek-v4-pro")!;
context.messages.push({ role: "user", content: "用一句话总结上面的对话。", timestamp: Date.now() });
const summary = await models.complete(ds, context);
console.log(summary.content);