From db05a64caade049f1e769f319981698b466d016d Mon Sep 17 00:00:00 2001 From: Prodman Devokadev Date: Tue, 1 Sep 2026 09:51:03 +0530 Subject: [PATCH] feat: add LiteLLM as a model provider (#2923) * feat(provider): add LiteLLM as an OpenAI-compatible provider * docs: use revive-compliant comment for ProviderLiteLLM const * fix(provider): stop defaulting LiteLLM to localhost under SSRF The catalog URL localhost:4000 is blocked by default SSRF and does not match DetectProvider, so UI/CLI could neither save nor auto-detect it. Use a placeholder host, keep thinking aligned with generic, and sync ModelSource across types, Swagger, and docs. --------- Co-authored-by: wizardchen --- client/model.go | 4 +- docs/BUILTIN_MODELS.md | 2 +- docs/api/model.md | 3 + docs/docs.go | 4 ++ docs/swagger.json | 4 ++ docs/swagger.yaml | 4 ++ docs/wiki/核心功能/内置模型管理.md | 2 +- frontend/src/components/ModelEditorDialog.vue | 11 ++++ frontend/src/i18n/locales/en-US.ts | 4 ++ frontend/src/i18n/locales/ko-KR.ts | 4 ++ frontend/src/i18n/locales/ru-RU.ts | 4 ++ frontend/src/i18n/locales/zh-CN.ts | 4 ++ frontend/src/utils/thinkingControl.test.ts | 1 + frontend/src/utils/thinkingControl.ts | 1 + internal/models/chat/provider.go | 8 ++- internal/models/chat/provider_test.go | 1 + internal/models/provider/litellm.go | 55 ++++++++++++++++ internal/models/provider/provider.go | 8 +++ internal/models/provider/provider_test.go | 62 +++++++++++++++++++ internal/types/model.go | 1 + .../01-getting-started/01-introduction.md | 2 +- website-docs/03-features/06-models.md | 3 +- .../06-development/03-extension-points.md | 2 +- 23 files changed, 187 insertions(+), 7 deletions(-) create mode 100644 internal/models/provider/litellm.go diff --git a/client/model.go b/client/model.go index b3a1ef7f..a6d67cc9 100644 --- a/client/model.go +++ b/client/model.go @@ -48,6 +48,7 @@ const ( ModelSourceSiliconFlow ModelSource = "siliconflow" // SiliconFlow model ModelSourceJina ModelSource = "jina" // Jina AI model ModelSourceOpenRouter ModelSource = "openrouter" // OpenRouter model + ModelSourceLiteLLM ModelSource = "litellm" // LiteLLM proxy model ModelSourceRequesty ModelSource = "requesty" // Requesty model ModelSourceNvidia ModelSource = "nvidia" // NVIDIA model ModelSourceNovita ModelSource = "novita" // Novita AI model @@ -64,7 +65,8 @@ func AllModelSources() []ModelSource { ModelSourceLocal, ModelSourceRemote, ModelSourceAliyun, ModelSourceZhipu, ModelSourceVolcengine, ModelSourceDeepseek, ModelSourceHunyuan, ModelSourceMinimax, ModelSourceOpenAI, ModelSourceGemini, ModelSourceMimo, ModelSourceSiliconFlow, - ModelSourceJina, ModelSourceOpenRouter, ModelSourceRequesty, ModelSourceNvidia, ModelSourceNovita, + ModelSourceJina, ModelSourceOpenRouter, ModelSourceLiteLLM, ModelSourceRequesty, + ModelSourceNvidia, ModelSourceNovita, ModelSourceAzureOpenAI, } } diff --git a/docs/BUILTIN_MODELS.md b/docs/BUILTIN_MODELS.md index 6816cd8b..b4fe9bba 100644 --- a/docs/BUILTIN_MODELS.md +++ b/docs/BUILTIN_MODELS.md @@ -167,7 +167,7 @@ services: ### 方式二:直接 SQL 插入 -支持的 