Skip to content

Soul-driven AI agent with permission-hardened tools, token budgets, and multi-channel access. Runs 24/7 from CLI, Telegram or More.

feat(providers): add CloudGPU OpenAI-compatible provider preset

Cursor Agent1b97a848cc5e
diff --git a/.env.example b/.env.example
index 15b18ddf..6443141a 100644
--- a/.env.example
+++ b/.env.example
@@ -38,6 +38,11 @@ RUNCRATE_API_KEY=
RUNCRATE_BASE_URL=https://api.runcrate.ai/v1
RUNCRATE_MODEL=deepseek-ai/DeepSeek-V4.1-Flash
+# CloudGPU (OpenAI-compatible model API gateway)
+CLOUDGPU_API_KEY=
+CLOUDGPU_BASE_URL=https://cloudgpu.app/v1
+CLOUDGPU_MODEL=deepseek-v4.1-flash
+
# AI/ML API (OpenAI-compatible API)
AIMLAPI_API_KEY=
AIMLAPI_BASE_URL=https://api.aimlapi.com/v1
@@ -75,7 +80,7 @@ LM_STUDIO_MODEL=
LM_STUDIO_ENABLED=false
MIMO_TOKEN_PLAN_ENABLED=false
-# Default provider: deepseek | openai | anthropic | grok | atlascloud | runcrate | aimlapi | mercuryCloud | ollamaCloud | ollamaLocal | openaiCompat | litellm | lmStudio | mimo | mimoTokenPlan
+# Default provider: deepseek | openai | anthropic | grok | atlascloud | runcrate | cloudgpu | aimlapi | mercuryCloud | ollamaCloud | ollamaLocal | openaiCompat | litellm | lmStudio | mimo | mimoTokenPlan
DEFAULT_PROVIDER=deepseek
# ── Telegram ──
diff --git a/README.md b/README.md
index da4d9527..b6fbf2dd 100644
--- a/README.md
+++ b/README.md
@@ -243,6 +243,7 @@ Mercury tries your default provider first and falls back to the others if it fai
| Grok (xAI) | `grok-4` | `GROK_API_KEY` |
| Atlas Cloud | `qwen/qwen3.5-flash` | `ATLASCLOUD_API_KEY` |
| Runcrate | `deepseek-ai/DeepSeek-V4.1-Flash` | `RUNCRATE_API_KEY` |
+| CloudGPU | `deepseek-v4.1-flash` | `CLOUDGPU_API_KEY` |
| AI/ML API | `anthropic/claude-sonnet-4.6` | `AIMLAPI_API_KEY` |
| MiMo (Xiaomi) | `mimo-v2.5-pro` | `MIMO_API_KEY` |
| Ollama Cloud | `gpt-oss:120b` | `OLLAMA_CLOUD_API_KEY` |
diff --git a/README.zh-CN.md b/README.zh-CN.md
index a0758cd7..273fd0f0 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -243,6 +243,7 @@ Mercury 优先使用默认提供方,失败时自动切换到其他提供方。
| Grok (xAI) | `grok-4` | `GROK_API_KEY` |
| Atlas Cloud | `qwen/qwen3.5-flash` | `ATLASCLOUD_API_KEY` |
| Runcrate | `deepseek-ai/DeepSeek-V4.1-Flash` | `RUNCRATE_API_KEY` |
+| CloudGPU | `deepseek-v4.1-flash` | `CLOUDGPU_API_KEY` |
| AI/ML API | `anthropic/claude-sonnet-4.6` | `AIMLAPI_API_KEY` |
| MiMo(小米) | `mimo-v2.5-pro` | `MIMO_API_KEY` |
| Ollama Cloud | `gpt-oss:120b` | `OLLAMA_CLOUD_API_KEY` |
diff --git a/src/capabilities/mcp/config.ts b/src/capabilities/mcp/config.ts
index 090edaea..3b8d9c19 100644
--- a/src/capabilities/mcp/config.ts
+++ b/src/capabilities/mcp/config.ts
@@ -73,7 +73,7 @@ function literalSecretWarnings(id: string, field: 'env' | 'headers', values: Rec
* (it may be deliberate), but always called out.
