feat(models): add Opper as a model vendor

README.md

@@ -60,7 +60,7 @@ Beyond the three modes:
 - **Memory and curation**: cross-session long-term memory keeps the profile, preferences and facts a user has confirmed. Folder uploads keep their directory tree, and retrieval chunks can be edited, diffed and rolled back.
 - **Data sources and formats**: auto-sync from Feishu wiki / Feishu Drive / Confluence / GitLab / Tencent IMA / Notion / Yuque / DingTalk Docs / RSS, with more on the way. 10+ document formats including PDF, Word, images, Excel and XMind; Office files are parsed in-process by anydoc.
 - **Channels and integrations**: Q&A in WeCom, Feishu, Slack, Telegram and other IM apps; an embed widget for external websites; a built-in MCP Server for Cursor, Claude and other AI tools; scoped API keys with a principal model for programmatic access.
-- **Models**: 27 built-in vendors with a generated model catalog, including OpenAI, DeepSeek, Qwen (Alibaba Cloud), Zhipu, Hunyuan, Gemini, MiniMax, NVIDIA, LiteLLM and Ollama.
+- **Models**: 28 built-in vendors with a generated model catalog, including OpenAI, DeepSeek, Qwen (Alibaba Cloud), Zhipu, Hunyuan, Gemini, MiniMax, NVIDIA, LiteLLM and Ollama.
 - **Permissions and operations**: multi-workspace RBAC (four roles, per-resource ownership, per-workspace audit log), several storage instances per workspace, a runtime task-queue dashboard with worker-pool governance, and Langfuse tracing for agent steps, token usage and pipelines.
 - **Deployment**: LLMs, vector databases and storage backends are all swappable. Deploy locally or on a private cloud and keep the data in your own environment.
 
@@ -313,7 +313,7 @@ A modular pipeline from document parsing, vectorization and retrieval to LLM inf
 
 | Component | Options |
 |-----------|---------|
-| [LLMs](https://weknora.weixin.qq.com/docs/03-features/06-models) | 27 built-in vendors, including OpenAI / Azure OpenAI / Anthropic / DeepSeek / Qwen / Zhipu / Hunyuan / Doubao / Gemini / MiniMax / NVIDIA / SiliconFlow / OpenRouter / LiteLLM / Ollama |
+| [LLMs](https://weknora.weixin.qq.com/docs/03-features/06-models) | 28 built-in vendors, including OpenAI / Azure OpenAI / Anthropic / DeepSeek / Qwen / Zhipu / Hunyuan / Doubao / Gemini / MiniMax / NVIDIA / SiliconFlow / OpenRouter / LiteLLM / Ollama |
 | Embeddings | Ollama / BGE / GTE / Zhipu / OpenAI-compatible APIs |
 | Vector databases | PostgreSQL (pgvector) / Elasticsearch / OpenSearch / Milvus / Weaviate / Qdrant / Apache Doris / Tencent VectorDB |
 | [Object storage](https://weknora.weixin.qq.com/docs/03-features/19-storage-backends) | Local / Tencent Cloud COS / MinIO / AWS S3 / Volcengine TOS / Alibaba Cloud OSS / Kingsoft Cloud KS3 / Huawei Cloud OBS |

README_CN.md

@@ -60,7 +60,7 @@ https://github.com/user-attachments/assets/5722b10d-d04d-49ed-a6cc-635a8c77d91f
 - **记忆与知识整理**：跨会话长期记忆保存用户确认过的个人信息、偏好与事实；文件夹上传保留原始目录结构；检索分块可以编辑、比对与回滚。
 - **数据源与格式**：飞书知识库 / 飞书云盘 / Confluence / GitLab / 腾讯 IMA / Notion / 语雀 / 钉钉文档 / RSS 自动同步，更多数据源持续接入中；覆盖 PDF、Word、图片、Excel、XMind 等十余种格式，Office 文档由 anydoc 在 Go 进程内解析。
 - **渠道与集成**：企业微信、飞书、Slack、Telegram 等 IM 频道内直接问答；网站嵌入 Widget 把智能体发布到外部站点；内置 MCP Server 供 Cursor、Claude 等 AI 工具连接；权限范围 API Key 与 Principal 模型用于程序化集成。
-- **模型**：内置 27 家模型厂商与自动生成的模型目录，兼容 OpenAI、DeepSeek、Qwen（阿里云）、智谱、混元、Gemini、MiniMax、NVIDIA、LiteLLM、Ollama 等。
+- **模型**：内置 28 家模型厂商与自动生成的模型目录，兼容 OpenAI、DeepSeek、Qwen（阿里云）、智谱、混元、Gemini、MiniMax、NVIDIA、LiteLLM、Ollama 等。
 - **权限与运维**：多空间 RBAC（四级角色、资源归属、空间审计日志）、每空间多实例存储后端、运行时任务队列面板与 Worker 池治理，并通过 Langfuse 追踪 Agent 步骤、Token 用量与任务流水线。
 - **部署**：大模型、向量数据库、存储后端均可替换，支持本地与私有云部署，数据留在你自己的环境中。
 
