Skip to content

🤖 Open-source, local-first alternative to Grok Bot, OpenAI dots and Claude Duet. Your own AI bots on any model: GPT, Claude, Gemini, Grok, DeepSeek, Kimi, GLM and 70+ more through one Modelflare key. Reach them in the app, WeChat or Feishu; they remember you, work in the background, run commands and use your computer. Pay per use.

fix(models): 连接 Modelflare 时启用全部对话模型

dyk98ccfd68e29f2c
diff --git a/src/app.ts b/src/app.ts
index 51cc6de..01940b3 100644
--- a/src/app.ts
+++ b/src/app.ts
@@ -109,6 +109,7 @@ export class App {
account!: AccountManager;
private notices: Array<{ at: number; level: "info" | "warn"; text: string }> = [];
private accountTimer: NodeJS.Timeout | undefined;
+ private picksTimer: NodeJS.Timeout | undefined;
private modelInfoCache: { at: number; data: Record<string, ModelInfo> } | undefined;
constructor(readonly config: AppConfig, readonly opts: AppOptions = {}) {
@@ -176,6 +177,27 @@ export class App {
return routeModel(model, this.magpie.hasProvider(own) ? own : (this.magpie.modelflareProviderId ?? MODELFLARE_PROVIDER));
}
+ /**
+ * 让每个 Modelflare 连接启用的模型等于 Modelflare 当前列出的全部对话模型(见 setModelPicks):
+ * 0.1.8 及以前连接的补上,新上架的加入,下架的移出。读不到列表时什么都不改,下次再试。
+ */
+ private async syncModelPicks(): Promise<void> {
+ const magpie = this.magpie;
+ if (!magpie || magpie.state !== "running") return;
+ const baseUrl = modelflareBaseUrl(this.config.modelflare.line);
+ const ids = [magpie.modelflareProviderId, ...magpie.botProviderIds()].filter((id): id is string => !!id);
+ for (const id of ids) {
+ const key = magpie.providerKey(id);
+ if (!key) continue;
+ try {
+ const r = await verifyModelflareKey(baseUrl, key, { line: this.config.modelflare.line });
+ if (r.ok) await magpie.setModelPicks(id, r.chatModels);
+ } catch (e) {
+ log.warn("sync model picks failed", { provider: id, message: (e as Error).message });
+ }
+ }
+ }
+
/** 该机器人请求 Modelflare 用的 Key(只在本进程内使用) */
private botKey(botId: string): string {
if (!this.magpie) return "";
@@ -202,7 +224,10 @@ export class App {
disconnect: () => this.disconnectModelflare(),
setBotKey: async (botId, key) => {
if (!this.magpie) throw new Error("外部模型端点模式下不管理 Modelflare 连接");
- await this.magpie.setBotProvider(botId, key, modelflareBaseUrl(this.config.modelflare.line), this.allowedModelflareUrls(), this.requestHeaders());
+ const baseUrl = modelflareBaseUrl(this.config.modelflare.line);
+ const r = await verifyModelflareKey(baseUrl, key, { line: this.config.modelflare.line });
+ if (!r.ok) throw new Error(r.message);
+ await this.magpie.setBotProvider(botId, key, baseUrl, this.allowedModelflareUrls(), this.requestHeaders(), r.chatModels);
},
removeBotKey: async (botId) => {
await this.magpie?.removeBotProvider(botId);
@@ -293,6 +318,12 @@ export class App {
this.seedDefaultBot();
for (const bot of this.store.listBots()) await this.startBot(bot.id);
if (!this.opts.noWeb) this.web = await startWebServer(this);
+ if (this.magpie) {
+ // 启动时和之后每小时核对一次启用的模型(新上架的加入,下架的移出)
+ void this.syncModelPicks();
+ this.picksTimer = setInterval(() => void this.syncModelPicks(), 3600_000);
+ this.picksTimer.unref();
+ }
if (this.magpie && !this.opts.noWeb) {
// 余额与低余额提醒:启动后读一次,之后每 5 分钟一次
const tick = () => {
@@ -306,6 +337,7 @@ export class App {
async stop(): Promise<void> {
if (this.accountTimer) clearInterval(this.accountTimer);
+ if (this.picksTimer) clearInterval(this.picksTimer);
this.account?.cancelLogin();
await this.web?.close();
for (const id of [...this.channels.keys()]) await this.stopBot(id);
@@ -340,7 +372,7 @@ export class App {
const baseUrl = modelflareBaseUrl(line);