*/
const MERCURY_OWN_SECRETS = new Set([
- 'AIMLAPI_API_KEY', 'ANTHROPIC_API_KEY', 'ATLASCLOUD_API_KEY', 'BOTS_WEBHOOK_SECRET', 'DEEPSEEK_API_KEY',
+ 'AIMLAPI_API_KEY', 'ANTHROPIC_API_KEY', 'ATLASCLOUD_API_KEY', 'BOTS_WEBHOOK_SECRET', 'CLOUDGPU_API_KEY', 'DEEPSEEK_API_KEY',
'DISCORD_BOT_TOKEN', 'GEMINI_API_KEY', 'GOOGLE_API_KEY', 'GROK_API_KEY', 'LITELLM_API_KEY', 'LM_STUDIO_API_KEY',
'MERCURY_CLOUD_ACCESS_KEY', 'MERCURY_CLOUD_AGENT_API_KEY', 'MERCURY_CLOUD_REFRESH_TOKEN', 'MIMO_API_KEY',
'MIMO_TOKEN_PLAN_API_KEY', 'OLLAMA_CLOUD_API_KEY', 'OPENAI_API_KEY', 'OPENAI_COMPAT_API_KEY', 'OPENROUTER_API_KEY',
diff --git a/src/cloud/pairing-flow.ts b/src/cloud/pairing-flow.ts
index 3952040b..0158f2c8 100644
--- a/src/cloud/pairing-flow.ts
+++ b/src/cloud/pairing-flow.ts
@@ -414,7 +414,7 @@ function isProviderConfiguredSafe(p: import('../utils/config.js').ProviderConfig
if (!p.enabled) return false;
if (p.name === 'ollamaLocal') return p.baseUrl.length > 0 && p.model.length > 0;
if (p.name === 'ollamaCloud') return p.apiKey.length > 0 && p.baseUrl.length > 0;
- if (p.name === 'atlascloud' || p.name === 'runcrate') return p.apiKey.length > 0 && p.baseUrl.length > 0 && p.model.length > 0;
+ if (p.name === 'atlascloud' || p.name === 'runcrate' || p.name === 'cloudgpu') return p.apiKey.length > 0 && p.baseUrl.length > 0 && p.model.length > 0;
if (p.name === 'openaiCompat') return p.baseUrl.length > 0 && p.model.length > 0;
if (p.name === 'chatgptWeb' || p.name === 'githubCopilot') return p.model.length > 0;
return p.apiKey.length > 0;
diff --git a/src/index.ts b/src/index.ts
index 666d4f85..f69a5e44 100644
--- a/src/index.ts
+++ b/src/index.ts
@@ -204,6 +204,7 @@ const PROVIDER_OPTIONS: Array<{ key: ProviderName; label: string }> = [
{ key: 'grok', label: 'Grok (xAI)' },
{ key: 'atlascloud', label: 'Atlas Cloud' },
{ key: 'runcrate', label: 'Runcrate' },
+ { key: 'cloudgpu', label: 'CloudGPU' },
{ key: 'ollamaCloud', label: 'Ollama Cloud' },
{ key: 'ollamaLocal', label: 'Ollama Local' },
{ key: 'atomicChat', label: 'Atomic Chat' },
@@ -1514,6 +1515,23 @@ async function configure(existingConfig?: MercuryConfig): Promise<void> {
continue;
}
+ if (provider === 'cloudgpu') {
+ const mask = isReconfig && config.providers.cloudgpu.apiKey ? ` [${maskKey(config.providers.cloudgpu.apiKey)}]` : '';
+ const result = await promptApiKeyWithModelSelection(
+ config,
+ 'cloudgpu',
+ 'CloudGPU',
+ chalk.white(` CloudGPU API key${mask}${isReconfig ? '' : ' (Enter to skip)'}: `),
+ isReconfig,
+ );
+ if (!result.skipped && result.apiKey && result.model) {
+ config.providers.cloudgpu.apiKey = result.apiKey;
+ config.providers.cloudgpu.model = result.model;
+ config.providers.cloudgpu.enabled = true;
+ }
+ continue;
+ }
+
if (provider === 'ollamaCloud') {
const mask = isReconfig && config.providers.ollamaCloud.apiKey ? ` [${maskKey(config.providers.ollamaCloud.apiKey)}]` : '';
const result = await promptApiKeyWithModelSelection(
diff --git a/src/providers/provider-meta.ts b/src/providers/provider-meta.ts
index 734b8372..cf0d79ac 100644
--- a/src/providers/provider-meta.ts
+++ b/src/providers/provider-meta.ts
@@ -34,6 +34,7 @@ export const PROVIDER_META: Record<ProviderName, ProviderMeta> = {
grok: { label: 'Grok (xAI)', kind: 'cloud', blurb: 'Grok models. The same key powers Grok voice.', needsKey: true, needsBaseUrl: false, keyUrl: 'https://console.x.ai', keyHint: 'xai-…' },
