RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.
Go agent-harnessagent-native-databaseagentic-aiagentic-retrievalagentic-searchaiai-agentscontext-enginecontext-engineeringcontext-managementharness-engineeringknowledge-compilationragretrieval-augmented-generationretrieval-harnesssearch-harness
feat: add llmman as a model provider (#19030)
Eric Curtinc051f40593fb
diff --git a/conf/llm_factories.json b/conf/llm_factories.jsonindex 7789232644..11e9e21f8a 100644--- a/conf/llm_factories.json+++ b/conf/llm_factories.json@@ -3010,6 +3010,13 @@],"url": "https://integrate.api.nvidia.com/v1"},+ {+ "name": "llmman",+ "logo": "",+ "tags": "LLM,TEXT EMBEDDING",+ "status": "1",+ "llm": []+ },{"name": "LM-Studio","logo": "",diff --git a/conf/models/llmman.json b/conf/models/llmman.jsonnew file mode 100644index 0000000000..804dc96473--- /dev/null+++ b/conf/models/llmman.json@@ -0,0 +1,9 @@+{+ "name": "llmman",+ "url_suffix": {+ "chat": "chat/completions",+ "models": "models",+ "embedding": "embeddings"+ },+ "class": "local"+}diff --git a/docs/guides/models/supported_models.mdx b/docs/guides/models/supported_models.mdxindex b7c3d9bfb6..5608857452 100644--- a/docs/guides/models/supported_models.mdx+++ b/docs/guides/models/supported_models.mdx@@ -46,6 +46,7 @@ A complete list of model providers supported by RAGFlow, which will continue to| HuggingFace | `https://huggingface.co` || Jiekou.AI | `https://jiekou.ai` || Jina | `https://jina.ai` |+| llmman | `https://github.com/llmmanorg/llmman` || LM-Studio | `https://lmstudio.ai` || LocalAI | `https://localai.io` || LongCat | `https://longcat.chat` |diff --git a/internal/entity/models/factory.go b/internal/entity/models/factory.goindex 64853c1e70..efb6e1d662 100644--- a/internal/entity/models/factory.go+++ b/internal/entity/models/factory.go@@ -63,6 +63,8 @@ func (f *ModelFactory) CreateModelDriver(providerName string, baseURL map[stringreturn NewXAIModel(baseURL, urlSuffix), nilcase "lm-studio":return NewLmStudioModel(baseURL, urlSuffix), nil+ case "llmman":+ return NewLlmmanModel(baseURL, urlSuffix), nilcase "ollama":return NewOllamaModel(baseURL, urlSuffix), nilcase "openai":diff --git a/internal/entity/models/llmman.go b/internal/entity/models/llmman.gonew file mode 100644index 0000000000..5ccd48bb8f--- /dev/null+++ b/internal/entity/models/llmman.go@@ -0,0 +1,40 @@+//+// 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++// LlmmanModel drives a local llmman server. It speaks the OpenAI wire format,+// so the embedded OpenAIAPICompatibleModel covers every call.+type LlmmanModel struct {+ *OpenAIAPICompatibleModel+}++// NewLlmmanModel creates an llmman driver.+func NewLlmmanModel(baseURL map[string]string, urlSuffix URLSuffix) *LlmmanModel {+ return &LlmmanModel{+ OpenAIAPICompatibleModel: NewOpenAIAPICompatibleModel(baseURL, urlSuffix),+ }+}++// Name returns the driver name.+func (m *LlmmanModel) Name() string {+ return "llmman"+}++// NewInstance returns a driver bound to the given base URL.+func (m *LlmmanModel) NewInstance(baseURL map[string]string) ModelDriver {+ return NewLlmmanModel(baseURL, m.baseModel.URLSuffix)+}diff --git a/internal/entity/models/llmman_test.go b/internal/entity/models/llmman_test.gonew file mode 100644index 0000000000..d390780cde--- /dev/null+++ b/internal/entity/models/llmman_test.go@@ -0,0 +1,116 @@+package models++import (+ "encoding/json"+ "net/http"+ "net/http/httptest"+ "ragflow/internal/common"+ "testing"+)++func newLlmmanForTest(baseURL string) *LlmmanModel {+ return NewLlmmanModel(+ map[string]string{"default": baseURL},+ URLSuffix{Chat: "chat/completions", Models: "models", Embedding: "embeddings"},+ )+}++func TestLlmmanName(t *testing.T) {+ if got := newLlmmanForTest("http://unused").Name(); got != "llmman" {+ t.Errorf("Name()=%q", got)+ }+}++func TestLlmmanFactory(t *testing.T) {+ driver, err := NewModelFactory().CreateModelDriver("llmman", map[string]string{"default": "http://unused"}, URLSuffix{})+ if err != nil {+ t.Fatalf("CreateModelDriver: %v", err)+ }+ if _, ok := driver.