293 lines
9.7 KiB
JavaScript
293 lines
9.7 KiB
JavaScript
import { NextResponse } from 'next/server';
|
||
|
||
// 通用模型列表拉取 — 支持 OpenAI 兼容格式和 Gemini 原生格式
|
||
export async function POST(request) {
|
||
try {
|
||
const { apiKey, baseUrl, provider, embedOnly } = await request.json();
|
||
|
||
if (!apiKey) {
|
||
return NextResponse.json(
|
||
{ error: '请先填入 API Key' },
|
||
{ status: 400 }
|
||
);
|
||
}
|
||
|
||
// Gemini 原生格式
|
||
if (['gemini-native', 'custom-gemini'].includes(provider)) {
|
||
return await fetchGeminiModels(apiKey, baseUrl, embedOnly);
|
||
}
|
||
|
||
// Claude/Anthropic（多策略拉取）
|
||
if (['claude', 'custom-claude'].includes(provider)) {
|
||
return await fetchClaudeModels(apiKey, baseUrl);
|
||
}
|
||
|
||
// OpenAI 兼容格式（适用于所有其他供应商）
|
||
return await fetchOpenAIModels(apiKey, baseUrl, embedOnly);
|
||
|
||
} catch (error) {
|
||
console.error('拉取模型列表错误:', error);
|
||
return NextResponse.json(
|
||
{ error: '网络连接失败，请检查 API 地址' },
|
||
{ status: 500 }
|
||
);
|
||
}
|
||
}
|
||
|
||
// Gemini 原生格式拉取模型 — 支持分页，兼容中转
|
||
async function fetchGeminiModels(apiKey, baseUrl, embedOnly) {
|
||
const base = (baseUrl || 'https://generativelanguage.googleapis.com/v1beta').replace(/\/$/, '');
|
||
let allModels = [];
|
||
let pageToken = '';
|
||
|
||
// 循环分页拉取
|
||
do {
|
||
const url = `${base}/models?key=${apiKey}&pageSize=1000${pageToken ? `&pageToken=${pageToken}` : ''}`;
|
||
|
||
const response = await fetch(url, {
|
||
method: 'GET',
|
||
headers: { 'Content-Type': 'application/json' },
|
||
});
|
||
|
||
if (!response.ok) {
|
||
// 分页请求失败时，尝试不带 pageSize 参数（有些中转不支持）
|
||
if (allModels.length === 0) {
|
||
const fallbackUrl = `${base}/models?key=${apiKey}`;
|
||
const fallbackRes = await fetch(fallbackUrl, {
|
||
method: 'GET',
|
||
headers: { 'Content-Type': 'application/json' },
|
||
});
|
||
if (!fallbackRes.ok) {
|
||
return handleFetchError(fallbackRes);
|
||
}
|
||
const fallbackData = await fallbackRes.json();
|
||
allModels = extractModelArray(fallbackData);
|
||
break;
|
||
}
|
||
break;
|
||
}
|
||
|
||
const data = await response.json();
|
||
// 兼容不同中转返回格式：models[] 或 data[]
|
||
allModels = allModels.concat(extractModelArray(data));
|
||
pageToken = data.nextPageToken || '';
|
||
} while (pageToken);
|
||
|
||
let models = allModels;
|
||
|
||
// 过滤逻辑：有 supportedGenerationMethods 时按能力过滤，没有时全部保留
|
||
const hasCapabilityInfo = models.some(m => m.supportedGenerationMethods?.length > 0);
|
||
|
||
if (hasCapabilityInfo) {
|
||
if (embedOnly) {
|
||
models = models.filter(m => m.supportedGenerationMethods?.includes('embedContent'));
|
||
} else {
|
||
models = models.filter(m =>
|
||
!m.supportedGenerationMethods ||
|
||
m.supportedGenerationMethods.includes('generateContent') ||
|
||
m.supportedGenerationMethods.includes('embedContent')
|
||
);
|
||
}
|
||
} else if (embedOnly) {
|
||
// 无能力信息时，按名称匹配嵌入模型
|
||
models = models.filter(m => {
|
||
