feat: add MiniMax as LLM provider with M3 model support (M3 default, M2.7 kept)

CLAUDE.md

@@ -143,7 +143,7 @@ make atlas-hash
 Before deployment, configure AI models in `backend/conf/model/`:
 1. Copy template from `backend/conf/model/template/`
 2. Set `id`, `meta.conn_config.api_key`, and `meta.conn_config.model`
-3. Supported providers: OpenAI, Volcengine Ark, Claude, Gemini, Qwen, DeepSeek, Ollama
+3. Supported providers: OpenAI, Volcengine Ark, Claude, Gemini, Qwen, DeepSeek, MiniMax, Ollama
 
 ## Testing Strategy
 

README.md

@@ -28,7 +28,7 @@ The backend of Coze Studio is developed using Golang, the frontend uses React +
 ## Feature list
 | **Module** | **Feature** |
 | --- | --- |
-| Model service | Manage the model list, integrate services such as OpenAI and Volcengine |
+| Model service | Manage the model list, integrate services such as OpenAI, Volcengine, Claude, DeepSeek, Gemini, Qwen, MiniMax, and Ollama |
 | Build agent | * Build, publish, and manage agent <br> * Support configuring workflows, knowledge bases, and other resources |
 | Build apps | * Create and publish apps <br> * Build business logic through workflows |
 | Build a workflow | Create, modify, publish, and delete workflows |

README.zh_CN.md

@@ -29,7 +29,7 @@ Coze Studio 的后端采用 Golang 开发，前端使用 React + TypeScript，
 ## 功能清单
 | **功能模块** | **功能点** |
 | --- | --- |
-| 模型服务 | 管理模型列表，可接入OpenAI、火山方舟 等在线或离线模型服务 |
+| 模型服务 | 管理模型列表，可接入 OpenAI、火山方舟、Claude、DeepSeek、Gemini、Qwen、MiniMax、Ollama 等在线或离线模型服务 |
 | 搭建智能体 | * 编排、发布、管理智能体 <br> * 支持配置工作流、知识库等资源 |
 | 搭建应用 | * 创建、发布应用 <br> * 通过工作流搭建业务逻辑 |
 | 搭建工作流 | 创建、修改、发布、删除工作流 |

backend/bizpkg/config/modelmgr/deprecate_model_get.go

@@ -204,6 +204,8 @@ func strProtocolToModelClass(protocol Protocol) developer_api.ModelClass {
 		modelClass = developer_api.ModelClass_Llama
 	case ProtocolQwen:
 		modelClass = developer_api.ModelClass_QWen
+	case ProtocolMiniMax:
+		modelClass = developer_api.ModelClass_MiniMax
 	default:
 		modelClass = developer_api.ModelClass_SEED
 	}
@@ -348,6 +350,7 @@ const (
 	ProtocolArk      Protocol = "ark"
 	ProtocolOllama   Protocol = "ollama"
 	ProtocolQwen     Protocol = "qwen"
+	ProtocolMiniMax  Protocol = "minimax"
 )
 
 type MultilingualText struct {

backend/bizpkg/config/modelmgr/mode_provider.go

@@ -107,6 +107,18 @@ func getModelProviderList() []*config.ModelProvider {
 			},
 			ModelClass: developer_api.ModelClass_QWen,
 		},
+		{
+			Name: &config.I18nText{
+				ZhCn: "MiniMax 模型",
+				EnUs: "MiniMax Model",
+			},
+			IconURI: "default_icon/minimax.png",
+			Description: &config.I18nText{
+				ZhCn: "MiniMax 模型家族",
+				EnUs: "MiniMax model family",
+			},
+			ModelClass: developer_api.ModelClass_MiniMax,
+		},
 	}
 }
 

backend/bizpkg/llm/modelbuilder/minimax.go

@@ -0,0 +1,105 @@
+/*
+ * Copyright 2025 coze-dev Authors
+ *
+ * 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 modelbuilder
+
+import (
+	"context"
+
+	"github.com/cloudwego/eino-ext/components/model/openai"
+
+	"github.com/coze-dev/coze-studio/backend/api/model/admin/config"
+	"github.com/coze-dev/coze-studio/backend/pkg/lang/ptr"
+)
+
+const defaultMiniMaxBaseURL = "https://api.minimax.io/v1"
+
+type minimaxModelBuilder struct {
+	cfg *config.Model
+}
+
+func newMiniMaxModelBuilder(cfg *config.Model) Service {
+	return &minimaxModelBuilder{
+		cfg: cfg,
+	}
+}
+
+func (m *minimaxModelBuilder) getDefaultConfig() *openai.ChatModelConfig {
+	return &openai.ChatModelConfig{
+		BaseURL: defaultMiniMaxBaseURL,
+		ResponseFormat: &openai.ChatCompletionResponseFormat{
+			Type: openai.ChatCompletionResponseFormatTypeText,
+		},
+	}
+}
+
+// clampTemperature ensures temperature stays within MiniMax's valid range (0, 1.0].
+// MiniMax rejects temperature=0, so we clamp it to a small positive value.
+func clampTemperature(t float32) float32 {
+	if t <= 0 {
+		return 0.01
+	}
+	if t > 1.0 {
+		return 1.0
+	}
+	return t
+}
+
+func (m *minimaxModelBuilder) applyParamsToConfig(conf *openai.ChatModelConfig, params *LLMParams) {
+	if params == nil {
+		return
+	}
+
+	if params.Temperature != nil {
+		clamped := clampTemperature(*params.Temperature)
+		conf.Temperature = ptr.Of(clamped)
+	}
+
+	if params.MaxTokens != 0 {
+		conf.MaxCompletionTokens = ptr.Of(params.MaxTokens)
+	}
+
+	if params.FrequencyPenalty != 0 {
+		conf.FrequencyPenalty = ptr.Of(params.FrequencyPenalty)
+	}
+
+	if params.PresencePenalty != 0 {
+		conf.PresencePenalty = ptr.Of(params.PresencePenalty)
+	}
+
+	conf.TopP = params.TopP
+
+	// MiniMax does not support response_format (JSON mode), always use text.
+	conf.ResponseFormat = &openai.ChatCompletionResponseFormat{
+		Type: openai.ChatCompletionResponseFormatTypeText,
+	}
+}
+
+func (m *minimaxModelBuilder) Build(ctx context.Context, params *LLMParams) (ToolCallingChatModel, error) {
+	base := m.cfg.Connection.BaseConnInfo
+
+	conf := m.getDefaultConfig()
+	conf.APIKey = base.APIKey
+	conf.Model = base.Model
+
+	if base.BaseURL != "" {
+		conf.BaseURL = base.BaseURL
+	}
+
+	m.applyParamsToConfig(conf, params)
+
+	return openai.NewChatModel(ctx, conf)
+}