const r = await verifyModelflareKey(baseUrl, apiKey.trim(), { line });
if (!r.ok) return r;
- await this.magpie.connectModelflare(apiKey, baseUrl, this.allowedModelflareUrls(), this.requestHeaders());
+ await this.magpie.connectModelflare(apiKey, baseUrl, this.allowedModelflareUrls(), this.requestHeaders(), r.chatModels);
this.modelInfoCache = undefined;
this.setModelflareLineConfig(line);
await this.runtime.listModels(true).catch(() => []);
diff --git a/src/magpie/supervisor.ts b/src/magpie/supervisor.ts
index 6c3a5da..cd19dac 100644
--- a/src/magpie/supervisor.ts
+++ b/src/magpie/supervisor.ts
@@ -366,17 +366,33 @@ export class MagpieSupervisor {
* (key 不出现在进程参数里)。重复连接 = 覆盖。baseUrl 必须是白名单里的地址。
* headers 是每个请求额外带给 Modelflare 的请求头(调用归因)。
*/
- async connectModelflare(apiKey: string, baseUrl: string, allowed: string[], headers: Record<string, string> = {}): Promise<void> {
- await this.upsertProvider(this.modelflareId(), MODELFLARE_PROVIDER, "Modelflare", apiKey, baseUrl, allowed, headers);
+ async connectModelflare(apiKey: string, baseUrl: string, allowed: string[], headers: Record<string, string> = {}, picks: string[] = []): Promise<void> {
+ await this.upsertProvider(this.modelflareId(), MODELFLARE_PROVIDER, "Modelflare", apiKey, baseUrl, allowed, headers, picks);
}
/** 机器人自己的 provider(预算 Key):id 固定为 mfb-<botId>,机器人的请求都显式发到这个 provider */
- async setBotProvider(botId: string, apiKey: string, baseUrl: string, allowed: string[], headers: Record<string, string> = {}): Promise<void> {
+ async setBotProvider(botId: string, apiKey: string, baseUrl: string, allowed: string[], headers: Record<string, string> = {}, picks: string[] = []): Promise<void> {
const id = botProviderId(botId);
- await this.upsertProvider(this.hasProvider(id) ? id : undefined, id, `Modelflare ${botId}`, apiKey, baseUrl, allowed, headers);
+ await this.upsertProvider(this.hasProvider(id) ? id : undefined, id, `Modelflare ${botId}`, apiKey, baseUrl, allowed, headers, picks);
}
- private async upsertProvider(existing: string | undefined, wantId: string, name: string, apiKey: string, baseUrl: string, allowed: string[], headers: Record<string, string>): Promise<void> {
+ /**
+ * 启用的模型(picks)。内置网关对超过 24 个模型、又没有 picks 的服务商只放出列表前 24 个,没放出的
+ * 连 "provider/model" 也不服务,所以由我们写明全部对话模型。pickedFrom 由网关自己记录;Modelflare 的
+ * 列表还有不能对话的决策模型不在 picks 里,网关因此不会自动加入新模型,由 App 定时核对(syncModelPicks)。
+ */
+ async setModelPicks(id: string, picks: string[]): Promise<void> {
+ const cur = this.modelPicks(id);
+ if (!picks.length || (cur.length === picks.length && picks.every((m) => cur.includes(m)))) return;
+ await this.saveProvider(id, { models: picks });
+ }
+
+ modelPicks(id: string): string[] {
+ const p = this.readProviderRaw(id);
+ return Array.isArray(p?.models) ? p.models.map(String) : [];
+ }
+
+ private async upsertProvider(existing: string | undefined, wantId: string, name: string, apiKey: string, baseUrl: string, allowed: string[], headers: Record<string, string>, picks: string[]): Promise<void> {
if (!allowed.includes(baseUrl)) throw new Error("不支持的 Modelflare 地址");
if (!apiKey.trim()) throw new Error("缺少 API Key");
let id = existing;
@@ -385,7 +401,7 @@ export class MagpieSupervisor {
if (add.code !== 0) throw new Error(`创建 Modelflare 连接失败:${(add.stderr || add.stdout).trim().slice(0, 300)}`);
id = this.hasProvider(wantId) ? wantId : await this.findProviderId(wantId);
}
- await this.saveProvider(id, { chat: baseUrl, key: apiKey.trim(), headers });
+ await this.saveProvider(id, { chat: baseUrl, key: apiKey.trim(), headers, ...(picks.length ? { models: picks } : {}) });