const id = (m.name || m.id || '').toLowerCase();
|
||
return /embed|text-embed/i.test(id);
|
||
});
|
||
}
|
||
|
||
models = models.map(m => ({
|
||
id: (m.name?.replace('models/', '') || m.id || m.name || '').trim(),
|
||
displayName: m.displayName || m.display_name || m.name?.replace('models/', '') || m.id || '',
|
||
}))
|
||
.filter(m => m.id) // 过滤掉空 ID
|
||
.sort((a, b) => a.id.localeCompare(b.id));
|
||
|
||
return NextResponse.json({ models });
|
||
}
|
||
|
||
// 从不同格式的响应中提取模型数组
|
||
function extractModelArray(data) {
|
||
// Gemini 原生格式: { models: [...] }
|
||
if (Array.isArray(data.models)) return data.models;
|
||
// OpenAI 兼容格式: { data: [...] }
|
||
if (Array.isArray(data.data)) return data.data;
|
||
// 直接是数组
|
||
if (Array.isArray(data)) return data;
|
||
return [];
|
||
}
|
||
|
||
// OpenAI 兼容格式拉取模型（/v1/models）
|
||
async function fetchOpenAIModels(apiKey, baseUrl, embedOnly) {
|
||
const base = (baseUrl || '').replace(/\/$/, '');
|
||
if (!base) {
|
||
return NextResponse.json(
|
||
{ error: '请先填写 API 地址' },
|
||
{ status: 400 }
|
||
);
|
||
}
|
||
|
||
const url = `${base}/models`;
|
||
|
||
const response = await fetch(url, {
|
||
method: 'GET',
|
||
headers: {
|
||
'Content-Type': 'application/json',
|
||
'Authorization': `Bearer ${apiKey}`,
|
||
},
|
||
});
|
||
|
||
if (!response.ok) {
|
||
return handleFetchError(response);
|
||
}
|
||
|
||
const data = await response.json();
|
||
let models = (data.data || []);
|
||
|
||
if (embedOnly) {
|
||
// 匹配常见嵌入模型：embed*, bge-*, bce-*, e5-*, gte-* 等
|
||
models = models.filter(m => /embed|\/bge[-_]|\/bce[-_]|\/e5[-_]|\/gte[-_]|text-embedding/i.test(m.id));
|
||
}
|
||
|
||
models = models.map(m => ({
|
||
id: m.id,
|
||
displayName: m.id,
|
||
}))
|
||
.sort((a, b) => a.id.localeCompare(b.id));
|
||
|
||
return NextResponse.json({ models });
|
||
}
|
||
|
||
async function handleFetchError(response) {
|
||
const errorText = await response.text();
|
||
if (response.status === 401 || response.status === 403) {
|
||
return NextResponse.json(
|
||
{ error: 'API Key 无效或无权限' },
|
||
{ status: 401 }
|
||
);
|
||
}
|
||
let errMsg = `拉取失败(${response.status})`;
|
||
try {
|
||
const errObj = JSON.parse(errorText);
|
||
errMsg = errObj?.error?.message || errMsg;
|
||
} catch { /* ignore */ }
|
||
return NextResponse.json(
|
||
{ error: errMsg },
|
||
{ status: response.status }
|
||
);
|
||
}
|
||
|
||
// Claude/Anthropic 模型列表 — 多策略拉取
|
||
// 策略1: Anthropic 原生 API（直连 api.anthropic.com）
|
||
// 策略2: OpenAI 兼容格式（中转/代理）
|
||
// 策略3: 预定义列表（兜底）
|
||
async function fetchClaudeModels(apiKey, baseUrl) {
|
||
const base = (baseUrl || 'https://api.anthropic.com').replace(/\/$/, '');
|
||
|
||
// 策略1: 尝试 Anthropic 原生 /v1/models
|
||
if (apiKey) {
|
||
try {
|
||
const models = await tryAnthropicNativeModels(apiKey, base);
|
||
if (models.length > 0) {
|
||
return NextResponse.json({ models });
|
||
}
|
||
} catch (err) {
|
||