atlascloud: { label: 'Atlas Cloud', kind: 'gateway', blurb: 'Open models (Qwen, DeepSeek, Llama) on one key.', needsKey: true, needsBaseUrl: false, keyUrl: 'https://www.atlascloud.ai' },
runcrate: { label: 'Runcrate', kind: 'gateway', blurb: 'Open models (DeepSeek, GLM, Qwen, Kimi) on one key.', needsKey: true, needsBaseUrl: false, keyUrl: 'https://www.runcrate.ai/dashboard/api-keys' },
+ cloudgpu: { label: 'CloudGPU', kind: 'gateway', blurb: 'OpenAI-compatible model API gateway.', needsKey: true, needsBaseUrl: false, keyUrl: 'https://cloudgpu.app/api/console' },
ollamaCloud: { label: 'Ollama Cloud', kind: 'cloud', blurb: 'Large open models hosted by Ollama.', needsKey: true, needsBaseUrl: false, keyUrl: 'https://ollama.com/settings/keys' },
ollamaLocal: { label: 'Ollama (this computer)', kind: 'local', blurb: 'Models you run yourself with Ollama. Free and private.', needsKey: false, needsBaseUrl: false },
lmStudio: { label: 'LM Studio', kind: 'local', blurb: 'Models served by LM Studio on this computer.', needsKey: false, needsBaseUrl: false },
diff --git a/src/providers/registry.ts b/src/providers/registry.ts
index 90f27d95..492dc888 100644
--- a/src/providers/registry.ts
+++ b/src/providers/registry.ts
@@ -26,7 +26,7 @@ export async function createProvider(pc: ProviderConfig, tokenStore?: import('..
// catalogue is mostly third-party models reached through that surface.
const { OpenAICompatProvider } = await import('./openai-compat.js');
return new OpenAICompatProvider(pc, { useChatApi: true });
- } else if (pc.name === 'atlascloud' || pc.name === 'runcrate' || pc.name === 'ollamaCloud' || pc.name === 'openaiCompat' || pc.name === 'litellm' || pc.name === 'lmStudio') {
+ } else if (pc.name === 'atlascloud' || pc.name === 'runcrate' || pc.name === 'cloudgpu' || pc.name === 'ollamaCloud' || pc.name === 'openaiCompat' || pc.name === 'litellm' || pc.name === 'lmStudio') {
const { OpenAICompatProvider } = await import('./openai-compat.js');
return new OpenAICompatProvider(pc, { useChatApi: true });
} else if (pc.name === 'mimo' || pc.name === 'mimoTokenPlan') {
@@ -76,6 +76,7 @@ export class ProviderRegistry {
config.providers.grok,
config.providers.atlascloud,
config.providers.runcrate,
+ config.providers.cloudgpu,
config.providers.ollamaCloud,
config.providers.ollamaLocal,
config.providers.atomicChat,
diff --git a/src/utils/config.ts b/src/utils/config.ts
index aa7f5bf9..f4c8a529 100644
--- a/src/utils/config.ts
+++ b/src/utils/config.ts
@@ -72,6 +72,7 @@ export type ProviderName =
| 'grok'
| 'atlascloud'
| 'runcrate'
+ | 'cloudgpu'
| 'ollamaCloud'
| 'ollamaLocal'
| 'atomicChat'
@@ -117,6 +118,7 @@ export interface MercuryConfig {
grok: ProviderConfig;
atlascloud: ProviderConfig;
runcrate: ProviderConfig;
+ cloudgpu: ProviderConfig;
ollamaCloud: ProviderConfig;
ollamaLocal: ProviderConfig;
atomicChat: ProviderConfig;
@@ -405,6 +407,13 @@ export function getDefaultConfig(): MercuryConfig {
model: getEnv('RUNCRATE_MODEL', 'deepseek-ai/DeepSeek-V4.1-Flash'),
enabled: getEnvBool('RUNCRATE_ENABLED', true),
},
+ cloudgpu: {
+ name: 'cloudgpu',
+ apiKey: getEnv('CLOUDGPU_API_KEY', ''),
+ baseUrl: getEnv('CLOUDGPU_BASE_URL', 'https://cloudgpu.app/v1'),
+ model: getEnv('CLOUDGPU_MODEL', 'deepseek-v4.1-flash'),
+ enabled: getEnvBool('CLOUDGPU_ENABLED', true),
+ },
ollamaCloud: {
name: 'ollamaCloud',