}
/** 切换线路:只改地址,key 不变(机器人自己的 provider 一起切) */
diff --git a/src/modelflare.ts b/src/modelflare.ts
index a1017c8..f80f534 100644
--- a/src/modelflare.ts
+++ b/src/modelflare.ts
@@ -90,7 +90,16 @@ export function modelflareBaseUrl(line: ModelflareLine, env: NodeJS.ProcessEnv =
export type VerifyFailure = "invalid_key" | "forbidden" | "rate_limited" | "upstream" | "network" | "bad_response";
export type VerifyResult =
- | { ok: true; models: string[] }
+ | {
+ ok: true;
+ models: string[];
+ /**
+ * 能对话的模型(Modelflare 标为 kind=text 的;旧版网关没有 kind 时为全部)。
+ * 连接时把它们全部写进内置网关的 picks:网关对超过 24 个模型、又没有 picks 的服务商
+ * 只放出列表前 24 个,而 Modelflare 的列表按字母排,后面的模型会被截掉。
+ */
+ chatModels: string[];
+ }
| { ok: false; code: VerifyFailure; status?: number; message: string; suggestLine?: ModelflareLine };
/** 校验密钥:只发一次 GET /models,不重试、不换线路。 */
@@ -120,9 +129,15 @@ export async function verifyModelflareKey(
if (res.status === 429) return { ok: false, code: "rate_limited", status: 429, message: `Modelflare 暂时不可用(HTTP 429)${retryAfter(res)}` };
if (!res.ok) return { ok: false, code: "upstream", status: res.status, message: `Modelflare 暂时不可用(HTTP ${res.status})` };
try {
- const body = (await res.json()) as { data?: Array<{ id?: unknown }> };
+ const body = (await res.json()) as { data?: Array<{ id?: unknown; kind?: unknown }> };
if (!Array.isArray(body.data)) throw new Error("no data[]");
- return { ok: true, models: body.data.map((m) => String(m.id ?? "")).filter(Boolean).sort() };
+ const rows = body.data.map((m) => ({ id: String(m.id ?? ""), kind: typeof m.kind === "string" ? m.kind : "" })).filter((m) => m.id);
+ const classified = rows.some((m) => m.kind);
+ return {
+ ok: true,
+ models: rows.map((m) => m.id).sort(),
+ chatModels: rows.filter((m) => !classified || m.kind === "text").map((m) => m.id).sort(),
+ };
} catch {
return { ok: false, code: "bad_response", status: res.status, message: "Modelflare 返回了无法识别的模型列表" };
}
diff --git a/test/helpers.ts b/test/helpers.ts
index 2ba1624..a09e44f 100644
--- a/test/helpers.ts
+++ b/test/helpers.ts
@@ -23,7 +23,7 @@ async function body(req: http.IncomingMessage): Promise<any> {
}
/** 模拟 OpenAI 兼容端点:/v1/models + 流式 /v1/chat/completions(回声 + 模型名) */
-export async function mockOpenAI(opts: { models?: string[]; key?: string | string[]; reply?: (msgs: any[], model: string) => string } = {}) {
+export async function mockOpenAI(opts: { models?: string[]; key?: string | string[]; reply?: (msgs: any[], model: string) => string; kinds?: Record<string, string> } = {}) {
const models = opts.models ?? ["mock-a", "mock-b"];
const keys = opts.key === undefined ? undefined : ([] as string[]).concat(opts.key);
const calls: Array<{ model: string; messages: any[]; auth?: string; skill?: string }> = [];
@@ -34,7 +34,9 @@ export async function mockOpenAI(opts: { models?: string[]; key?: string | strin
return;
}
if (req.url?.endsWith("/models") && req.method === "GET") {
- res.writeHead(200, { "content-type": "application/json" }).end(JSON.stringify({ object: "list", data: models.map((id) => ({ id, object: "model" })) }));
+ res.writeHead(200, { "content-type": "application/json" }).end(
+ JSON.stringify({ object: "list", data: models.map((id) => ({ id, object: "model", ...(opts.kinds?.[id] !== undefined ? { kind: opts.kinds[id] } : {}) })) }),
+ );
return;
}
if (req.url?.endsWith("/chat/completions") && req.method === "POST") {
diff --git a/test/model-picks.integration.test.ts b/test/model-picks.integration.test.ts
new file mode 100644
index 0000000..53057a8
--- /dev/null
+++ b/test/model-picks.integration.test.ts