console.warn('Claude native API failed:', err.message);
|
||
}
|
||
|
||
// 策略2: 尝试 OpenAI 兼容格式 /v1/models（很多中转用这种格式）
|
||
try {
|
||
const models = await tryOpenAICompatModels(apiKey, base);
|
||
if (models.length > 0) {
|
||
return NextResponse.json({ models });
|
||
}
|
||
} catch (err) {
|
||
console.warn('Claude OpenAI-compat API failed:', err.message);
|
||
}
|
||
}
|
||
|
||
// 策略3: 预定义列表
|
||
const models = [
|
||
{ id: 'claude-opus-4-20250514', displayName: 'Claude Opus 4' },
|
||
{ id: 'claude-sonnet-4-20250514', displayName: 'Claude Sonnet 4' },
|
||
{ id: 'claude-3-7-sonnet-20250219', displayName: 'Claude 3.7 Sonnet' },
|
||
{ id: 'claude-3-5-haiku-20241022', displayName: 'Claude 3.5 Haiku' },
|
||
{ id: 'claude-3-5-sonnet-20241022', displayName: 'Claude 3.5 Sonnet v2' },
|
||
{ id: 'claude-3-5-sonnet-20240620', displayName: 'Claude 3.5 Sonnet' },
|
||
{ id: 'claude-3-opus-20240229', displayName: 'Claude 3 Opus' },
|
||
{ id: 'claude-3-haiku-20240307', displayName: 'Claude 3 Haiku' },
|
||
];
|
||
return NextResponse.json({ models });
|
||
}
|
||
|
||
// Anthropic 原生格式：x-api-key + anthropic-version header
|
||
async function tryAnthropicNativeModels(apiKey, base) {
|
||
let allModels = [];
|
||
let hasMore = true;
|
||
let afterId = null;
|
||
|
||
while (hasMore) {
|
||
let url = `${base}/v1/models?limit=100`;
|
||
if (afterId) url += `&after_id=${afterId}`;
|
||
|
||
const response = await fetch(url, {
|
||
method: 'GET',
|
||
headers: {
|
||
'x-api-key': apiKey,
|
||
'anthropic-version': '2023-06-01',
|
||
'Content-Type': 'application/json',
|
||
},
|
||
});
|
||
|
||
if (!response.ok) return [];
|
||
|
||
const data = await response.json();
|
||
const pageModels = data.data || [];
|
||
allModels = allModels.concat(pageModels);
|
||
|
||
hasMore = data.has_more === true;
|
||
if (hasMore && pageModels.length > 0) {
|
||
afterId = pageModels[pageModels.length - 1].id;
|
||
} else {
|
||
hasMore = false;
|
||
}
|
||
}
|
||
|
||
return allModels.map(m => ({
|
||
id: m.id,
|
||
displayName: m.display_name || m.id,
|
||
})).sort((a, b) => a.id.localeCompare(b.id));
|
||
}
|
||
|
||
// OpenAI 兼容格式：Bearer token + /v1/models（多数中转使用此格式）
|
||
async function tryOpenAICompatModels(apiKey, base) {
|
||
// 尝试 /v1/models 和 /models 两种路径
|
||
for (const path of ['/v1/models', '/models']) {
|
||
try {
|
||
const url = `${base}${path}`;
|
||
const response = await fetch(url, {
|
||
method: 'GET',
|
||
headers: {
|
||
'Content-Type': 'application/json',
|
||
'Authorization': `Bearer ${apiKey}`,
|
||
},
|
||
});
|
||
|
||
if (!response.ok) continue;
|
||
|
||
const data = await response.json();
|
||
const rawModels = data.data || [];
|
||
if (rawModels.length === 0) continue;
|
||
|
||
return rawModels.map(m => ({
|
||
id: m.id,
|
||
displayName: m.id,
|
||
})).sort((a, b) => a.id.localeCompare(b.id));
|
||
} catch {
|
||
continue;
|
||
}
|
||
}
|
||
|
||
return [];
|
||
}
|