From 432a9f45d778a670423e3b814a8861f1bb112204 Mon Sep 17 00:00:00 2001
From: majiayu000 <1835304752@qq.com>
Date: Tue, 30 Dec 2025 19:53:50 +0800
Subject: [PATCH 1/8] feat: add Ollama provider support

Add LMMEngineOllama class to support local Ollama models.
Ollama uses OpenAI-compatible API with default endpoint http://localhost:11434/v1.

Closes #44

Signed-off-by: majiayu000 <1835304752@qq.com>
---
 gui_agents/s3/core/engine.py | 46 ++++++++++++++++++++++++++++++++++++
 gui_agents/s3/core/mllm.py   |  4 ++++
 2 files changed, 50 insertions(+)

diff --git a/gui_agents/s3/core/engine.py b/gui_agents/s3/core/engine.py
index 7bf90f14..197e885d 100644
--- a/gui_agents/s3/core/engine.py
+++ b/gui_agents/s3/core/engine.py
@@ -443,3 +443,49 @@ def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
             .choices[0]
             .message.content
         )
+
+
+class LMMEngineOllama(LMMEngine):
+    def __init__(
+        self,
+        base_url=None,
+        api_key=None,
+        model=None,
+        rate_limit=-1,
+        temperature=None,
+        **kwargs,
+    ):
+        assert model is not None, "model must be provided"
+        self.model = model
+        self.api_key = api_key
+        self.base_url = base_url
+        self.request_interval = 0 if rate_limit == -1 else 60.0 / rate_limit
+        self.llm_client = None
+        self.temperature = temperature
+
+    @backoff.on_exception(
+        backoff.expo, (APIConnectionError, APIError, RateLimitError), max_time=60
+    )
+    def generate(
+        self,
+        messages,
+        temperature=0.0,
+        max_new_tokens=None,
+        **kwargs,
+    ):
+        base_url = self.base_url or os.getenv("OLLAMA_HOST")
+        if base_url is None:
+            base_url = "http://localhost:11434/v1"
+        elif not base_url.endswith("/v1"):
+            base_url = base_url.rstrip("/") + "/v1"
+        api_key = self.api_key or "ollama"
+        if not self.llm_client:
+            self.llm_client = OpenAI(base_url=base_url, api_key=api_key)
+        temp = self.temperature if self.temperature is not None else temperature
+        completion = self.llm_client.chat.completions.create(
+            model=self.model,
+            messages=messages,
+            max_tokens=max_new_tokens if max_new_tokens else 4096,
+            temperature=temp,
+        )
+        return completion.choices[0].message.content
diff --git a/gui_agents/s3/core/mllm.py b/gui_agents/s3/core/mllm.py
index fb49e4b0..2540a401 100644
--- a/gui_agents/s3/core/mllm.py
+++ b/gui_agents/s3/core/mllm.py
@@ -6,6 +6,7 @@
     LMMEngineAnthropic,
     LMMEngineAzureOpenAI,
     LMMEngineHuggingFace,
+    LMMEngineOllama,
     LMMEngineOpenAI,
     LMMEngineOpenRouter,
     LMMEngineParasail,
@@ -35,6 +36,8 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                     self.engine = LMMEngineOpenRouter(**engine_params)
                 elif engine_type == "parasail":
                     self.engine = LMMEngineParasail(**engine_params)
+                elif engine_type == "ollama":
+                    self.engine = LMMEngineOllama(**engine_params)
                 else:
                     raise ValueError(f"engine_type '{engine_type}' is not supported")
             else:
@@ -129,6 +132,7 @@ def add_message(
                 LMMEngineGemini,
                 LMMEngineOpenRouter,
                 LMMEngineParasail,
+                LMMEngineOllama,
             ),
         ):
             # infer role from previous message

From c55f3947df4c419d5a712862833016e0c3c7d0cf Mon Sep 17 00:00:00 2001
From: majiayu000 <1835304752@qq.com>
Date: Tue, 30 Dec 2025 23:32:11 +0800
Subject: [PATCH 2/8] fix: forward **kwargs to chat.completions.create() for
 Ollama engine
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This ensures consistency with other engine implementations (OpenAI, Gemini,
OpenRouter, etc.) and allows callers to pass additional parameters like
stop, top_p, etc.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---
 gui_agents/s3/core/engine.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/gui_agents/s3/core/engine.py b/gui_agents/s3/core/engine.py
index 197e885d..4476f567 100644
--- a/gui_agents/s3/core/engine.py
+++ b/gui_agents/s3/core/engine.py
@@ -487,5 +487,6 @@ def generate(
             messages=messages,
             max_tokens=max_new_tokens if max_new_tokens else 4096,
             temperature=temp,
+            **kwargs,
         )
         return completion.choices[0].message.content

