Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

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

CloudGPU0cfa864cc69c
diff --git a/conf/llm_factories.json b/conf/llm_factories.json
index fc656450c4..a2a232aa67 100644
--- a/conf/llm_factories.json
+++ b/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": "",
diff --git a/conf/model.md b/conf/model.md
index a5191abcd0..d67b12143b 100644
--- a/conf/model.md
+++ b/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 |
diff --git a/conf/models/cloudgpu.json b/conf/models/cloudgpu.json
new file mode 100644
index 0000000000..53a420a36e
--- /dev/null
+++ b/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
+ }
+ }
+ ]
+}
diff --git a/internal/entity/models/cloudgpu.go b/internal/entity/models/cloudgpu.go
new file mode 100644
index 0000000000..c3debb167b
--- /dev/null
+++ b/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
+}
diff --git a/internal/entity/models/cloudgpu_test.go b/internal/entity/models/cloudgpu_test.go
new file mode 100644
index 0000000000..840e72022c
--- /dev/null
+++ b/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])
+ }
+ }
+}
diff --git a/internal/entity/models/factory.go b/internal/entity/models/factory.go
index 7dcda49b28..32beef1341 100644
--- a/internal/entity/models/factory.go
+++ b/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:
diff --git a/web/src/assets/svg/llm/cloudgpu.svg b/web/src/assets/svg/llm/cloudgpu.svg
new file mode 100644
index 0000000000..bf817f0536
--- /dev/null
+++ b/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>
diff --git a/web/src/components/svg-icon.tsx b/web/src/components/svg-icon.tsx
index 0775a9601c..f85be88ed3 100644
--- a/web/src/components/svg-icon.tsx
+++ b/web/src/components/svg-icon.tsx
@@ -122,6 +122,7 @@ const svgIcons = [
LLMFactory.EmpirioLabs,
LLMFactory.AIStupidLevel,
LLMFactory.Runcrate,
+ LLMFactory.CloudGPU,
];
export const LlmIcon = ({
diff --git a/web/src/constants/llm.ts b/web/src/constants/llm.ts
index 1e509788b1..bd9a4f218b 100644
--- a/web/src/constants/llm.ts
+++ b/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',