@@ -313,7 +313,7 @@ docker compose up -d    # 用新镜像重建容器
 
 | 组件 | 可选项 |
 |------|--------|
-| [模型厂商](https://weknora.weixin.qq.com/docs/03-features/06-models) | 内置 27 家，包括 OpenAI / Azure OpenAI / Anthropic / DeepSeek / Qwen（阿里云）/ 智谱 / 混元 / 豆包（火山引擎）/ Gemini / MiniMax / NVIDIA / SiliconFlow / OpenRouter / LiteLLM / Ollama |
+| [模型厂商](https://weknora.weixin.qq.com/docs/03-features/06-models) | 内置 28 家，包括 OpenAI / Azure OpenAI / Anthropic / DeepSeek / Qwen（阿里云）/ 智谱 / 混元 / 豆包（火山引擎）/ Gemini / MiniMax / NVIDIA / SiliconFlow / OpenRouter / LiteLLM / Ollama |
 | Embedding | Ollama / BGE / GTE / 智谱 / OpenAI 兼容接口 |
 | 向量数据库 | PostgreSQL (pgvector) / Elasticsearch / OpenSearch / Milvus / Weaviate / Qdrant / Apache Doris / 腾讯云 VectorDB |
 | [对象存储](https://weknora.weixin.qq.com/docs/03-features/19-storage-backends) | 本地 / 腾讯云 COS / MinIO / AWS S3 / 火山引擎 TOS / 阿里云 OSS / 金山云 KS3 / 华为云 OBS |

README_JA.md

@@ -60,7 +60,7 @@ https://github.com/user-attachments/assets/5722b10d-d04d-49ed-a6cc-635a8c77d91f
 - **メモリとナレッジ整理**：クロスセッション長期メモリが、ユーザーが確認したプロフィール・好み・事実を保持します。フォルダーアップロードは元のディレクトリ構造を保ち、検索チャンクは編集・差分比較・ロールバックできます。
 - **データソースとフォーマット**：Feishu ナレッジベース / Feishu クラウドドライブ / Confluence / GitLab / Tencent IMA / Notion / Yuque / DingTalk Docs / RSS の自動同期（順次拡充中）。PDF、Word、画像、Excel、XMind など 10 以上のフォーマットに対応し、Office 文書は anydoc でプロセス内解析します。
 - **チャネルと連携**：WeChat Work、Feishu、Slack、Telegram などの IM で直接 Q&A、ウェブサイト埋め込み Widget で外部サイトにエージェントを公開、組み込み MCP Server で Cursor や Claude などの AI ツールと接続、スコープ付き API キーと Principal モデルでプログラム連携。
-- **モデル**：27 の組み込みベンダーと自動生成のモデルカタログ。OpenAI、DeepSeek、Qwen（Alibaba Cloud）、Zhipu、Hunyuan、Gemini、MiniMax、NVIDIA、LiteLLM、Ollama などに対応。
+- **モデル**：28 の組み込みベンダーと自動生成のモデルカタログ。OpenAI、DeepSeek、Qwen（Alibaba Cloud）、Zhipu、Hunyuan、Gemini、MiniMax、NVIDIA、LiteLLM、Ollama などに対応。
 - **権限と運用**：マルチワークスペース RBAC（4 階層ロール、リソース所有権、ワークスペース監査ログ）、ワークスペースごとの複数ストレージインスタンス、ランタイムタスクキューダッシュボードと Worker プール統治、Langfuse による Agent ステップ・トークン消費・パイプラインのトレーシング。
 - **デプロイ**：LLM、ベクトルデータベース、ストレージバックエンドはすべて差し替え可能。ローカルまたはプライベートクラウドにデプロイし、データを自社環境に置けます。
 