@@ -0,0 +1,56 @@
+/**
+ * 集成测试:真实内置 magpie。它对超过 24 个模型、又没有 picks 的服务商只放出列表前 24 个(没放出的
+ * 连 "provider/model" 也不服务);连接 Modelflare 时要把全部对话模型写成 picks,决策模型不算。
+ */
+import { describe, it, expect, beforeAll, afterAll } from "vitest";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+import { App } from "../src/app.js";
+import { loadConfig } from "../src/config.js";
+import { resolveMagpieBinary } from "../src/magpie/supervisor.js";
+import { FakeFeishu, mockOpenAI, until } from "./helpers.js";
+
+const d = resolveMagpieBinary() ? describe : describe.skip;
+
+// 按字母排序时第 24 个之后的模型,正是以前被截掉的那些
+const chat = Array.from({ length: 30 }, (_, i) => `model-${String(i).padStart(2, "0")}`);
+const kinds: Record<string, string> = { ...Object.fromEntries(chat.map((m) => [m, "text"])), "jev-decide": "", "img-gen": "image" };
+
+d("Modelflare 模型很多时全部启用", () => {
+ let llm: Awaited<ReturnType<typeof mockOpenAI>>;
+ let app: App;
+
+ beforeAll(async () => {
+ llm = await mockOpenAI({ key: "mf-many-key", models: [...chat, "jev-decide", "img-gen"], kinds });
+ process.env.AMB_MODELFLARE_BASE_URL = llm.url + "/v1";
+ const dataDir = fs.mkdtempSync(path.join(os.tmpdir(), "amb-picks-"));
+ app = new App(loadConfig({ dataDir, env: { AMB_UI_PORT: "0", AMB_WEIXIN_API_BASE: "http://127.0.0.1:9", AMB_AGENT: "off" } }), {
+ feishuTransport: () => new FakeFeishu(),
+ printQrToTerminal: false,
+ noWeb: true,
+ });
+ await app.start();
+ }, 60_000);
+
+ afterAll(async () => {
+ delete process.env.AMB_MODELFLARE_BASE_URL;
+ await app?.stop();
+ await llm?.close();
+ });
+
+ it("列出全部 30 个对话模型,不含决策与生图模型;第 24 个之后的也能对话", async () => {
+ const r = await app.connectModelflare("mf-many-key", "auto");
+ expect(r.ok && r.chatModels).toEqual(chat);
+ let list: string[] = [];
+ await until(async () => {
+ list = await app.runtime.listModels(true).catch(() => []);
+ return list.length >= chat.length;
+ }, 15_000, 200);
+ expect(list.map((m) => m.replace(/^modelflare\//, "")).sort()).toEqual(chat);
+ app.store.updateBot("default", { defaultModel: "model-29" });
+ const out = { text: "", async apply(op: any) { if (op.type === "stream.finish") this.text = op.text; } };
+ await app.runtime.handle({ id: "e1", channel: "web", botId: "default", conversation: { kind: "dm", id: "c" }, sender: { id: "u" }, parts: [{ type: "text", text: "你好" }], replyHandle: {}, receivedAt: Date.now() }, out);
+ expect(out.text).toContain("[model-29] 你说:你好");
+ }, 40_000);
+});
diff --git a/test/unit.test.ts b/test/unit.test.ts
index 7c92925..0ac90b4 100644
--- a/test/unit.test.ts
+++ b/test/unit.test.ts
@@ -249,10 +249,14 @@ describe("Modelflare", () => {
});
try {
const base = srv.url + "/v1";
- expect(await verifyModelflareKey(base, "k1")).toEqual({ ok: true, models: ["a", "b"] });
+ // 没有 kind 的旧版网关:全部算对话模型
+ expect(await verifyModelflareKey(base, "k1")).toEqual({ ok: true, models: ["a", "b"], chatModels: ["a", "b"] });
expect(auth.at(-1)).toBe("Bearer k1");
+ // 有 kind 时只有 text 是对话模型(生图、生视频、决策模型不算)
+ body = JSON.stringify({ data: [{ id: "t", kind: "text" }, { id: "i", kind: "image" }, { id: "v", kind: "video" }, { id: "d" }] });
+ expect(await verifyModelflareKey(base, "k1")).toEqual({ ok: true, models: ["d", "i", "t", "v"], chatModels: ["t"] });
body = JSON.stringify({ data: [] });
- expect(await verifyModelflareKey(base, "k1")).toEqual({ ok: true, models: [] });
+ expect(await verifyModelflareKey(base, "k1")).toEqual({ ok: true, models: [], chatModels: [] });
status = 401;
expect(await verifyModelflareKey(base, "k")).toMatchObject({ ok: false, code: "invalid_key", status: 401 });
status = 403;