provider:`generic`(自定义)、`openai`、`aliyun`、`zhipu`、`volcengine`、`hunyuan`、`deepseek`、`minimax`、`mimo`、`siliconflow`、`jina`、`openrouter`、`requesty`、`gemini`、`modelscope`、`moonshot`、`qianfan`、`qiniu`、`longcat`、`gpustack` +支持的 provider:`generic`(自定义)、`openai`、`aliyun`、`zhipu`、`volcengine`、`hunyuan`、`deepseek`、`minimax`、`mimo`、`siliconflow`、`jina`、`openrouter`、`litellm`、`requesty`、`gemini`、`modelscope`、`moonshot`、`qianfan`、`qiniu`、`longcat`、`gpustack` ```sql -- 示例:LLM 内置模型 diff --git a/docs/api/model.md b/docs/api/model.md index 9024d161..61f6a165 100644 --- a/docs/api/model.md +++ b/docs/api/model.md @@ -33,6 +33,7 @@ WeKnora 支持多种主流 AI 模型服务商,在创建模型时可通过 `par | `siliconflow` | 硅基流动 SiliconFlow | Chat, Embedding, Rerank, VLLM | | `jina` | Jina | Embedding, Rerank | | `openrouter` | OpenRouter | Chat, VLLM | +| `litellm` | LiteLLM (self-hosted proxy) | Chat, Embedding, VLLM | | `requesty` | Requesty | Chat, Embedding, VLLM | | `gemini` | Google Gemini | Chat | | `modelscope` | 魔搭 ModelScope | Chat, Embedding, VLLM | @@ -43,6 +44,8 @@ WeKnora 支持多种主流 AI 模型服务商,在创建模型时可通过 `par | `gpustack` | GPUStack | Chat, Embedding, Rerank, VLLM | > 实际可用的服务商以 `GET /models/providers` 返回为准。 +> +> `litellm` 的目录默认地址是占位符 `http://your_litellm_proxy/v1`,保存前请换成实际可解析的主机名。`localhost` / `127.0.0.1` / `host.docker.internal` 默认会被 SSRF 拦截;自托管时把该主机加入 `SSRF_WHITELIST`。创建模型时 `source` 仍为 `remote`,厂商写在 `parameters.provider`。 ## GET `/models/providers` - 获取模型服务商列表 diff --git a/docs/docs.go b/docs/docs.go index acc16b18..0e42773c 100644 --- a/docs/docs.go +++ b/docs/docs.go @@ -19908,6 +19908,7 @@ const docTemplate = `{ "siliconflow", "jina", "openrouter", + "litellm", "requesty", "nvidia", "novita", @@ -19927,6 +19928,7 @@ const docTemplate = `{ "ModelSourceNvidia": "NVIDIA model", "ModelSourceOpenAI": "OpenAI model", "ModelSourceOpenRouter": "OpenRouter model", + "ModelSourceLiteLLM": "LiteLLM proxy model", "ModelSourceRemote": "Remote model", "ModelSourceRequesty": "Requesty model", "ModelSourceSiliconFlow": "SiliconFlow model", @@ -19948,6 +19950,7 @@ const docTemplate = `{ "SiliconFlow model", "Jina AI model", "OpenRouter model", + "LiteLLM proxy model", "Requesty model", "NVIDIA model", "Novita AI model", @@ -19968,6 +19971,7 @@ const docTemplate = `{ "ModelSourceSiliconFlow", "ModelSourceJina", "ModelSourceOpenRouter", + "ModelSourceLiteLLM", "ModelSourceRequesty", "ModelSourceNvidia", "ModelSourceNovita", diff --git a/docs/swagger.json b/docs/swagger.json index 46ee83e6..e7f9e01e 100644 --- a/docs/swagger.json +++ b/docs/swagger.json @@ -19901,6 +19901,7 @@ "siliconflow", "jina", "openrouter", + "litellm", "requesty", "nvidia", "novita", @@ -19920,6 +19921,7 @@ "ModelSourceNvidia": "NVIDIA model", "ModelSourceOpenAI": "OpenAI model", "ModelSourceOpenRouter": "OpenRouter model", + "ModelSourceLiteLLM": "LiteLLM proxy model", "ModelSourceRemote": "Remote model", "ModelSourceRequesty": "Requesty model", "ModelSourceSiliconFlow": "SiliconFlow model", @@ -19941,6 +19943,7 @@ "SiliconFlow model", "Jina AI model", "OpenRouter model", + "LiteLLM proxy model", "Requesty model", "NVIDIA model", "Novita AI model", @@ -19961,6 +19964,7 @@ "ModelSourceSiliconFlow", "ModelSourceJina", "ModelSourceOpenRouter", + "ModelSourceLiteLLM", "ModelSourceRequesty", "ModelSourceNvidia", "ModelSourceNovita", diff --git a/docs/swagger.yaml b/docs/swagger.yaml index bee859ab..49ed133c 100644 --- a/docs/swagger.yaml +++ b/docs/swagger.yaml @@ -2468,6 +2468,7 @@ definitions: - siliconflow - jina - openrouter + - litellm - requesty - nvidia - novita @@ -2487,6 +2488,7 @@ definitions: ModelSourceNvidia: NVIDIA model ModelSourceOpenAI: OpenAI model ModelSourceOpenRouter: OpenRouter model + ModelSourceLiteLLM: LiteLLM proxy model ModelSourceRemote: Remote model ModelSourceRequesty: Requesty model ModelSourceSiliconFlow: SiliconFlow model @@ -2507,6 +2509,7 @@ definitions: - SiliconFlow model - Jina AI model - OpenRouter model + - LiteLLM proxy model - Requesty model - NVIDIA model - Novita AI model @@ -2526,6 +2529,7 @@ definitions: - ModelSourceSiliconFlow - ModelSourceJina - ModelSourceOpenRouter + - ModelSourceLiteLLM - ModelSourceRequesty - ModelSourceNvidia - ModelSourceNovita diff --git a/docs/wiki/核心功能/内置模型管理.md b/docs/wiki/核心功能/内置模型管理.md index 8526c355..871b9676 100644 --- a/docs/wiki/核心功能/内置模型管理.md +++ b/docs/wiki/核心功能/内置模型管理.md @@ -30,7 +30,7 @@ source: BUILTIN_MODELS.md - 模型参数(parameters):包括 base_url、api_key、provider 等 - 空间ID(tenant_id):建议使用小于10000的空间ID -**支持的服务商(provider)**:`generic`、`openai`、`aliyun`、`zhipu`、`volcengine`、`hunyuan`、`deepseek`、`minimax`、`mimo`、`siliconflow`、`jina`、`openrouter`、`requesty`、`gemini`、`modelscope`、`moonshot`、`qianfan`、`qiniu`、`longcat`、`gpustack` +**支持的服务商(provider)**:`generic`、`openai`、`aliyun`、`zhipu`、`volcengine`、`hunyuan`、`deepseek`、`minimax`、`mimo`、`siliconflow`、`jina`、`openrouter`、`litellm`、`requesty`、`gemini`、`modelscope`、`moonshot`、`qianfan`、`qiniu`、`longcat`、`gpustack` ### 2. 执行 SQL 插入语句 diff --git a/frontend/src/components/ModelEditorDialog.vue b/frontend/src/components/ModelEditorDialog.vue index 47834661..94fcb12d 100644 --- a/frontend/src/components/ModelEditorDialog.vue +++ b/frontend/src/components/ModelEditorDialog.vue @@ -550,6 +550,17 @@ const fallbackProviderOptions = computed(() => [ description: t('model.editor.providers.openrouter.description'), modelTypes: ['chat', 'embedding'] }, + { + value: 'litellm', + label: t('model.editor.providers.litellm.label'), + defaultUrls: { + chat: 'http://your_litellm_proxy/v1', + embedding: 'http://your_litellm_proxy/v1', + vllm: 'http://your_litellm_proxy/v1' + }, + description: t('model.editor.providers.litellm.description'), + modelTypes: ['chat', 'embedding', 'vllm'] + }, { value: 'requesty', label: t('model.editor.providers.requesty.label'), diff --git a/frontend/src/i18n/locales/en-US.ts b/frontend/src/i18n/locales/en-US.ts index 8b7bfaa9..63fb7123 100755 --- a/frontend/src/i18n/locales/en-US.ts +++ b/frontend/src/i18n/locales/en-US.ts @@ -4272,6 +4272,10 @@ export default { label: 'OpenRouter', description: 'openai/gpt-5.2-chat, google/gemini-3-flash-preview, etc.' }, + litellm: { + label: 'LiteLLM', + description: 'Self-hosted proxy to 100+ providers (OpenAI, Anthropic, Gemini, Bedrock, etc.). Replace the placeholder URL; loopback hosts need