feat: add CloudGPU as a model provider (#21543)

conf/llm_factories.json

@@ -3532,6 +3532,14 @@
             "llm": [],
             "url": "https://api.runcrate.ai/v1"
         },
+        {
+            "name": "CloudGPU",
+            "logo": "",
+            "tags": "LLM,IMAGE2TEXT",
+            "status": "1",
+            "llm": [],
+            "url": "https://cloudgpu.app/v1"
+        },
         {
             "name": "StepFun",
             "logo": "",

conf/model.md

@@ -176,6 +176,7 @@ The following providers are **aggregators** — they host models from multiple u
 | **AstraFlow** | Multi-provider aggregator |
 | **Avian** | Multi-provider aggregator |
 | **Baidu (Qianwen)** | Ernie + third-party models |
+| **CloudGPU** | GPU cloud with an OpenAI-compatible inference API for open models |
 | **CometAPI** | Multi-provider aggregator |
 | **DeepInfra** | Open-source model hosting |
 | **Demon Route** | OpenAI-compatible router |

conf/models/cloudgpu.json

@@ -0,0 +1,108 @@
+{
+  "name": "CloudGPU",
+  "url": {
+    "default": "https://cloudgpu.app/v1"
+  },
+  "url_suffix": {
+    "chat": "chat/completions",
+    "models": "models"
+  },
+  "class": "cloudgpu",
+  "models": [
+    {
+      "name": "deepseek-v4.1-flash",
+      "context_length": 1048576,
+      "model_types": [
+        "chat",
+        "vision"
+      ],
+      "tools": {
+        "support": true
+      }
+    },
+    {
+      "name": "deepseek-v4-flash",
+      "context_length": 1048576,
+      "model_types": [
+        "chat"
+      ],
+      "tools": {
+        "support": true
+      }
+    },
+    {
+      "name": "deepseek-v4-pro",
+      "context_length": 1048576,
+      "model_types": [
+        "chat"
+      ],
+      "tools": {
+        "support": true
+      }
+    },
+    {
+      "name": "kimi-k2.7-code",
+      "context_length": 262144,
+      "model_types": [
+        "chat",
+        "vision"
+      ],
+      "tools": {
+        "support": true
+      }
+    },
+    {
+      "name": "kimi-k2.6",
+      "context_length": 262144,
+      "model_types": [
+        "chat",
+        "vision"
+      ],
+      "tools": {
+        "support": true
+      }
+    },
+    {
+      "name": "glm-5.3-flash",
+      "context_length": 1048576,
+      "model_types": [
+        "chat",
+        "vision"
+      ],
+      "tools": {
+        "support": true
+      }
+    },
+    {
+      "name": "glm-5.2",
+      "context_length": 1048576,
+      "model_types": [
+        "chat"
+      ],
+      "tools": {
+        "support": true
+      }
+    },
+    {
+      "name": "qwen3-coder-480b",
+      "context_length": 262144,
+      "model_types": [
+        "chat"
+      ],
+      "tools": {
+        "support": true
+      }
+    },
+    {
+      "name": "minimax-m3",
+      "context_length": 1048576,
+      "model_types": [
+        "chat",
+        "vision"
+      ],
+      "tools": {
+        "support": true
+      }
+    }
+  ]
+}

internal/entity/models/cloudgpu.go

@@ -0,0 +1,70 @@
+//
+//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+//
+
+package models
+
+import "context"
+
+// CloudGPUModel drives CloudGPU (https://cloudgpu.app/v1), a GPU cloud whose
+// inference API serves open models through OpenAI-compatible chat and
+// model-list endpoints, so the embedded OpenAIAPICompatibleModel covers
+// every call.
+type CloudGPUModel struct {
+	*OpenAIAPICompatibleModel
+}
+
+// NewCloudGPUModel creates a CloudGPU driver.
+func NewCloudGPUModel(baseURL map[string]string, urlSuffix URLSuffix) *CloudGPUModel {
+	return &CloudGPUModel{
+		OpenAIAPICompatibleModel: NewOpenAIAPICompatibleModel(baseURL, urlSuffix),
+	}
+}
+
+// Name returns the driver name.
+func (m *CloudGPUModel) Name() string {
+	return "cloudgpu"
+}
+
+// NewInstance returns a driver bound to the given base URL.
+func (m *CloudGPUModel) NewInstance(baseURL map[string]string) ModelDriver {
+	return NewCloudGPUModel(baseURL, m.baseModel.URLSuffix)
+}
+
+// ListModels lists CloudGPU models and, for ids in the CloudGPU catalog
+// (conf/models/cloudgpu.json), returns the catalogued model types so vision
+// models keep their image-input capability. Unknown ids keep the name-based
+// inference of the embedded driver.
+func (m *CloudGPUModel) ListModels(ctx context.Context, apiConfig *APIConfig) ([]ListModelResponse, error) {
+	models, err := m.OpenAIAPICompatibleModel.ListModels(ctx, apiConfig)
+	if err != nil {
+		return nil, err
+	}
+
+	pm := GetProviderManager()
+	if pm == nil {
+		return models, nil
+	}
+	provider := pm.FindProvider("CloudGPU")
+	if provider == nil {
+		return models, nil
+	}
+	for i := range models {
+		if known := pm.FindModel(provider, models[i].Name); known != nil && len(known.ModelTypes) > 0 {
+			models[i].ModelTypes = append([]string(nil), known.ModelTypes...)
+		}
+	}
+	return models, nil
+}

