Skip to content

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

feat: add llmman as a model provider (#19030)

Eric Curtinc051f40593fb
diff --git a/conf/llm_factories.json b/conf/llm_factories.json
index 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.json
new file mode 100644
index 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.mdx
index 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.go
index 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[string
return NewXAIModel(baseURL, urlSuffix), nil
case "lm-studio":
return NewLmStudioModel(baseURL, urlSuffix), nil
+ case "llmman":
+ return NewLlmmanModel(baseURL, urlSuffix), nil
case "ollama":
return NewOllamaModel(baseURL, urlSuffix), nil
case "openai":
diff --git a/internal/entity/models/llmman.go b/internal/entity/models/llmman.go
new file mode 100644
index 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.go
new file mode 100644
index 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.py
index 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.py
index 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.py
index 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"