feat(vercel-ai-gateway): support embeddings, images and credit usage

Extend Vercel AI Gateway beyond chat: add OpenAI-compatible embeddings and image generation endpoints, credit balance fetch on the usage dashboard, retry on 429, and models catalog fetcher. Thinking/reasoning mapping is omitted pending a project-wide refactor. Co-authored-by: Ngô Tấn Tài <tantai@newnol.io.vn> Co-authored-by: Cursor <cursoragent@cursor.com>
2026-06-13 10:54:51 +07:00
parent d9b030011f
commit b33cbb0280
10 changed files with 164 additions and 2 deletions
--- a/open-sse/config/providers.js
+++ b/open-sse/config/providers.js
@@ -129,7 +129,8 @@ export const PROVIDERS = {
  },
  "vercel-ai-gateway": {
    baseUrl: "https://ai-gateway.vercel.sh/v1/chat/completions",
-    format: "openai"
+    format: "openai",
+    retry: { 429: 2 }
  },
  glm: {
    baseUrl: "https://api.z.ai/api/anthropic/v1/messages",
--- a/open-sse/handlers/embeddingProviders/index.js
+++ b/open-sse/handlers/embeddingProviders/index.js
@@ -6,6 +6,7 @@ import openaiCompatNode from "./openaiCompatNode.js";
 const OPENAI_COMPAT_PROVIDERS = [
  "openai", "openrouter", "mistral", "voyage-ai", "fireworks",
  "together", "nebius", "github", "nvidia", "jina-ai",
+  "vercel-ai-gateway",
 ];

 const ADAPTERS = {
--- a/open-sse/handlers/embeddingProviders/openai.js
+++ b/open-sse/handlers/embeddingProviders/openai.js
@@ -12,6 +12,7 @@ const ENDPOINTS = {
  github: "https://models.github.ai/inference/embeddings",
  nvidia: "https://integrate.api.nvidia.com/v1/embeddings",
  "jina-ai": "https://api.jina.ai/v1/embeddings",
+  "vercel-ai-gateway": "https://ai-gateway.vercel.sh/v1/embeddings",
 };

 export default function createOpenAIEmbeddingAdapter(providerId) {
--- a/open-sse/handlers/imageProviders/index.js
+++ b/open-sse/handlers/imageProviders/index.js
@@ -17,6 +17,7 @@ const ADAPTERS = {
  minimax: createOpenAIAdapter("minimax"),
  openrouter: createOpenAIAdapter("openrouter"),
  recraft: createOpenAIAdapter("recraft"),
+  "vercel-ai-gateway": createOpenAIAdapter("vercel-ai-gateway"),
  xai: createOpenAIAdapter("xai"),
  gemini,
  codex,
--- a/open-sse/handlers/imageProviders/openai.js
+++ b/open-sse/handlers/imageProviders/openai.js
@@ -5,6 +5,7 @@ const ENDPOINTS = {
  minimax: "https://api.minimaxi.com/v1/images/generations",
  openrouter: "https://openrouter.ai/api/v1/images/generations",
  recraft: "https://external.api.recraft.ai/v1/images/generations",
+  "vercel-ai-gateway": "https://ai-gateway.vercel.sh/v1/images/generations",
  xai: "https://api.x.ai/v1/images/generations",
 };

--- a/open-sse/services/usage.js
+++ b/open-sse/services/usage.js
@@ -30,6 +30,11 @@ const MINIMAX_USAGE_URLS = {
  ],
 };

+// Vercel AI Gateway credits endpoint
+// Returns { balance: "95.50", total_used: "4.50" } (USD as decimal strings).
+// Docs: https://vercel.com/docs/ai-gateway/usage
+const VERCEL_AI_GATEWAY_CREDITS_URL = "https://ai-gateway.vercel.sh/v1/credits";
+
 // Antigravity API config (from Quotio)
 const ANTIGRAVITY_CONFIG = {
  quotaApiUrl: "https://cloudcode-pa.googleapis.com/v1internal:fetchAvailableModels",
@@ -92,6 +97,8 @@ export async function getUsageForProvider(connection, proxyOptions = null) {
    case "minimax":
    case "minimax-cn":
      return await getMiniMaxUsage(apiKey, provider, proxyOptions);
+    case "vercel-ai-gateway":
+      return await getVercelAiGatewayUsage(apiKey, proxyOptions);
    default:
      return { message: `Usage API not implemented for ${provider}` };
  }
@@ -1203,6 +1210,89 @@ async function getMiniMaxUsage(apiKey, provider, proxyOptions = null) {
  return { message: lastErrorMessage ? `MiniMax connected. Unable to fetch usage: ${lastErrorMessage}` : "MiniMax connected. Unable to fetch usage." };
 }

