Skip to content

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

feat: add Hubris as a model provider (#19341)

Konstantin Romankov4731ee332297
diff --git a/conf/llm_factories.json b/conf/llm_factories.json
index e149c4a80f..5ea9bb36d3 100644
--- a/conf/llm_factories.json
+++ b/conf/llm_factories.json
@@ -5384,6 +5384,13 @@
"status": "1",
"llm": []
},
+ {
+ "name": "Hubris",
+ "logo": "",
+ "tags": "LLM,TEXT EMBEDDING",
+ "status": "1",
+ "llm": []
+ },
{
"name": "HuggingFace",
"logo": "",
diff --git a/conf/models/hubris.json b/conf/models/hubris.json
new file mode 100644
index 0000000000..05cff89509
--- /dev/null
+++ b/conf/models/hubris.json
@@ -0,0 +1,249 @@
+{
+ "name": "Hubris",
+ "url": {
+ "default": "https://api.hubris.pw/v1"
+ },
+ "url_suffix": {
+ "chat": "chat/completions",
+ "models": "models",
+ "embedding": "embeddings"
+ },
+ "class": "hubris",
+ "models": [
+ {
+ "name": "anthropic/claude-haiku-4.5",
+ "context_length": 200000,
+ "max_output": 64000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": false,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "anthropic/claude-opus-5",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "anthropic/claude-sonnet-5",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": false,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "deepseek/deepseek-v4-flash",
+ "context_length": 1048576,
+ "max_output": 384000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "deepseek/deepseek-v4-pro",
+ "context_length": 1048576,
+ "max_output": 384000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": false,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "google/gemini-3.7-flash",
+ "context_length": 1048576,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "google/gemini-3.8-flash",
+ "context_length": 1048576,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "minimax/minimax-m3",
+ "context_length": 1048576,
+ "max_output": 512000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": false,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "moonshotai/kimi-k3",
+ "context_length": 1048576,
+ "max_output": 943718,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "openai/gpt-5.4-mini",
+ "context_length": 400000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": false,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "openai/gpt-5.6-luna",
+ "context_length": 1050000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": false,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "openai/gpt-6-astra",
+ "context_length": 1050000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": false,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "x-ai/grok-4.6",
+ "context_length": 500000,
+ "max_output": 450000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "z-ai/glm-5.3",
+ "context_length": 1310720,
+ "max_output": 943718,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "openai/text-embedding-3-large",
+ "max_tokens": 8192,
+ "model_types": [
+ "embedding"
+ ]
+ },
+ {
+ "name": "openai/text-embedding-3-small",
+ "max_tokens": 8192,
+ "model_types": [
+ "embedding"
+ ]
+ }
+ ]
+}
diff --git a/docs/guides/models/supported_models.mdx b/docs/guides/models/supported_models.mdx
index 5608857452..fad33b8725 100644
--- a/docs/guides/models/supported_models.mdx
+++ b/docs/guides/models/supported_models.mdx
@@ -43,6 +43,7 @@ A complete list of model providers supported by RAGFlow, which will continue to
| GPUStack | `https://gpustack.ai` |
| GreenPT | `https://greenpt.ai` |
| Groq | `https://groq.com` |
+| Hubris | `https://hubris.pw` |
| HuggingFace | `https://huggingface.co` |
| Jiekou.AI | `https://jiekou.ai` |
| Jina | `https://jina.ai` |
diff --git a/internal/entity/models/factory.go b/internal/entity/models/factory.go
index 63101bcfd8..5bcd22df91 100644
--- a/internal/entity/models/factory.go
+++ b/internal/entity/models/factory.go
@@ -61,6 +61,8 @@ func (f *ModelFactory) CreateModelDriver(providerName string, baseURL map[string
return NewMWSModel(baseURL, urlSuffix), nil
case "xai":
return NewXAIModel(baseURL, urlSuffix), nil
+ case "hubris":
+ return NewHubrisModel(baseURL, urlSuffix), nil
case "lm-studio":
return NewLmStudioModel(baseURL, urlSuffix), nil
case "llmman":
diff --git a/internal/entity/models/hubris.go b/internal/entity/models/hubris.go
new file mode 100644
index 0000000000..8887fc25ca
--- /dev/null
+++ b/internal/entity/models/hubris.go
@@ -0,0 +1,41 @@
+//
+// 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
+
+// HubrisModel drives Hubris, a hosted gateway that fronts models from several
+// vendors behind one OpenAI-compatible endpoint, so the embedded
+// OpenAIAPICompatibleModel covers every call.