internal/entity/models/cloudgpu_test.go

@@ -0,0 +1,194 @@
+//
+//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+//
+
+package models
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"ragflow/internal/common"
+	"strings"
+	"testing"
+)
+
+// newCloudGPUForTest creates a CloudGPUModel instance for testing.
+func newCloudGPUForTest(baseURL string) *CloudGPUModel {
+	return NewCloudGPUModel(
+		map[string]string{"default": baseURL},
+		URLSuffix{Chat: "chat/completions", Models: "models"},
+	)
+}
+
+// TestCloudGPUName tests the driver name.
+func TestCloudGPUName(t *testing.T) {
+	if got := newCloudGPUForTest("http://unused").Name(); got != "cloudgpu" {
+		t.Errorf("Name()=%q", got)
+	}
+}
+
+// TestCloudGPUFactory tests creating the driver from factory.
+func TestCloudGPUFactory(t *testing.T) {
+	driver, err := NewModelFactory().CreateModelDriver("cloudgpu", map[string]string{"default": "http://unused"}, URLSuffix{})
+	if err != nil {
+		t.Fatalf("CreateModelDriver: %v", err)
+	}
+	if _, ok := driver.(*CloudGPUModel); !ok {
+		t.Fatalf("driver type=%T, want *CloudGPUModel", driver)
+	}
+	if _, ok := driver.NewInstance(map[string]string{"default": "http://other"}).(*CloudGPUModel); !ok {
+		t.Fatal("NewInstance did not return *CloudGPUModel")
+	}
+}
+
+// The shipped config must resolve to the CloudGPU driver. CreateModelDriver
+// dispatches on strings.ToLower(name), so the "CloudGPU" provider name has
+// to match the "cloudgpu" case; anything else silently falls through to
+// DummyModel.
+func TestCloudGPUProviderConfig(t *testing.T) {
+	dir, restore := setupProviderTestDir(t, "cloudgpu.json")
+	defer restore()
+
+	if err := InitProviderManager(dir); err != nil {
+		t.Fatalf("InitProviderManager: %v", err)
+	}
+
+	provider := GetProviderManager().FindProvider("CloudGPU")
+	if provider == nil {
+		t.Fatal("CloudGPU provider not found")
+	}
+	if _, ok := provider.ModelDriver.(*CloudGPUModel); !ok {
+		t.Fatalf("ModelDriver=%T, want *models.CloudGPUModel", provider.ModelDriver)
+	}
+}
+
+// TestCloudGPUChatSendsAPIKey tests chat completion authentication and that
+// the model id reaches the API unchanged.
+func TestCloudGPUChatSendsAPIKey(t *testing.T) {
+	withSSRFBypass(t)
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/chat/completions" {
+			t.Errorf("path=%s", r.URL.Path)
+		}
+		if got := r.Header.Get("Authorization"); got != "Bearer sk-gw-test" {
+			t.Errorf("Authorization=%q", got)
+		}
+		var body struct {
+			Model string `json:"model"`
+		}
+		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+			t.Errorf("decode request: %v", err)
+		}
+		if body.Model != "deepseek-v4.1-flash" {
+			t.Errorf("model=%q", body.Model)
+		}
+		_ = json.NewEncoder(w).Encode(map[string]any{
+			"id":      "chat-cloudgpu",
+			"choices": []map[string]any{{"message": map[string]any{"content": "pong"}}},
+			"usage":   map[string]any{"prompt_tokens": 3, "completion_tokens": 5, "total_tokens": 8},
+		})
+	}))
+	defer srv.Close()
+
+	usage := &common.ModelUsage{}
+	resp, err := newCloudGPUForTest(srv.URL).ChatWithMessages(
+		t.Context(),
+		"deepseek-v4.1-flash",
+		[]Message{{Role: "user", Content: "ping"}},
+		&APIConfig{ApiKey: &testAPIKey},
+		nil,
+		usage,
+	)
+	if err != nil {
+		t.Fatalf("ChatWithMessages: %v", err)
+	}
+	if *resp.Answer != "pong" {
+		t.Errorf("Answer=%q", *resp.Answer)
+	}
+	assertModelUsage(t, usage, 3, 5, 8)
+}
+
+// TestCloudGPUListModelsAndCheckConnection tests model listing and connectivity check.
+func TestCloudGPUListModelsAndCheckConnection(t *testing.T) {
+	withSSRFBypass(t)
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.Method != http.MethodGet || r.URL.Path != "/models" {
+			t.Errorf("%s %s", r.Method, r.URL.Path)
+		}
+		_ = json.NewEncoder(w).Encode(map[string]any{
+			"data": []map[string]any{
+				{"id": "deepseek-v4.1-flash"},
+				{"id": "kimi-k2.6"},
+			},
+		})
+	}))
+	defer srv.Close()
+
+	model := newCloudGPUForTest(srv.URL)
+	list, err := model.ListModels(t.Context(), &APIConfig{ApiKey: &testAPIKey})
+	if err != nil {
+		t.Fatalf("ListModels: %v", err)
+	}
+	if joinModelNames(list, ",") != "deepseek-v4.1-flash,kimi-k2.6" {
+		t.Errorf("models=%v", list)
+	}
+	if err := model.CheckConnection(t.Context(), &APIConfig{ApiKey: &testAPIKey}); err != nil {
+		t.Fatalf("CheckConnection: %v", err)
+	}
+}
+
+// Known CloudGPU models must report their catalogued types, so the vision
+// models are not downgraded to chat-only by name inference. Unknown ids keep
+// the name-based fallback.
+func TestCloudGPUListModelsUsesCatalogTypes(t *testing.T) {
+	withSSRFBypass(t)
+	dir, restore := setupProviderTestDir(t, "cloudgpu.json")
+	defer restore()
+	if err := InitProviderManager(dir); err != nil {
+		t.Fatalf("InitProviderManager: %v", err)
+	}
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_ = json.NewEncoder(w).Encode(map[string]any{
+			"data": []map[string]any{
+				{"id": "deepseek-v4.1-flash"},
+				{"id": "kimi-k2.6"},
+				{"id": "glm-5.2"},
+				{"id": "some-new-vl-model"},
+			},
+		})
+	}))
+	defer srv.Close()
+
+	list, err := newCloudGPUForTest(srv.URL).ListModels(t.Context(), &APIConfig{ApiKey: &testAPIKey})
+	if err != nil {
+		t.Fatalf("ListModels: %v", err)
+	}
+	want := map[string]string{
+		"deepseek-v4.1-flash": "chat,vision",
+		"kimi-k2.6":           "chat,vision",
+		"glm-5.2":             "chat",
+		"some-new-vl-model":   "chat,vision",
+	}
+	if len(list) != len(want) {
+		t.Fatalf("models=%v", list)
+	}
+	for _, model := range list {
+		if got := strings.Join(model.ModelTypes, ","); got != want[model.Name] {
+			t.Errorf("%s types=%q, want %q", model.Name, got, want[model.Name])
+		}
+	}
+}