+
+/**
+ * Vercel AI Gateway usage — credit balance for the API key
+ *
+ * Calls GET /v1/credits which returns:
+ *   { "balance": "95.50", "total_used": "4.50" }   (USD as decimal strings)
+ *
+ * We surface this as a single "Balance ($)" quota row so the existing
+ * QuotaTable / progress-bar UI can render it. used = total_used,
+ * total = balance + total_used (the original credit allotment), so the
+ * remaining percentage equals balance / total.
+ *
+ * Docs: https://vercel.com/docs/ai-gateway/usage
+ */
+async function getVercelAiGatewayUsage(apiKey, proxyOptions = null) {
+  if (!apiKey) {
+    return { message: "Vercel AI Gateway API key not available." };
+  }
+
+  try {
+    const response = await proxyAwareFetch(VERCEL_AI_GATEWAY_CREDITS_URL, {
+      method: "GET",
+      headers: {
+        Authorization: `Bearer ${apiKey}`,
+        Accept: "application/json",
+      },
+    }, proxyOptions);
+
+    if (response.status === 401 || response.status === 403) {
+      return { message: "Vercel AI Gateway API key invalid or expired." };
+    }
+
+    if (!response.ok) {
+      const errorText = await response.text().catch(() => "");
+      const trimmed = errorText ? `: ${errorText.slice(0, 200)}` : "";
+      return { message: `Vercel AI Gateway credits API error (${response.status})${trimmed}` };
+    }
+
+    const data = await response.json();
+
+    // Vercel returns numeric strings; coerce safely.
+    const balance = Number(data?.balance) || 0;
+    const totalUsed = Number(data?.total_used) || 0;
+
+    // Vercel gives $5/month free credit. The API doesn't return the
+    // monthly allocation so we use the known constant as the denominator.
+    const MONTHLY_CREDIT = 5;
+    const remainingPercentage = (balance / MONTHLY_CREDIT) * 100;
+
+    if (balance <= 0 && totalUsed <= 0) {
+      return {
+        plan: "Pay-as-you-go",
+        message: "Vercel AI Gateway connected. No credit allocation found (BYOK or unfunded account).",
+        quotas: {},
+      };
+    }
+
+    // "Used (USD)": how much has been spent this month (no fixed cap → unlimited).
+    // "Remaining (USD)": balance remaining out of the $5 monthly allocation.
+    return {
+      plan: "Pay-as-you-go",
+      quotas: {
+        "Used (USD)": {
+          used: totalUsed,
+          total: 0,
+          remaining: 0,
+          remainingPercentage: 100,
+          unlimited: true,
+        },
+        "Remaining (USD)": {
+          used: balance,
+          total: MONTHLY_CREDIT,
+          remaining: balance,
+          remainingPercentage,
+          unlimited: false,
+        },
+      },
+    };
+  } catch (error) {
+    return { message: `Vercel AI Gateway error: ${error.message}` };
+  }
+}
+
 async function getQoderUsage(accessToken, proxyOptions = null) {
  if (!accessToken) {
    return { message: "Qoder usage unavailable: no access token" };
--- a/src/app/(dashboard)/dashboard/usage/components/ProviderLimits/utils.js
+++ b/src/app/(dashboard)/dashboard/usage/components/ProviderLimits/utils.js
@@ -207,6 +207,24 @@ export function parseQuotaData(provider, data) {
        }
        break;

