把用量与单价对齐,方便对照各模型实际开销。 Co-authored-by: Cursor <cursoragent@cursor.com>
276 lines
7.6 KiB
TypeScript
276 lines
7.6 KiB
TypeScript
import { readFileSync, writeFileSync } from "node:fs";
|
|
import path from "node:path";
|
|
import { randomUUID } from "node:crypto";
|
|
import { ensureUserDataDirs, getUserDataDir } from "../config/user-data-dir.js";
|
|
import {
|
|
parsePricingDocument,
|
|
type ModelRate,
|
|
} from "./pricing-parse.js";
|
|
|
|
export type PricingSource = {
|
|
id: string;
|
|
name: string;
|
|
url: string;
|
|
createdAt: string;
|
|
fetchedAt?: string;
|
|
fetchError?: string;
|
|
title?: string;
|
|
rates: ModelRate[];
|
|
};
|
|
|
|
type PricingFile = {
|
|
version: 1;
|
|
sources: PricingSource[];
|
|
};
|
|
|
|
const FILE_NAME = "pricing-sources.json";
|
|
const FETCH_TIMEOUT_MS = 20_000;
|
|
const MAX_BYTES = 2_000_000;
|
|
|
|
export const SUGGESTED_PRICING_DOCS: Array<{ name: string; url: string }> = [
|
|
{
|
|
name: "DeepSeek Models & Pricing",
|
|
url: "https://api-docs.deepseek.com/quick_start/pricing",
|
|
},
|
|
{
|
|
name: "DeepSeek 模型与价格",
|
|
url: "https://api-docs.deepseek.com/zh-cn/quick_start/pricing",
|
|
},
|
|
{
|
|
name: "智谱 GLM API 定价",
|
|
url: "https://docs.bigmodel.cn/cn/guide/start/pricing",
|
|
},
|
|
{
|
|
name: "Z.AI GLM Pricing",
|
|
url: "https://docs.z.ai/guides/overview/pricing",
|
|
},
|
|
];
|
|
|
|
function filePath(): string {
|
|
return path.join(getUserDataDir(), FILE_NAME);
|
|
}
|
|
|
|
function emptyFile(): PricingFile {
|
|
return { version: 1, sources: [] };
|
|
}
|
|
|
|
function readFile(): PricingFile {
|
|
ensureUserDataDirs();
|
|
try {
|
|
const raw = readFileSync(filePath(), "utf8");
|
|
const parsed = JSON.parse(raw) as PricingFile;
|
|
if (parsed.version !== 1 || !Array.isArray(parsed.sources)) return emptyFile();
|
|
return parsed;
|
|
} catch {
|
|
return emptyFile();
|
|
}
|
|
}
|
|
|
|
function writeFile(data: PricingFile): void {
|
|
ensureUserDataDirs();
|
|
writeFileSync(filePath(), JSON.stringify(data, null, 2), "utf8");
|
|
}
|
|
|
|
export function listPricingSources(): PricingSource[] {
|
|
return readFile().sources;
|
|
}
|
|
|
|
export function getPricingSource(id: string): PricingSource | null {
|
|
return listPricingSources().find((s) => s.id === id) ?? null;
|
|
}
|
|
|
|
export function normalizePricingUrl(raw: string): string {
|
|
const trimmed = raw.trim();
|
|
let url: URL;
|
|
try {
|
|
url = new URL(trimmed);
|
|
} catch {
|
|
throw new Error("不是合法 URL");
|
|
}
|
|
if (url.protocol !== "http:" && url.protocol !== "https:") {
|
|
throw new Error("只支持 http/https 官方文档");
|
|
}
|
|
url.hash = "";
|
|
if (url.pathname.length > 1) {
|
|
url.pathname = url.pathname.replace(/\/+$/, "");
|
|
}
|
|
return url.toString();
|
|
}
|
|
|
|
export async function addPricingSource(input: {
|
|
url: string;
|
|
name?: string;
|
|
}): Promise<PricingSource> {
|
|
const url = normalizePricingUrl(input.url);
|
|
const data = readFile();
|
|
const existing = data.sources.find((s) => s.url === url);
|
|
if (existing) {
|
|
return refreshPricingSource(existing.id);
|
|
}
|
|
const source: PricingSource = {
|
|
id: randomUUID(),
|
|
name: input.name?.trim() || hostLabel(url),
|
|
url,
|
|
createdAt: new Date().toISOString(),
|
|
rates: [],
|
|
};
|
|
data.sources.push(source);
|
|
writeFile(data);
|
|
return refreshPricingSource(source.id);
|
|
}
|
|
|
|
export function deletePricingSource(id: string): boolean {
|
|
const data = readFile();
|
|
const next = data.sources.filter((s) => s.id !== id);
|
|
if (next.length === data.sources.length) return false;
|
|
writeFile({ version: 1, sources: next });
|
|
return true;
|
|
}
|
|
|
|
export async function refreshPricingSource(id: string): Promise<PricingSource> {
|
|
const data = readFile();
|
|
const source = data.sources.find((s) => s.id === id);
|
|
if (!source) throw new Error("价目来源不存在");
|
|
try {
|
|
const parsed = await fetchAndParsePricing(source.url);
|
|
source.fetchedAt = new Date().toISOString();
|
|
source.title = parsed.title;
|
|
source.rates = parsed.rates;
|
|
if (parsed.title && looksGenericName(source.name)) {
|
|