+type HubrisModel struct {
+ *OpenAIAPICompatibleModel
+}
+
+// NewHubrisModel creates a Hubris driver.
+func NewHubrisModel(baseURL map[string]string, urlSuffix URLSuffix) *HubrisModel {
+ return &HubrisModel{
+ OpenAIAPICompatibleModel: NewOpenAIAPICompatibleModel(baseURL, urlSuffix),
+ }
+}
+
+// Name returns the driver name.
+func (m *HubrisModel) Name() string {
+ return "hubris"
+}
+
+// NewInstance returns a driver bound to the given base URL.
+func (m *HubrisModel) NewInstance(baseURL map[string]string) ModelDriver {
+ return NewHubrisModel(baseURL, m.baseModel.URLSuffix)
+}
diff --git a/internal/entity/models/hubris_test.go b/internal/entity/models/hubris_test.go
new file mode 100644
index 0000000000..735a1f34e0
--- /dev/null
+++ b/internal/entity/models/hubris_test.go
@@ -0,0 +1,132 @@
+package models
+
+import (
+ "encoding/json"
+ "net/http"
+ "net/http/httptest"
+ "ragflow/internal/common"
+ "testing"
+)
+
+var testAPIKey = "sk-gw-test"
+
+func newHubrisForTest(baseURL string) *HubrisModel {
+ return NewHubrisModel(
+ map[string]string{"default": baseURL},
+ URLSuffix{Chat: "chat/completions", Models: "models", Embedding: "embeddings"},
+ )
+}
+
+func TestHubrisName(t *testing.T) {
+ if got := newHubrisForTest("http://unused").Name(); got != "hubris" {
+ t.Errorf("Name()=%q", got)
+ }
+}
+
+func TestHubrisFactory(t *testing.T) {
+ driver, err := NewModelFactory().CreateModelDriver("hubris", map[string]string{"default": "http://unused"}, URLSuffix{})
+ if err != nil {
+ t.Fatalf("CreateModelDriver: %v", err)
+ }
+ if _, ok := driver.(*HubrisModel); !ok {
+ t.Fatalf("driver type=%T, want *HubrisModel", driver)
+ }
+ if _, ok := driver.NewInstance(map[string]string{"default": "http://other"}).(*HubrisModel); !ok {
+ t.Fatal("NewInstance did not return *HubrisModel")
+ }
+}
+
+// Hubris is a hosted gateway, so unlike a local server it must receive the
+// tenant's key as a bearer token on every call.