apiKey: getEnv('OLLAMA_CLOUD_API_KEY', ''),
@@ -824,7 +833,7 @@ export function isProviderConfigured(provider: ProviderConfig): boolean {
if (provider.name === 'ollamaCloud') {
return provider.apiKey.length > 0 && provider.baseUrl.length > 0;
}
- if (provider.name === 'atlascloud' || provider.name === 'runcrate') {
+ if (provider.name === 'atlascloud' || provider.name === 'runcrate' || provider.name === 'cloudgpu') {
return provider.apiKey.length > 0 && provider.baseUrl.length > 0 && provider.model.length > 0;
}
if (provider.name === 'openaiCompat' || provider.name === 'litellm') {
diff --git a/src/utils/model-prefs.ts b/src/utils/model-prefs.ts
index 3a637cbb..7899783c 100644
--- a/src/utils/model-prefs.ts
+++ b/src/utils/model-prefs.ts
@@ -186,6 +186,7 @@ const PROVIDER_LABELS: Record<string, string> = {
aimlapi: 'AI/ML API',
atlascloud: 'Atlas Cloud',
runcrate: 'Runcrate',
+ cloudgpu: 'CloudGPU',
mimo: 'MiMo',
mimoTokenPlan: 'MiMo Token Plan',
githubCopilot: 'GitHub Copilot',
diff --git a/src/utils/provider-models.test.ts b/src/utils/provider-models.test.ts
index 30b03f17..2d55683e 100644
--- a/src/utils/provider-models.test.ts
+++ b/src/utils/provider-models.test.ts
@@ -110,6 +110,20 @@ describe('buildModelCatalog', () => {
expect(catalog.models).toContain('another-chat-model');
});
+ it('prefers the CloudGPU DeepSeek V4.1 Flash model when available', () => {
+ const catalog = buildModelCatalog('cloudgpu', [
+ 'kimi-k3',
+ 'deepseek-v4.1-flash',
+ 'deepseek-v4-pro',
+ 'another-chat-model',
+ ]);
+
+ expect(catalog.recommendedModel).toBe('deepseek-v4.1-flash');
+ expect(catalog.models[0]).toBe('deepseek-v4-pro');
+ expect(catalog.models).toContain('kimi-k3');
+ expect(catalog.models).toContain('another-chat-model');
+ });
+
it('falls back to the current Ollama Cloud model when no preferred default is available', () => {
const catalog = buildModelCatalog(
'ollamaCloud',
@@ -220,6 +234,34 @@ describe('fetchProviderModelCatalog', () => {
expect(catalog.recommendedModel).toBe('deepseek-ai/DeepSeek-V4.1-Flash');
expect(catalog.models).toEqual(['zai-org/GLM-5.3-Flash']);
});
+
+ it('keeps CloudGPU chat models and sends the key as a bearer token', async () => {
+ const fetchMock = vi.fn().mockResolvedValue(new Response(JSON.stringify({
+ data: [
+ { id: 'deepseek-v4.1-flash' },
+ { id: 'glm-5.2' },
+ { id: 'flux-1-schnell' },
+ { id: 'kokoro-tts' },
+ { id: 'whisper-v3' },
+ { id: 'qwen3-asr' },
+ ],
+ }), { status: 200 }));
+ vi.stubGlobal('fetch', fetchMock);
+
+ const catalog = await fetchProviderModelCatalog('cloudgpu', {
+ name: 'cloudgpu',
+ apiKey: 'cgw-sk-test',
+ baseUrl: 'https://cloudgpu.app/v1/',
+ model: 'deepseek-v4.1-flash',
+ enabled: true,
+ });
+
+ const [url, init] = fetchMock.mock.calls[0];
+ expect(url).toBe('https://cloudgpu.app/v1/models');
+ expect(init.headers).toMatchObject({ Authorization: 'Bearer cgw-sk-test' });
+ expect(catalog.recommendedModel).toBe('deepseek-v4.1-flash');
+ expect(catalog.models).toEqual(['glm-5.2']);
+ });
});
describe('chat models only', () => {
diff --git a/src/utils/provider-models.ts b/src/utils/provider-models.ts
index 2e058b92..0c732c0b 100644
--- a/src/utils/provider-models.ts
+++ b/src/utils/provider-models.ts
@@ -67,6 +67,19 @@ const RUNCRATE_PREFERRED_MODELS = [
'moonshotai/Kimi-K2.6',
] as const;
+// Verified present on GET https://cloudgpu.app/v1/models on 2026-11-01.