source.name = shortenTitle(parsed.title);
|
|
}
|
|
source.fetchError = parsed.rates.length
|
|
? undefined
|
|
: parsed.warnings[0] || "未能从文档解析出型号价格";
|
|
} catch (err) {
|
|
source.fetchedAt = new Date().toISOString();
|
|
source.fetchError = err instanceof Error ? err.message : String(err);
|
|
}
|
|
writeFile(data);
|
|
return source;
|
|
}
|
|
|
|
export type FetchedDoc = { contentType: string; text: string };
|
|
|
|
export function markdownDocFallbacks(url: string): string[] {
|
|
try {
|
|
const parsed = new URL(url);
|
|
if (parsed.protocol !== "http:" && parsed.protocol !== "https:") return [];
|
|
if (parsed.pathname.endsWith(".md")) return [];
|
|
const docPath = parsed.pathname.replace(/\/+$/, "") || "/";
|
|
parsed.hash = "";
|
|
parsed.search = "";
|
|
parsed.pathname = `${docPath}.md`;
|
|
return [parsed.toString()];
|
|
} catch {
|
|
return [];
|
|
}
|
|
}
|
|
|
|
export async function fetchAndParsePricing(url: string): Promise<{
|
|
title?: string;
|
|
rates: ModelRate[];
|
|
warnings: string[];
|
|
}> {
|
|
const candidates = [url, ...markdownDocFallbacks(url)];
|
|
let lastError = "未能从文档解析出型号价格";
|
|
let lastTitle: string | undefined;
|
|
for (const candidate of candidates) {
|
|
try {
|
|
const doc = await fetchPricingDocument(candidate);
|
|
const parsed = parsePricingDocument(doc.text, doc.contentType);
|
|
lastTitle = parsed.title ?? lastTitle;
|
|
if (parsed.rates.length) return parsed;
|
|
lastError = parsed.warnings[0] || lastError;
|
|
} catch (err) {
|
|
lastError = err instanceof Error ? err.message : String(err);
|
|
}
|
|
}
|
|
return { title: lastTitle, rates: [], warnings: [lastError] };
|
|
}
|
|
|
|
export async function fetchPricingDocument(url: string): Promise<FetchedDoc> {
|
|
const res = await fetch(url, {
|
|
redirect: "follow",
|
|
headers: {
|
|
Accept: "text/html, text/markdown, application/json, text/plain;q=0.9",
|
|
"User-Agent": "WritingAgent/0.1 (local token cost)",
|
|
},
|
|
signal: AbortSignal.timeout(FETCH_TIMEOUT_MS),
|
|
});
|
|
if (!res.ok) {
|
|
throw new Error(`拉取失败 HTTP ${res.status}`);
|
|
}
|
|
const buf = Buffer.from(await res.arrayBuffer());
|
|
if (buf.length > MAX_BYTES) {
|
|
throw new Error("文档过大");
|
|
}
|
|
const contentType = res.headers.get("content-type") ?? "text/html";
|
|
return { contentType, text: buf.toString("utf8") };
|
|
}
|
|
|
|
export type ResolvedRate = ModelRate & {
|
|
sourceId: string;
|
|
sourceName: string;
|
|
sourceUrl: string;
|
|
};
|
|
|
|
export function resolveRateCatalog(sources = listPricingSources()): {
|
|
byKey: Map<string, ResolvedRate>;
|
|
rates: ResolvedRate[];
|
|
} {
|
|
const byKey = new Map<string, ResolvedRate>();
|
|
for (const source of sources) {
|
|
for (const rate of source.rates) {
|
|
const resolved: ResolvedRate = {
|
|
...rate,
|
|
sourceId: source.id,
|
|
sourceName: source.name,
|
|
sourceUrl: source.url,
|
|
};
|
|
byKey.set(rate.model.toLowerCase(), resolved);
|
|
for (const alias of rate.aliases) {
|
|
byKey.set(alias.toLowerCase(), resolved);
|
|
}
|
|
}
|
|
}
|
|
const seen = new Set<string>();
|
|
const rates: ResolvedRate[] = [];
|
|
for (const rate of byKey.values()) {
|
|
const id = `${rate.sourceId}:${rate.model.toLowerCase()}`;
|
|
if (seen.has(id)) continue;
|
|
seen.add(id);
|
|
rates.push(rate);
|
|
}
|
|
return { byKey, rates };
|
|
}
|
|
|
|
export function findModelRate(
|
|
model: string,
|
|
catalog = resolveRateCatalog(),
|
|
): ResolvedRate | undefined {
|
|
const key = model.trim().toLowerCase();
|
|
if (!key) return undefined;
|
|
const exact = catalog.byKey.get(key);
|
|
if (exact) return exact;
|
|
const tail = key.split("/").pop();
|
|
if (tail && tail !== key) return catalog.byKey.get(tail);
|
|
return undefined;
|
|
}
|
|
|
|
function hostLabel(url: string): string {
|
|
try {
|
|
return new URL(url).hostname.replace(/^www\./, "");
|
|
} catch {
|
|
return url;
|
|
}
|
|
}
|
|
|
|
function looksGenericName(name: string): boolean {
|
|
return /^(https?:|www\.|[a-z0-9.-]+\.[a-z]{2,}$)/i.test(name);
|
|
}
|
|
|
|
function shortenTitle(title: string): string {
|
|
return title.replace(/\s*[·|]\s*.*$/, "").trim() || title;
|
|
}
|