SSRF_WHITELIST.' + }, requesty: { label: 'Requesty', description: 'openai/gpt-4o-mini, anthropic/claude-sonnet-4-5, etc.' diff --git a/frontend/src/i18n/locales/ko-KR.ts b/frontend/src/i18n/locales/ko-KR.ts index c9d44dbb..8deaf14a 100755 --- a/frontend/src/i18n/locales/ko-KR.ts +++ b/frontend/src/i18n/locales/ko-KR.ts @@ -2452,6 +2452,10 @@ export default { label: 'OpenRouter', description: 'openai/gpt-5.2-chat, google/gemini-3-flash-preview 등' }, + litellm: { + label: 'LiteLLM', + description: '자체 호스팅 프록시로 OpenAI, Anthropic, Gemini, Bedrock 등 100+ 공급자를 연결합니다. 플레이스홀더 URL을 실제 주소로 바꾸세요. localhost는 SSRF_WHITELIST에 추가해야 합니다.' + }, zhipu: { label: 'Zhipu BigModel', description: 'glm-4.7, embedding-3, rerank, etc.' diff --git a/frontend/src/i18n/locales/ru-RU.ts b/frontend/src/i18n/locales/ru-RU.ts index a31edb83..c972d3dc 100755 --- a/frontend/src/i18n/locales/ru-RU.ts +++ b/frontend/src/i18n/locales/ru-RU.ts @@ -2452,6 +2452,10 @@ export default { label: 'OpenRouter', description: 'openai/gpt-5.2-chat, google/gemini-3-flash-preview, etc.' }, + litellm: { + label: 'LiteLLM', + description: 'Self-hosted прокси к 100+ провайдерам (OpenAI, Anthropic, Gemini, Bedrock и др.). Замените URL-заглушку; localhost нужно добавить в SSRF_WHITELIST.' + }, zhipu: { label: 'Zhipu BigModel', description: 'glm-4.7, embedding-3, rerank, etc.' diff --git a/frontend/src/i18n/locales/zh-CN.ts b/frontend/src/i18n/locales/zh-CN.ts index c849ecd2..30cf1922 100755 --- a/frontend/src/i18n/locales/zh-CN.ts +++ b/frontend/src/i18n/locales/zh-CN.ts @@ -2454,6 +2454,10 @@ export default { label: 'OpenRouter', description: 'openai/gpt-5.2-chat, google/gemini-3-flash-preview, etc.' }, + litellm: { + label: 'LiteLLM', + description: '自托管代理,统一接入 OpenAI、Anthropic、Gemini、Bedrock 等 100+ 厂商。请将占位 URL 换成可访问地址;localhost 需加入 SSRF_WHITELIST。' + }, zhipu: { label: '智谱 BigModel', description: 'glm-4.7, embedding-3, rerank, etc.' diff --git a/frontend/src/utils/thinkingControl.test.ts b/frontend/src/utils/thinkingControl.test.ts index 6502bc0d..1840e34e 100644 --- a/frontend/src/utils/thinkingControl.test.ts +++ b/frontend/src/utils/thinkingControl.test.ts @@ -8,6 +8,7 @@ test('defaultThinkingControl matches backend provider adapters', () => { const cases: Array<[string, string, ReturnType]> = [ ['generic', 'anything', 'chat_template_kwargs'], ['nvidia', 'anything', 'chat_template_kwargs'], + ['litellm', 'anything', 'chat_template_kwargs'], ['volcengine', 'doubao', 'thinking_type'], ['aliyun', 'qwen3-32b', 'enable_thinking'], ['aliyun', 'qwen-plus', 'enable_thinking'], diff --git a/frontend/src/utils/thinkingControl.ts b/frontend/src/utils/thinkingControl.ts index 122bced7..9ec34cbb 100644 --- a/frontend/src/utils/thinkingControl.ts +++ b/frontend/src/utils/thinkingControl.ts @@ -47,6 +47,7 @@ export function defaultThinkingControl( return 'thinking_type' case 'generic': case 'nvidia': + case 'litellm': return 'chat_template_kwargs' case 'volcengine': return 'thinking_type' diff --git a/internal/models/chat/provider.go b/internal/models/chat/provider.go index c596a354..be1a4f79 100644 --- a/internal/models/chat/provider.go +++ b/internal/models/chat/provider.go @@ -156,7 +156,7 @@ func (deepseekProvider) ShapeRequest(req *openai.ChatCompletionRequest, opts *Ch } } -// --- Generic (vLLM) / NVIDIA: thinking via chat_template_kwargs --- +// --- Generic (vLLM) / NVIDIA / LiteLLM: thinking via chat_template_kwargs --- type genericProvider struct{ baseProvider } @@ -168,6 +168,11 @@ type nvidiaProvider struct{ baseProvider } func (nvidiaProvider) Name() provider.ProviderName { return provider.ProviderNvidia } func (nvidiaProvider) Thinking() ThinkingStrategy { return chatTemplateKwargs{} } +type liteLLMProvider struct{ baseProvider } + +func (liteLLMProvider) Name() provider.ProviderName { return provider.ProviderLiteLLM } +func (liteLLMProvider) Thinking() ThinkingStrategy { return chatTemplateKwargs{} } + // --- Gemini OpenAI compatibility: tool thought signatures live in extra_content --- type geminiProvider struct{ baseProvider } @@ -277,6 +282,7 @@ var providerRegistry = []providerAdapter{ lkeapProvider{}, deepseekProvider{}, genericProvider{}, + liteLLMProvider{}, geminiProvider{}, volcengineProvider{}, nvidiaProvider{}, diff --git a/internal/models/chat/provider_test.go b/internal/models/chat/provider_test.go index bc1049fb..68848572 100644 --- a/internal/models/chat/provider_test.go +++ b/internal/models/chat/provider_test.go @@ -26,6 +26,7 @@ func TestResolveProvider(t *testing.T) { {"lkeap r1 falls back", provider.ProviderLKEAP, "deepseek-r1", baseProvider{}}, {"qwen thinking", provider.ProviderAliyun, "qwen3-32b", qwenThinkingProvider{}}, {"generic", provider.ProviderGeneric, "anything", genericProvider{}}, + {"litellm", provider.ProviderLiteLLM, "anything", liteLLMProvider{}}, {"gemini", provider.ProviderGemini, "gemini-3-flash-preview", geminiProvider{}}, {"nvidia", provider.ProviderNvidia, "anything", nvidiaProvider{}}, {"volcengine", provider.ProviderVolcengine, "doubao", volcengineProvider{}}, diff --git a/internal/models/provider/litellm.go b/internal/models/provider/litellm.go new file mode 100644 index 00000000..c06ca6ab --- /dev/null +++ b/internal/models/provider/litellm.go @@ -0,0 +1,55 @@ +package provider + +import ( + "fmt" + + "github.com/Tencent/WeKnora/internal/types" +) + +const ( + // LiteLLMBaseURL is a placeholder the user must replace with a reachable + // LiteLLM proxy. Loopback defaults (localhost:4000) are rejected by SSRF + // unless the host is added to SSRF_WHITELIST; Docker also cannot reach a + // proxy on the host via localhost. The hostname contains "litellm" so + // DetectProvider recognises the catalog default. + LiteLLMBaseURL = "http://your_litellm_proxy/v1" +) + +// LiteLLMProvider implements the Provider interface for LiteLLM +// (https://github.com/BerriAI/litellm). LiteLLM exposes a single +// OpenAI-compatible endpoint that routes to 100+ providers (OpenAI, Anthropic, +// Gemini, Bedrock, Vertex, Azure, ...) behind one base URL and key, so it plugs +// into WeKnora's OpenAI-compatible transport like the other gateway providers. +type LiteLLMProvider struct{} + +func init() { + Register(&LiteLLMProvider{}) +} + +// Info returns the