+func TestHubrisChatSendsAPIKey(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 != "Bearer sk-gw-test" {
+ t.Errorf("Authorization=%q", got)
+ }
+ _ = json.NewEncoder(w).Encode(map[string]any{
+ "id": "chat-hubris",
+ "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 := newHubrisForTest(srv.URL).ChatWithMessages(
+ t.Context(),
+ "anthropic/claude-sonnet-5",
+ []Message{{Role: "user", Content: "ping"}},
+ &APIConfig{ApiKey: &testAPIKey},
+ 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 TestHubrisEmbed(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 := "openai/text-embedding-3-small"
+ embeddings, err := newHubrisForTest(srv.URL).Embed(
+ t.Context(),
+ &modelName,
+ EmbedRequest{Texts: []string{"document"}},
+ &APIConfig{ApiKey: &testAPIKey},
+ 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 TestHubrisListModelsAndCheckConnection(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": "anthropic/claude-sonnet-5"},
+ {"id": "openai/text-embedding-3-small"},
+ },
+ })
+ }))
+ defer srv.Close()
+
+ model := newHubrisForTest(srv.URL)
+ list, err := model.ListModels(t.Context(), &APIConfig{ApiKey: &testAPIKey})
+ if err != nil {
+ t.Fatalf("ListModels: %v", err)
+ }
+ if joinModelNames(list, ",") != "anthropic/claude-sonnet-5,openai/text-embedding-3-small" {
+ t.Errorf("models=%v", list)
+ }
+ if err := model.CheckConnection(t.Context(), &APIConfig{ApiKey: &testAPIKey}); err != nil {
+ t.Fatalf("CheckConnection: %v", err)
+ }
+}
diff --git a/rag/llm/chat_model.py b/rag/llm/chat_model.py
index 9018544611..33d308679d 100644
--- a/rag/llm/chat_model.py
+++ b/rag/llm/chat_model.py
@@ -1169,6 +1169,28 @@ class LlmmanChat(Base):
super().__init__("llmman", model_name, base_url, **kwargs)
+class HubrisChat(Base):
+ """Hubris OpenAI-compatible chat adapter.
+
+ The endpoint is fixed rather than configurable. Hubris is a hosted gateway
+ on one known host, so a tenant-supplied ``base_url`` would have no
+ legitimate use and would send the Hubris API key to whatever host was
+ configured.
+ """
+
+ _FACTORY_NAME = "Hubris"
+
+ _BASE_URL = "https://api.hubris.pw/v1"
+
+ def __init__(self, key, model_name, base_url=None, **kwargs):
+ """Build the client against the fixed Hubris endpoint.
+
+ ``base_url`` is accepted for signature compatibility with the other
+ chat adapters and deliberately ignored.
+ """
+ super().__init__(key, model_name, self._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 208eda97ce..db4e1d2c62 100644
--- a/rag/llm/embedding_model.py
+++ b/rag/llm/embedding_model.py
@@ -918,6 +918,22 @@ class LlmmanEmbed(LocalAIEmbed):
_FACTORY_NAME = "llmman"
+class HubrisEmbed(OpenAIEmbed):
+ """Hubris embeddings.
+
+ The endpoint is fixed rather than configurable, matching HubrisChat. Hubris
+ is a hosted gateway on one known host, so a tenant-supplied ``base_url``
+ would have no legitimate use and would send the tenant's key elsewhere.
+ """
+
+ _FACTORY_NAME = "Hubris"
+
+ _BASE_URL = "https://api.hubris.pw/v1"
+
+ def __init__(self, key, model_name, base_url=None):
+ super().__init__(key, model_name, self._BASE_URL)
+
+
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 aafa5dad9b..157de63a07 100644
--- a/rag/llm/model_meta.py
+++ b/rag/llm/model_meta.py
@@ -1132,6 +1132,23 @@ class Llmman(OpenAIAPICompatible):
_FACTORY_NAME = "llmman"
+class Hubris(OpenAIAPICompatible):
+ """Hubris model metadata.
+
+ ``_get_model_list_url`` is pinned for the same reason the chat and
+ embedding classes pin their endpoint: the catalogue must be read from the
+ gateway itself, never from a host supplied by the tenant.
+ """
+
+ _FACTORY_NAME = "Hubris"
+
+ _BASE_URL = "https://api.hubris.pw/v1"
+
+ def _get_model_list_url(self):
+ """Return the catalogue URL, ignoring any tenant-configured base URL."""
+ return f"{self._BASE_URL}/models"
+
+
class NewAPI(OpenAIAPICompatible):
_FACTORY_NAME = "New API"