@@ -313,7 +313,7 @@ docker compose up -d    # 新しいイメージでコンテナを再作成
 
 | コンポーネント | 選択肢 |
 |----------------|--------|
-| [大規模モデル](https://weknora.weixin.qq.com/docs/03-features/06-models) | 27 の組み込みベンダー。OpenAI / Azure OpenAI / Anthropic / DeepSeek / Qwen（Alibaba Cloud）/ Zhipu / Hunyuan / Doubao（Volcengine）/ Gemini / MiniMax / NVIDIA / SiliconFlow / OpenRouter / LiteLLM / Ollama など |
+| [大規模モデル](https://weknora.weixin.qq.com/docs/03-features/06-models) | 28 の組み込みベンダー。OpenAI / Azure OpenAI / Anthropic / DeepSeek / Qwen（Alibaba Cloud）/ Zhipu / Hunyuan / Doubao（Volcengine）/ Gemini / MiniMax / NVIDIA / SiliconFlow / OpenRouter / LiteLLM / Ollama など |
 | Embedding | Ollama / BGE / GTE / Zhipu / OpenAI 互換 API |
 | ベクトル DB | PostgreSQL (pgvector) / Elasticsearch / OpenSearch / Milvus / Weaviate / Qdrant / Apache Doris / Tencent VectorDB |
 | [オブジェクトストレージ](https://weknora.weixin.qq.com/docs/03-features/19-storage-backends) | ローカル / Tencent Cloud COS / MinIO / AWS S3 / 火山引擎 TOS / Alibaba Cloud OSS / 金山雲 KS3 / 華為雲 OBS |

README_KO.md

@@ -60,7 +60,7 @@ https://github.com/user-attachments/assets/5722b10d-d04d-49ed-a6cc-635a8c77d91f
 - **메모리와 지식 정리**: 크로스 세션 장기 메모리가 사용자가 확인한 프로필, 선호, 사실을 보관합니다. 폴더 업로드는 원래 디렉터리 구조를 유지하고, 검색 청크는 편집·비교·롤백할 수 있습니다.
 - **데이터 소스와 포맷**: Feishu 지식베이스 / Feishu 클라우드 드라이브 / Confluence / GitLab / Tencent IMA / Notion / Yuque / DingTalk Docs / RSS 자동 동기화(지속 확장 중). PDF, Word, 이미지, Excel, XMind 등 10가지 이상의 포맷을 지원하며, Office 문서는 anydoc으로 프로세스 내 파싱합니다.
 - **채널과 연동**: WeChat Work, Feishu, Slack, Telegram 등 IM에서 바로 Q&A, 웹사이트 임베드 Widget으로 외부 사이트에 에이전트 게시, 내장 MCP Server로 Cursor·Claude 등 AI 도구와 연결, 범위 지정 API 키와 Principal 모델로 프로그램 연동.
-- **모델**: 27개 내장 벤더와 자동 생성 모델 카탈로그. OpenAI, DeepSeek, Qwen(Alibaba Cloud), Zhipu, Hunyuan, Gemini, MiniMax, NVIDIA, LiteLLM, Ollama 등을 지원합니다.
+- **모델**: 28개 내장 벤더와 자동 생성 모델 카탈로그. OpenAI, DeepSeek, Qwen(Alibaba Cloud), Zhipu, Hunyuan, Gemini, MiniMax, NVIDIA, LiteLLM, Ollama 등을 지원합니다.
 - **권한과 운영**: 멀티 워크스페이스 RBAC(4단계 역할, 리소스 소유권, 워크스페이스 감사 로그), 워크스페이스별 다중 스토리지 인스턴스, 런타임 작업 큐 대시보드와 Worker 풀 거버넌스, Langfuse를 통한 Agent 단계·토큰 사용량·파이프라인 추적.
 - **배포**: LLM, 벡터 데이터베이스, 스토리지 백엔드를 모두 교체할 수 있습니다. 로컬이나 프라이빗 클라우드에 배포해 데이터를 자체 환경에 둘 수 있습니다.
 