metadata for the LiteLLM provider. +func (p *LiteLLMProvider) Info() ProviderInfo { + return ProviderInfo{ + Name: ProviderLiteLLM, + DisplayName: "LiteLLM", + Description: "Self-hosted LiteLLM proxy: one OpenAI-compatible endpoint to 100+ providers.", + DefaultURLs: map[types.ModelType]string{ + types.ModelTypeKnowledgeQA: LiteLLMBaseURL, + types.ModelTypeEmbedding: LiteLLMBaseURL, + types.ModelTypeVLLM: LiteLLMBaseURL, + }, + ModelTypes: []types.ModelType{ + types.ModelTypeKnowledgeQA, + types.ModelTypeEmbedding, + types.ModelTypeVLLM, + }, + RequiresAuth: true, + } +} + +// ValidateConfig validates the LiteLLM provider configuration. +func (p *LiteLLMProvider) ValidateConfig(config *Config) error { + if config.APIKey == "" { + return fmt.Errorf("API key is required for LiteLLM provider") + } + return nil +} diff --git a/internal/models/provider/provider.go b/internal/models/provider/provider.go index 0cc13342..d9603662 100644 --- a/internal/models/provider/provider.go +++ b/internal/models/provider/provider.go @@ -23,6 +23,8 @@ const ( ProviderZhipu ProviderName = "zhipu" // OpenRouter ProviderOpenRouter ProviderName = "openrouter" + // ProviderLiteLLM is the LiteLLM self-hosted proxy (OpenAI-compatible gateway to 100+ providers). + ProviderLiteLLM ProviderName = "litellm" // Requesty ProviderRequesty ProviderName = "requesty" // 硅基流动 @@ -85,6 +87,7 @@ func AllProviders() []ProviderName { ProviderAnthropic, ProviderGemini, ProviderOpenRouter, + ProviderLiteLLM, ProviderRequesty, ProviderJina, ProviderMimo, @@ -227,6 +230,11 @@ func DetectProvider(baseURL string) ProviderName { return ProviderZhipu case containsAny(baseURL, "openrouter.ai"): return ProviderOpenRouter + // Hostname/path containing "litellm" (including the catalog placeholder + // your_litellm_proxy). Loopback URLs such as localhost:4000 stay generic + // because they are SSRF-blocked unless explicitly whitelisted. + case containsAny(baseURL, "litellm"): + return ProviderLiteLLM case containsAny(baseURL, "router.requesty.ai", "requesty.ai"): return ProviderRequesty case containsAny(baseURL, "siliconflow.cn"): diff --git a/internal/models/provider/provider_test.go b/internal/models/provider/provider_test.go index bf13b22f..a0a9320e 100644 --- a/internal/models/provider/provider_test.go +++ b/internal/models/provider/provider_test.go @@ -38,6 +38,9 @@ func TestDetectProvider(t *testing.T) { {"https://api.openai.com/v1", ProviderOpenAI}, {"https://api.anthropic.com/v1", ProviderAnthropic}, {"https://openrouter.ai/api/v1", ProviderOpenRouter}, + {"https://litellm.example.com/v1", ProviderLiteLLM}, + {LiteLLMBaseURL, ProviderLiteLLM}, + {"http://localhost:4000/v1", ProviderGeneric}, {"https://router.requesty.ai/v1", ProviderRequesty}, {"https://dashscope.aliyuncs.com/compatible-mode/v1", ProviderAliyun}, {"https://open.bigmodel.cn/api/paas/v4", ProviderZhipu}, @@ -317,3 +320,62 @@ func TestListByModelType(t *testing.T) { assert.True(t, found, "Gemini should support embedding via the native Gemini API") }) } + +func TestLiteLLMProviderValidation(t *testing.T) { + p := &LiteLLMProvider{} + + t.Run("valid