internal/entity/models/factory.go

@@ -201,6 +201,8 @@ func (f *ModelFactory) CreateModelDriver(providerName string, baseURL map[string
 		return NewEmpirioLabsModel(baseURL, urlSuffix), nil
 	case "runcrate":
 		return NewRuncrateModel(baseURL, urlSuffix), nil
+	case "cloudgpu":
+		return NewCloudGPUModel(baseURL, urlSuffix), nil
 	case "new api":
 		return NewNewAPIModel(baseURL, urlSuffix), nil
 	default:

web/src/assets/svg/llm/cloudgpu.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="48" height="48" viewBox="0 0 32 32"><rect x="1" y="1" width="30" height="30" rx="8" fill="#111827"/><g fill="none" stroke="#FFFFFF" stroke-linecap="round"><path d="M6 24H26" stroke-width="2" opacity=".38"/><path d="M6 18.5H26" stroke-width="2" opacity=".62"/><path d="M6 13C13 13 19 13 26 6" stroke-width="2.4"/></g></svg>

web/src/components/svg-icon.tsx

@@ -122,6 +122,7 @@ const svgIcons = [
   LLMFactory.EmpirioLabs,
   LLMFactory.AIStupidLevel,
   LLMFactory.Runcrate,
+  LLMFactory.CloudGPU,
 ];
 
 export const LlmIcon = ({

web/src/constants/llm.ts

@@ -117,6 +117,7 @@ export enum LLMFactory {
   EmpirioLabs = 'EmpirioLabs AI',
   AIStupidLevel = 'AI Stupid Level',
   Runcrate = 'Runcrate',
+  CloudGPU = 'CloudGPU',
   SystemOne = 'SystemOne',
 }
 
@@ -217,6 +218,7 @@ export const IconMap = {
   [LLMFactory.EmpirioLabs]: 'empiriolabs',
   [LLMFactory.AIStupidLevel]: 'aistupidlevel',
   [LLMFactory.Runcrate]: 'runcrate',
+  [LLMFactory.CloudGPU]: 'cloudgpu',
 };
 
 /**
@@ -265,6 +267,7 @@ export const APIMapUrl = {
   [LLMFactory.AIStupidLevel]: 'https://aistupidlevel.info/router/keys',
   [LLMFactory.Runcrate]:
     'https://www.runcrate.ai/dashboard/api-keys?utm_source=ragflow&utm_medium=directory&utm_campaign=gpu_pricing',
+  [LLMFactory.CloudGPU]: 'https://cloudgpu.app/api/console',
   [LLMFactory.MWS]:
     'https://mws.ru/docs/cloud-platform/gpt/general/inference-text.html',
   [LLMFactory.Anthropic]: 'https://console.anthropic.com/settings/keys',