@@ -323,7 +323,7 @@ docker compose up -d    # 새 이미지로 컨테이너 재생성
 
 | 구성 요소 | 선택지 |
 |-----------|--------|
-| [LLM](https://weknora.weixin.qq.com/docs/03-features/06-models) | 27개 내장 벤더. OpenAI / Azure OpenAI / Anthropic / DeepSeek / Qwen (Alibaba Cloud) / Zhipu / Hunyuan / Doubao (Volcengine) / Gemini / MiniMax / NVIDIA / SiliconFlow / OpenRouter / LiteLLM / Ollama 등 |
+| [LLM](https://weknora.weixin.qq.com/docs/03-features/06-models) | 28개 내장 벤더. OpenAI / Azure OpenAI / Anthropic / DeepSeek / Qwen (Alibaba Cloud) / Zhipu / Hunyuan / Doubao (Volcengine) / Gemini / MiniMax / NVIDIA / SiliconFlow / OpenRouter / LiteLLM / Ollama 등 |
 | Embedding | Ollama / BGE / GTE / Zhipu / OpenAI 호환 API |
 | 벡터 DB | PostgreSQL (pgvector) / Elasticsearch / OpenSearch / Milvus / Weaviate / Qdrant / Apache Doris / Tencent VectorDB |
 | [오브젝트 스토리지](https://weknora.weixin.qq.com/docs/03-features/19-storage-backends) | 로컬 / Tencent Cloud COS / MinIO / AWS S3 / Volcengine TOS / Alibaba Cloud OSS / Kingsoft Cloud KS3 / Huawei Cloud OBS |

internal/models/catalog/data/models.generated.json

@@ -4378,6 +4378,7 @@
         "source": "https://openrouter.ai/docs/api/api-reference/stt/create-transcription"
       }
     ],
+    "opper": [],
     "qianfan": [
       {
         "id": "ernie-5.0-thinking-preview",

internal/models/catalog/data/seed.json

@@ -3351,6 +3351,7 @@
         "source": "https://openrouter.ai/docs/api/api-reference/stt/create-transcription"
       }
     ],
+    "opper": [],
     "qianfan": [
       {
         "id": "ernie-5.0-thinking-preview",

internal/models/providers/builtin.go

@@ -25,6 +25,7 @@ func Builtins() []*Definition {
 		newNvidiaProvider(),
 		newOpenaiProvider(),
 		newOpenrouterProvider(),
+		newOpperProvider(),
 		newQianfanProvider(),
 		newQiniuProvider(),
 		newRequestyProvider(),

internal/models/providers/opper.go

@@ -0,0 +1,102 @@
+// Package providers registers the Opper gateway.
+//
+// Facts (https://docs.opper.ai/build/gateway/drop-in-sdks,
+// https://docs.opper.ai/build/gateway/completions,
+// https://docs.opper.ai/build/multimodal/vision-pdfs and the published
+// schema at https://api.opper.ai/v3/openapi.yaml):
+//   - chat, embeddings and the model list share the OpenAI-compatible base
+//     https://api.opper.ai/v3/compat, and auth is `Authorization: Bearer
+//     <key>` ("Opper reads the key from `Authorization: Bearer`");
+//   - a model id is a bare pool name (`claude-sonnet-4-6`, `gpt-5.5`), which
+//     Opper routes across the providers that serve the model, or
+//     `provider/model` (`anthropic/claude-sonnet-4-6`) to pin one route;
+//     GET /v3/compat/models lists both;
+//   - output cap is `max_tokens`, the field the chat guide documents ("Cap
+//     the response length"); `max_completion_tokens` is in the request
+//     schema without a description;
+//   - thinking is a flat `reasoning_effort` string: "Reasoning models (the
+//     GPT-5 family, Claude with extended thinking) also accept
+//     `reasoning_effort: "low" | "medium" | "high"`". minimal is not in that
+//     list, so the vendor level map sends it as low; xhigh and max stay off;
+//   - thinking text comes back in `reasoning_content`, and usage reports
+//     prompt_tokens_details.cached_tokens;
+//   - tool calls in a response carry `extra_content` (Gemini's
+//     thought_signature), and the request schema accepts it back on replayed
+//     assistant turns, so it is round-tripped opaquely;
+//   - images go in the chat messages as `image_url` parts ("Vision and PDF
+//     are model capabilities, so you send the media to a regular chat model
+//     that supports them"), so the VLM type shares the chat base URL;
+//   - embeddings take model, input, dimensions, encoding_format and user
+//     (https://docs.opper.ai/v3-api-reference/compatibility/create-embeddings);
+//   - the catalog ships no models: the pools follow the live list at
+//     GET /v3/compat/models, so the operator names the model.
+//
+// Unverified:
+//   - the rungs differ per model: GET /v3/models lists
+//     params.reasoning.supported (none, low, medium, high and xhigh for
+//     gpt-5.4-mini; low, medium, high and max for claude-sonnet-4-6). The
+//     vendor map keeps the three rungs the chat guide documents, and "off"
+//     sends no field, so the routed model's own default applies;
+//   - the docs do not say whether `temperature` is dropped for upstreams
+//     that reject it while reasoning, and with no catalog entries there is
+//     no per-model rule that turns sampling off.
+package providers
+
+import (
+	_ "embed"
+
+	"github.com/Tencent/WeKnora/internal/models/api"
+	"github.com/Tencent/WeKnora/internal/types"
+)
+
+//go:embed assets/opper.svg
+var opperIcon []byte
+
+// OpperID is the provider identifier stored on model rows.
+const OpperID = "opper"
+
+// OpperBaseURL is the OpenAI-compatible gateway endpoint.
+const OpperBaseURL = "https://api.opper.ai/v3/compat"
+
+func newOpperProvider() *Definition {
+	return &Definition{
+		ID:           OpperID,
+		Name:         "Opper",
+		Names:        map[string]string{"zh-CN": "Opper"},
+		Description:  "claude-sonnet-4-6, gpt-5.5, gemini-3.8-flash, deepseek-v4-pro, etc.",
+		Website:      "https://opper.ai",
+		Icon:         opperIcon,
+		API:          api.APIOpenAICompletions,
+		Order:        43,
+		RequiresAuth: true,
+		Auth:         AuthBearer,
+		URLPatterns:  []string{"api.opper.ai"},
+		DefaultBaseURLs: map[types.ModelType]string{
+			types.ModelTypeKnowledgeQA: OpperBaseURL,
+			types.ModelTypeEmbedding:   OpperBaseURL,
+			types.ModelTypeVLLM:        OpperBaseURL,
+		},
+		ModelTypes: []types.ModelType{
+			types.ModelTypeKnowledgeQA,
+			types.ModelTypeEmbedding,
+			types.ModelTypeVLLM,
+		},
+		Compat: VendorCompat{
+			Embeddings: api.EmbeddingsCompat{
+				SendEncodingFormat: api.Ptr(true),
+				DimensionsField:    api.Ptr("dimensions"),
+			},
+			OpenAICompletions: api.OpenAICompletionsCompat{
+				MaxTokensField:          api.Ptr("max_tokens"),
+				ThinkingFormat:          api.Ptr(api.ThinkingFormatOpenAI),
+				SupportsReasoningEffort: api.Ptr(true),
+				PromptCacheAccounting:   api.Ptr(true),
+				ToolCallExtraFields:     []string{"extra_content"},
+			},
+		},
+		// Opper documents low | medium | high; minimal folds onto low.
+		ThinkingLevels: api.ThinkingLevelMap{
+			api.ReasoningMinimal: api.StringPtr("low"),
+		},
+	}
+}