config", func(t *testing.T) { + config := &Config{ + APIKey: "test-key", + ModelName: "gpt-4.1-mini", + } + err := p.ValidateConfig(config) + assert.NoError(t, err) + }) + + t.Run("missing API key", func(t *testing.T) { + config := &Config{ + ModelName: "gpt-4.1-mini", + } + err := p.ValidateConfig(config) + assert.Error(t, err) + assert.Contains(t, err.Error(), "API key") + }) + + t.Run("info", func(t *testing.T) { + info := p.Info() + assert.Equal(t, ProviderLiteLLM, info.Name) + assert.Equal(t, "LiteLLM", info.DisplayName) + assert.True(t, info.RequiresAuth) + assert.Equal(t, LiteLLMBaseURL, info.DefaultURLs[types.ModelTypeKnowledgeQA]) + assert.Equal(t, LiteLLMBaseURL, info.DefaultURLs[types.ModelTypeEmbedding]) + assert.Equal(t, LiteLLMBaseURL, info.DefaultURLs[types.ModelTypeVLLM]) + assert.Contains(t, info.ModelTypes, types.ModelTypeKnowledgeQA) + assert.Contains(t, info.ModelTypes, types.ModelTypeEmbedding) + assert.Contains(t, info.ModelTypes, types.ModelTypeVLLM) + }) + + t.Run("registered and listed", func(t *testing.T) { + got, ok := Get(ProviderLiteLLM) + require.True(t, ok) + require.NotNil(t, got) + + foundChat := false + for _, info := range ListByModelType(types.ModelTypeKnowledgeQA) { + if info.Name == ProviderLiteLLM { + foundChat = true + break + } + } + assert.True(t, foundChat, "LiteLLM should appear for chat models") + + foundEmbed := false + for _, info := range ListByModelType(types.ModelTypeEmbedding) { + if info.Name == ProviderLiteLLM { + foundEmbed = true + break + } + } + assert.True(t, foundEmbed, "LiteLLM should appear for embedding models") + }) +} diff --git a/internal/types/model.go b/internal/types/model.go index 2af98283..17606043 100644 --- a/internal/types/model.go +++ b/internal/types/model.go @@ -49,6 +49,7 @@ const ( ModelSourceSiliconFlow ModelSource = "siliconflow" // SiliconFlow model ModelSourceJina ModelSource = "jina" // Jina AI model ModelSourceOpenRouter ModelSource = "openrouter" // OpenRouter model + ModelSourceLiteLLM ModelSource = "litellm" // LiteLLM proxy model ModelSourceRequesty ModelSource = "requesty" // Requesty model ModelSourceNvidia ModelSource = "nvidia" // NVIDIA model ModelSourceNovita ModelSource = "novita" // Novita AI model diff --git a/website-docs/01-getting-started/01-introduction.md b/website-docs/01-getting-started/01-introduction.md index d8ee8ae1..d0fca60c 100644 --- a/website-docs/01-getting-started/01-introduction.md +++ b/website-docs/01-getting-started/01-introduction.md @@ -56,7 +56,7 @@ WeKnora(维娜拉)是腾讯开源的知识库问答系统,做的事情是 | --- | --- | | 会话 Session | 一次多轮对话。记录 `LastRequestState`(上次提问时选中的 Agent、模型、知识库范围、Web 搜索、MCP 服务),重开会话时恢复;上下文压缩策略(`sliding_window` / `smart` LLM 摘要)来自 `ContextConfig` | | 消息 Message | `user` / `assistant` 角色消息,支持图片、附件、@提及(知识库/文档/标签/MCP/Skill),并统计 `TokenUsage`(含 prompt cache 命中情况) | -| 模型 Model | 模型注册项。`Type`:`KnowledgeQA`(对话 LLM)/ `Embedding` / `Rerank` / `VLLM`(视觉)/ `ASR`(语音);`Source`:`local`(Ollama)、`remote` 及 `openai`、`azure_openai`、`gemini`、`deepseek`、`aliyun`、`zhipu`、`volcengine`、`hunyuan`、`siliconflow`、`openrouter`、`jina` 等厂商;`ManagedBy: "yaml"` 表示由 `config/builtin_models.yaml` 声明式管理 | +| 模型 Model | 模型注册项。