(*LlmmanModel); !ok {+ t.Fatalf("driver type=%T, want *LlmmanModel", driver)+ }+ if _, ok := driver.NewInstance(map[string]string{"default": "http://other"}).(*LlmmanModel); !ok {+ t.Fatal("NewInstance did not return *LlmmanModel")+ }+}++// A local server needs no credential, so calls must work with an empty APIConfig.+func TestLlmmanChatWithoutAPIKey(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 != "" {+ t.Errorf("Authorization=%q, want empty", got)+ }+ _ = json.NewEncoder(w).Encode(map[string]any{+ "id": "chat-llmman",+ "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 := newLlmmanForTest(srv.URL).ChatWithMessages(+ t.Context(), "qwen3", []Message{{Role: "user", Content: "ping"}}, &APIConfig{}, 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)+}++func TestLlmmanEmbed(t *testing.T) {+ withSSRFBypass(t)+ srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {+ if r.URL.Path != "/embeddings" {+ t.Errorf("path=%s", r.URL.Path)+ }+ _ = json.NewEncoder(w).Encode(map[string]any{+ "data": []map[string]any{{"embedding": []float64{0.1, 0.2}, "index": 0}},+ "usage": map[string]any{"prompt_tokens": 7, "total_tokens": 7},+ })+ }))+ defer srv.Close()++ modelName := "nomic-embed-text"+ embeddings, err := newLlmmanForTest(srv.URL).Embed(+ t.Context(), &modelName, EmbedRequest{Texts: []string{"document"}}, &APIConfig{}, nil, &common.ModelUsage{},+ )+ if err != nil {+ t.Fatalf("Embed: %v", err)+ }+ if len(embeddings) != 1 || len(embeddings[0].Embedding) != 2 {+ t.Fatalf("embeddings=%#v", embeddings)+ }+}++func TestLlmmanListModelsAndCheckConnection(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": "qwen3"}, {"id": "nomic-embed-text"}},+ })+ }))+ defer srv.Close()++ model := newLlmmanForTest(srv.URL)+ list, err := model.ListModels(t.Context(), &APIConfig{})+ if err != nil {+ t.Fatalf("ListModels: %v", err)+ }+ if joinModelNames(list, ",") != "qwen3,nomic-embed-text" {+ t.Errorf("models=%v", list)+ }+ if err := model.CheckConnection(t.Context(), &APIConfig{}); err != nil {+ t.Fatalf("CheckConnection: %v", err)+ }+}diff --git a/rag/llm/chat_model.py b/rag/llm/chat_model.pyindex b5ccde4cf8..2e9d9ef618 100644--- a/rag/llm/chat_model.py+++ b/rag/llm/chat_model.py@@ -1120,6 +1120,16 @@ class LmStudioChat(Base):self.model_name = model_name+class LlmmanChat(Base):+ _FACTORY_NAME = "llmman"++ def __init__(self, key, model_name, base_url, **kwargs):+ if not base_url:+ raise ValueError("Local llm url cannot be None")+ # Local server ignores auth; a placeholder key keeps Base's sync and async clients identical.+ super().__init__("llmman", model_name, base_url, **kwargs)++class OpenAI_APIChat(Base):_FACTORY_NAME = ["VLLM", "OpenAI-API-Compatible"]diff --git a/rag/llm/embedding_model.py b/rag/llm/embedding_model.pyindex f6e4eaa7d0..0012715e39 100644--- a/rag/llm/embedding_model.py+++ b/rag/llm/embedding_model.py@@ -905,6 +905,10 @@ class LmStudioEmbed(LocalAIEmbed):self.model_name = model_name+class LlmmanEmbed(LocalAIEmbed):+ _FACTORY_NAME = "llmman"++class OpenAI_APIEmbed(OpenAIEmbed):_FACTORY_NAME = ["VLLM", "OpenAI-API-Compatible"]diff --git a/rag/llm/model_meta.py b/rag/llm/model_meta.pyindex 5c26ab6050..f4eb955122 100644--- a/rag/llm/model_meta.py+++ b/rag/llm/model_meta.py@@ -1128,6 +1128,10 @@ class LMStudio(OpenAIAPICompatible):_FACTORY_NAME = "LM-Studio"+class Llmman(OpenAIAPICompatible):+ _FACTORY_NAME = "llmman"++class NewAPI(OpenAIAPICompatible):_FACTORY_NAME = "New API"