internal/models/providers/providers_test.go

@@ -18,13 +18,13 @@ var expectedIDs = []string{
 	"aliyun", "zhipu", "volcengine", "hunyuan", "siliconflow", "deepseek",
 	"minimax", "moonshot", "mimo", "modelscope", "qianfan", "qiniu", "longcat", "lkeap",
 	"openai", "azure_openai", "anthropic", "gemini",
-	"openrouter", "litellm", "requesty",
+	"openrouter", "litellm", "requesty", "opper",
 	"jina", "nvidia", "novita", "gpustack",
 }
 
 func TestAllVendorsRegistered(t *testing.T) {
-	if len(expectedIDs) != 27 {
-		t.Fatalf("expected 27 vendor ids in the spec, got %d", len(expectedIDs))
+	if len(expectedIDs) != 28 {
+		t.Fatalf("expected 28 vendor ids in the spec, got %d", len(expectedIDs))
 	}
 	for _, id := range expectedIDs {
 		v, ok := modelruntime.Get(id)
@@ -298,6 +298,14 @@ func TestFamilyExpectations(t *testing.T) {
 		"anthropic" {
 		t.Error("openrouter anthropic/* family should use anthropic cache_control")
 	}
+	// Opper documents max_tokens and a reasoning_effort of low, medium or
+	// high, so the weakest rung has to go out as low.
+	if r := resolve(t, "opper", "claude-sonnet-4-6"); r.OpenAICompletions.MaxTokensField != "max_tokens" ||
+		r.OpenAICompletions.ThinkingFormat != api.ThinkingFormatOpenAI ||
+		!r.OpenAICompletions.SupportsReasoningEffort ||
+		r.ThinkingLevels.Value(api.ReasoningMinimal) != "low" {
+		t.Error("opper should send max_tokens and a top-level reasoning_effort, with minimal sent as low")
+	}
 	if r := resolve(t, "siliconflow", "deepseek-ai/DeepSeek-V4-Pro"); !r.OpenAICompletions.SupportsReasoningEffort ||
 		r.ThinkingLevels.Value(api.ReasoningMax) != "max" {
 		t.Error("siliconflow DeepSeek-V4-Pro should grade effort high/max")