`Type`:`KnowledgeQA`(对话 LLM)/ `Embedding` / `Rerank` / `VLLM`(视觉)/ `ASR`(语音);`Source`:`local`(Ollama)、`remote` 及 `openai`、`azure_openai`、`gemini`、`deepseek`、`aliyun`、`zhipu`、`volcengine`、`hunyuan`、`siliconflow`、`openrouter`、`litellm`、`jina` 等厂商;`ManagedBy: "yaml"` 表示由 `config/builtin_models.yaml` 声明式管理 | | Agent(自定义智能体) CustomAgent | 两种模式:`quick-answer`(经典 RAG 管线)与 `smart-reasoning`(ReAct 多步推理 + 工具调用)。smart-reasoning 下有类型预设 `AgentType`:`rag-qa` / `wiki-qa` / `hybrid-rag-wiki` / `data-analysis` / `custom`(定义见 `config/agent_type_presets.yaml`) | | 内置 Agent | 开箱可用:`builtin-quick-answer`(快速问答)、`builtin-smart-reasoning`(智能推理)、`builtin-data-analyst`(数据分析)、`builtin-wiki-researcher`(Wiki 研究员)、`builtin-wiki-fixer`(Wiki 修复员)等 | | MCP 服务 MCPService | Model Context Protocol 工具接入:`sse` / `http-streamable` / `stdio` 三种传输;认证支持 API Key / Bearer / OAuth2;Agent 可按 `all` / `selected` / `none` 选用其工具 | diff --git a/website-docs/03-features/06-models.md b/website-docs/03-features/06-models.md index 5eea1090..cda2ab0e 100644 --- a/website-docs/03-features/06-models.md +++ b/website-docs/03-features/06-models.md @@ -57,7 +57,7 @@ type Provider interface { ### 支持的厂商清单 -`AllProviders()`(`provider/provider.go`)返回的完整列表(共 26 个,每个厂商在自己的文件里 `init()` 注册)。表格最后一行的 Ollama 不在其中,它走 `source=local` 这条独立路径,列在这里只为方便对照: +`AllProviders()`(`provider/provider.go`)返回的完整列表(共 27 个,每个厂商在自己的文件里 `init()` 注册)。表格最后一行的 Ollama 不在其中,它走 `source=local` 这条独立路径,列在这里只为方便对照: | Provider 标识 | 名称 | 说明 | |---------------|------|------| @@ -78,6 +78,7 @@ type Provider interface { | `anthropic` | Anthropic Claude | 独立 Messages 协议实现 | | `gemini` | Google Gemini | Embedding 走专用 API | | `openrouter` | OpenRouter | | +| `litellm` | LiteLLM(自托管 OpenAI 兼容代理) | 默认 URL 为占位符,loopback 需加入 `SSRF_WHITELIST` | | `requesty` | Requesty | | | `jina` | Jina AI | Embedding 与 Rerank | | `mimo` | 小米 MiMo | | diff --git a/website-docs/06-development/03-extension-points.md b/website-docs/06-development/03-extension-points.md index 597df991..687dce40 100644 --- a/website-docs/06-development/03-extension-points.md +++ b/website-docs/06-development/03-extension-points.md @@ -342,7 +342,7 @@ type Reranker interface { ### 现有实现 -`internal/models/provider/provider.go` 中已定义 26 个 `ProviderName` 常量:openai、anthropic、aliyun、zhipu、openrouter、requesty、siliconflow、jina、generic、deepseek、gemini、volcengine、hunyuan、minimax、mimo、gpustack、moonshot、modelscope、qianfan、qiniu、longcat、lkeap、nvidia 等。具体 Provider 实现分布在 `internal/models/provider/` 下的各文件(如 `zhipu.go`、`gemini.go`、`hunyuan.go`、`generic.go`);特殊 embedding 实现如 `internal/models/embedding/jina.go`、`volcengine.go`、`nvidia.go`。 +`internal/models/provider/provider.go` 中已定义 27 个 `ProviderName` 常量:openai、anthropic、aliyun、zhipu、openrouter、litellm、requesty、siliconflow、jina、generic、deepseek、gemini、volcengine、hunyuan、minimax、mimo、gpustack、moonshot、modelscope、qianfan、qiniu、longcat、lkeap、nvidia 等。具体 Provider 实现分布在 `internal/models/provider/` 下的各文件(如 `zhipu.go`、`gemini.go`、`hunyuan.go`、`generic.go`);特殊 embedding 实现如 `internal/models/embedding/jina.go`、`volcengine.go`、`nvidia.go`。 ### 新增步骤