+      case "vercel-ai-gateway":
+        // Vercel returns currency credit balance, not request quotas.
+        // The 'Remaining (USD)' row needs explicit remainingPercentage because
+        // its used/total values would otherwise compute the wrong direction
+        // (e.g. used=95.5 / total=100 → 4% instead of 96%).
+        if (data.quotas) {
+          Object.entries(data.quotas).forEach(([name, quota]) => {
+            normalizedQuotas.push({
+              name,
+              used: quota.used || 0,
+              total: quota.total || 0,
+              resetAt: quota.resetAt || null,
+              remainingPercentage: quota.remainingPercentage,
+            });
+          });
+        }
+        break;
+
      default:
        // Generic fallback for unknown providers
        if (data.quotas) {
--- a/src/shared/constants/providers.js
+++ b/src/shared/constants/providers.js
@@ -80,7 +80,7 @@ export const APIKEY_PROVIDERS = {
  "xiaomi-tokenplan": { id: "xiaomi-tokenplan", alias: "xmtp", name: "Xiaomi MiMo (Token Plan)", icon: "smart_toy", color: "#FF6700", textIcon: "XT", website: "https://mimo.xiaomi.com", notice: { text: "Xiaomi MiMo Token Plan subscription (API key starts with tp-). Token Plan keys are cluster-specific — select the region matching your subscription.", apiKeyUrl: "https://mimo.xiaomi.com" }, hasProviderSpecificData: true, regions: [{ id: "sgp", label: "Singapore", baseUrl: "https://token-plan-sgp.xiaomimimo.com/v1" }, { id: "cn", label: "China", baseUrl: "https://token-plan-cn.xiaomimimo.com/v1" }, { id: "ams", label: "Europe", baseUrl: "https://token-plan-ams.xiaomimimo.com/v1" }], defaultRegion: "sgp" },
  "volcengine-ark": { id: "volcengine-ark", alias: "ark", name: "Volcengine Ark", icon: "cloud", color: "#1677FF", textIcon: "ARK", website: "https://ark.cn-beijing.volces.com", notice: { apiKeyUrl: "https://console.volcengine.com/ark/region:ark+cn-beijing/apiKey" } },
  openai: { id: "openai", alias: "openai", name: "OpenAI", icon: "auto_awesome", color: "#10A37F", textIcon: "OA", website: "https://platform.openai.com", notice: { apiKeyUrl: "https://platform.openai.com/api-keys" }, serviceKinds: ["llm", "embedding", "tts", "stt", "image", "imageToText", "webSearch"], thinkingConfig: THINKING_CONFIG.effort, searchViaChat: { defaultModel: "gpt-4o-mini", pricingUrl: "https://openai.com/api/pricing" }, ttsConfig: { baseUrl: "https://api.openai.com/v1/audio/speech", authType: "apikey", authHeader: "bearer", format: "openai", models: [{ id: "tts-1", name: "TTS-1" }, { id: "tts-1-hd", name: "TTS-1 HD" }, { id: "gpt-4o-mini-tts", name: "GPT-4o Mini TTS" }] }, sttConfig: { baseUrl: "https://api.openai.com/v1/audio/transcriptions", authType: "apikey", authHeader: "bearer", format: "openai", models: [{ id: "whisper-1", name: "Whisper 1" }, { id: "gpt-4o-transcribe", name: "GPT-4o Transcribe" }, { id: "gpt-4o-mini-transcribe", name: "GPT-4o Mini Transcribe" }] }, embeddingConfig: { baseUrl: "https://api.openai.com/v1/embeddings", authType: "apikey", authHeader: "bearer", models: [{ id: "text-embedding-3-small", name: "Text Embedding 3 Small", dimensions: 1536 }, { id: "text-embedding-3-large", name: "Text Embedding 3 Large", dimensions: 3072 }, { id: "text-embedding-ada-002", name: "Text Embedding Ada 002", dimensions: 1536 }] } },
-  "vercel-ai-gateway": { id: "vercel-ai-gateway", alias: "vercel", name: "Vercel AI Gateway", icon: "deployed_code", color: "#111827", textIcon: "VG", website: "https://vercel.com/ai-gateway", notice: { text: "Unified OpenAI-compatible endpoint from Vercel. Use your AI Gateway API key, then pick models with provider/model IDs like anthropic/claude-sonnet-4.6 or openai/gpt-5.4.", apiKeyUrl: "https://vercel.com/dashboard/~/ai-gateway" }, passthroughModels: true, serviceKinds: ["llm"] },
+  "vercel-ai-gateway": { id: "vercel-ai-gateway", alias: "vercel", name: "Vercel AI Gateway", icon: "deployed_code", color: "#111827", textIcon: "VG", website: "https://vercel.com/ai-gateway", notice: { text: "Unified OpenAI-compatible endpoint from Vercel. Use your AI Gateway API key, then pick models with provider/model IDs like anthropic/claude-sonnet-4.6 or openai/gpt-5.4.", apiKeyUrl: "https://vercel.com/dashboard/~/ai-gateway" }, passthroughModels: true, serviceKinds: ["llm", "embedding", "image", "imageToText", "webSearch"], searchViaChat: { defaultModel: "openai/gpt-4o-mini", pricingUrl: "https://vercel.com/docs/ai-gateway/pricing" }, modelsFetcher: { url: "https://ai-gateway.vercel.sh/v1/models", type: "openai" } },
  anthropic: { id: "anthropic", alias: "anthropic", name: "Anthropic", icon: "smart_toy", color: "#D97757", textIcon: "AN", website: "https://console.anthropic.com", notice: { apiKeyUrl: "https://console.anthropic.com/settings/keys" }, serviceKinds: ["llm", "imageToText"] },
  "opencode-go": { id: "opencode-go", alias: "ocg", name: "OpenCode Go", icon: "terminal", color: "#E87040", textIcon: "OC", website: "https://opencode.ai/auth", notice: { text: "OpenCode Go subscription: $5/mo (then $10/mo). Access to Kimi, GLM, Qwen, MiMo, MiniMax models.", apiKeyUrl: "https://opencode.ai/auth" } },
  azure: { id: "azure", alias: "azure", name: "Azure OpenAI", icon: "cloud", color: "#0078D4", textIcon: "AZ", website: "https://azure.microsoft.com/en-us/products/ai-services/openai-service", notice: { apiKeyUrl: "https://portal.azure.com/#view/Microsoft_Azure_ProjectOxford/CognitiveServicesHub/~/OpenAI" }, hasProviderSpecificData: true },
@@ -278,6 +278,7 @@ export const USAGE_SUPPORTED_PROVIDERS = [
  "glm-cn",
  "minimax",
  "minimax-cn",
+  "vercel-ai-gateway",
 ];