+// CloudGPU is an OpenAI-compatible model API gateway.
+const CLOUDGPU_PREFERRED_MODELS = [
+ 'deepseek-v4.1-flash',
+ 'deepseek-v4-pro',
+ 'kimi-k3',
+ 'kimi-k2.7-code',
+ 'glm-5.2',
+ 'minimax-m3',
+ 'qwen3-coder-480b',
+ 'qwen3.5-397b',
+] as const;
+
const OLLAMA_CLOUD_PREFERRED_MODELS = [
'gpt-oss:120b',
'gpt-oss:120b-cloud',
@@ -145,6 +158,7 @@ export function getPreferredModelsForProvider(provider: ProviderName): string[]
grok: GROK_PREFERRED_MODELS,
atlascloud: ATLASCLOUD_PREFERRED_MODELS,
runcrate: RUNCRATE_PREFERRED_MODELS,
+ cloudgpu: CLOUDGPU_PREFERRED_MODELS,
ollamaCloud: OLLAMA_CLOUD_PREFERRED_MODELS,
ollamaLocal: OLLAMA_LOCAL_PREFERRED_MODELS,
atomicChat: ATOMIC_CHAT_PREFERRED_MODELS,
@@ -274,6 +288,7 @@ function chooseRecommendedModel(
grok: GROK_PREFERRED_MODELS,
atlascloud: ATLASCLOUD_PREFERRED_MODELS,
runcrate: RUNCRATE_PREFERRED_MODELS,
+ cloudgpu: CLOUDGPU_PREFERRED_MODELS,
ollamaCloud: OLLAMA_CLOUD_PREFERRED_MODELS,
ollamaLocal: OLLAMA_LOCAL_PREFERRED_MODELS,
atomicChat: ATOMIC_CHAT_PREFERRED_MODELS,
@@ -325,6 +340,7 @@ export function buildModelCatalog(
grok: GROK_PREFERRED_MODELS,
atlascloud: ATLASCLOUD_PREFERRED_MODELS,
runcrate: RUNCRATE_PREFERRED_MODELS,
+ cloudgpu: CLOUDGPU_PREFERRED_MODELS,
ollamaCloud: OLLAMA_CLOUD_PREFERRED_MODELS,
ollamaLocal: OLLAMA_LOCAL_PREFERRED_MODELS,
atomicChat: ATOMIC_CHAT_PREFERRED_MODELS,
@@ -359,6 +375,8 @@ async function fetchOpenAICompatModels(provider: ProviderName, config: ProviderC
errorMessage = 'Mercury could not fetch models for this Atlas Cloud key. Please re-enter it.';
} else if (provider === 'runcrate') {
errorMessage = 'Mercury could not fetch models for this Runcrate key. Please re-enter it.';
+ } else if (provider === 'cloudgpu') {
+ errorMessage = 'Mercury could not fetch models for this CloudGPU key. Please re-enter it.';
} else if (provider === 'deepseek') {
errorMessage = 'Mercury could not fetch models for this DeepSeek key. Please re-enter it.';
} else if (provider === 'aimlapi') {
@@ -409,6 +427,17 @@ async function fetchOpenAICompatModels(provider: ProviderName, config: ProviderC
// endpoint; each row says what it outputs.
return id.length > 0 && (outputModalities ?? []).includes('text');
}
+ if (provider === 'cloudgpu') {
+ // CloudGPU lists image, speech and transcription models on the same
+ // endpoint. Rows do not declare an output modality, so those are dropped by id.
+ const lower = id.toLowerCase();
+ return id.length > 0
+ && !lower.startsWith('flux')
+ && !lower.includes('tts')
+ && !lower.includes('whisper')
+ && !lower.endsWith('-asr')
+ && !lower.includes('transcri');
+ }
return isOpenAIChatModel(id);
})
.map(({ id }) => id);