 // Subset that uses apikey auth (still surfaced on quota page)
@@ -286,4 +287,5 @@ export const USAGE_APIKEY_PROVIDERS = [
  "glm-cn",
  "minimax",
  "minimax-cn",
+  "vercel-ai-gateway",
 ];
--- a/tests/unit/embeddingsCore.test.js
+++ b/tests/unit/embeddingsCore.test.js
@@ -222,6 +222,21 @@ describe("buildEmbeddingsUrl", () => {
    expect(url).toBe("https://openrouter.ai/api/v1/embeddings");
  });

+  it("vercel-ai-gateway → https://ai-gateway.vercel.sh/v1/embeddings", async () => {
+    vi.mocked(fetch).mockResolvedValueOnce(makeProviderResponse(VALID_EMBEDDING_RESPONSE));
+
+    await handleEmbeddingsCore(makeOptions({
+      modelInfo: { provider: "vercel-ai-gateway", model: "openai/text-embedding-3-small" },
+      credentials: { apiKey: "vag-test-key" },
+    }));
+
+    const [url, init] = vi.mocked(fetch).mock.calls[0];
+    const sent = JSON.parse(init.body);
+    expect(url).toBe("https://ai-gateway.vercel.sh/v1/embeddings");
+    expect(init.headers.Authorization).toBe("Bearer vag-test-key");
+    expect(sent.model).toBe("openai/text-embedding-3-small");
+  });
+
  it("openai-compatible-* → uses baseUrl from providerSpecificData", async () => {
    vi.mocked(fetch).mockResolvedValueOnce(makeProviderResponse(VALID_EMBEDDING_RESPONSE));

--- a/tests/unit/image-generation.test.js
+++ b/tests/unit/image-generation.test.js
@@ -263,6 +263,38 @@ describe("handleImageGenerationCore", () => {
    );
  });

+  it("handles Vercel AI Gateway image generation as OpenAI-compatible", async () => {
+    global.fetch.mockResolvedValueOnce(
+      new Response(
+        JSON.stringify({
+          created: 1234567890,
+          data: [{ url: "https://example.com/vercel-image.png" }],
+        }),
+        { status: 200, headers: { "Content-Type": "application/json" } }
+      )
+    );
+
+    const result = await handleImageGenerationCore({
+      body: { prompt: "A watercolor castle", n: 1, size: "1024x1024" },
+      modelInfo: { provider: "vercel-ai-gateway", model: "openai/gpt-image-1" },
+      credentials: { apiKey: "vag-test-key" },
+      log: null,
+    });
+
+    expect(result.success).toBe(true);
+    expect(global.fetch).toHaveBeenCalledWith(
+      "https://ai-gateway.vercel.sh/v1/images/generations",
+      expect.objectContaining({
+        method: "POST",
+        headers: expect.objectContaining({
+          "Content-Type": "application/json",
+          Authorization: "Bearer vag-test-key",
+        }),
+        body: expect.stringContaining('"model":"openai/gpt-image-1"'),
+      })
+    );
+  });
+
  it("handles HuggingFace binary response", async () => {
    const imageBuffer = new Uint8Array([0x89, 0x50, 0x4e, 0x47]); // PNG header
    global.fetch.mockResolvedValueOnce(