From 02a9da5591c60ffcfd4d49555806784a8fed62d2 Mon Sep 17 00:00:00 2001
From: majiayu000 <1835304752@qq.com>
Date: Tue, 13 Jan 2026 22:55:16 +0800
Subject: [PATCH 3/8] refactor: reuse LMMEngineOpenAI for Ollama provider

---
 gui_agents/s3/core/engine.py | 45 ------------------------------------
 gui_agents/s3/core/mllm.py   | 11 ++++++---
 2 files changed, 8 insertions(+), 48 deletions(-)

diff --git a/gui_agents/s3/core/engine.py b/gui_agents/s3/core/engine.py
index 4476f567..21aa019e 100644
--- a/gui_agents/s3/core/engine.py
+++ b/gui_agents/s3/core/engine.py
@@ -445,48 +445,3 @@ def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
         )
 
 
-class LMMEngineOllama(LMMEngine):
-    def __init__(
-        self,
-        base_url=None,
-        api_key=None,
-        model=None,
-        rate_limit=-1,
-        temperature=None,
-        **kwargs,
-    ):
-        assert model is not None, "model must be provided"
-        self.model = model
-        self.api_key = api_key
-        self.base_url = base_url
-        self.request_interval = 0 if rate_limit == -1 else 60.0 / rate_limit
-        self.llm_client = None
-        self.temperature = temperature
-
-    @backoff.on_exception(
-        backoff.expo, (APIConnectionError, APIError, RateLimitError), max_time=60
-    )
-    def generate(
-        self,
-        messages,
-        temperature=0.0,
-        max_new_tokens=None,
-        **kwargs,
-    ):
-        base_url = self.base_url or os.getenv("OLLAMA_HOST")
-        if base_url is None:
-            base_url = "http://localhost:11434/v1"
-        elif not base_url.endswith("/v1"):
-            base_url = base_url.rstrip("/") + "/v1"
-        api_key = self.api_key or "ollama"
-        if not self.llm_client:
-            self.llm_client = OpenAI(base_url=base_url, api_key=api_key)
-        temp = self.temperature if self.temperature is not None else temperature
-        completion = self.llm_client.chat.completions.create(
-            model=self.model,
-            messages=messages,
-            max_tokens=max_new_tokens if max_new_tokens else 4096,
-            temperature=temp,
-            **kwargs,
-        )
-        return completion.choices[0].message.content
diff --git a/gui_agents/s3/core/mllm.py b/gui_agents/s3/core/mllm.py
index 2540a401..7c71d5f0 100644
--- a/gui_agents/s3/core/mllm.py
+++ b/gui_agents/s3/core/mllm.py
@@ -6,7 +6,7 @@
     LMMEngineAnthropic,
     LMMEngineAzureOpenAI,
     LMMEngineHuggingFace,
-    LMMEngineOllama,
+
     LMMEngineOpenAI,
     LMMEngineOpenRouter,
     LMMEngineParasail,
@@ -37,7 +37,12 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                 elif engine_type == "parasail":
                     self.engine = LMMEngineParasail(**engine_params)
                 elif engine_type == "ollama":
-                    self.engine = LMMEngineOllama(**engine_params)
+                    # Reuse LMMEngineOpenAI for Ollama, defaulting to localhost if not specified
+                    if "base_url" not in engine_params:
+                        engine_params["base_url"] = "http://localhost:11434/v1"
+                    if "api_key" not in engine_params:
+                        engine_params["api_key"] = "ollama"
+                    self.engine = LMMEngineOpenAI(**engine_params)
                 else:
                     raise ValueError(f"engine_type '{engine_type}' is not supported")
             else:
@@ -132,7 +137,7 @@ def add_message(
                 LMMEngineGemini,
                 LMMEngineOpenRouter,
                 LMMEngineParasail,
-                LMMEngineOllama,
+
             ),
         ):
             # infer role from previous message

From 234450bb0cbf00d7f1673cc52279f321f9e04db8 Mon Sep 17 00:00:00 2001
From: majiayu000 <1835304752@qq.com>
Date: Tue, 13 Jan 2026 23:16:29 +0800
Subject: [PATCH 4/8] fix: restore OLLAMA_HOST env var support

---
 gui_agents/s3/core/mllm.py | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/gui_agents/s3/core/mllm.py b/gui_agents/s3/core/mllm.py
index 7c71d5f0..d74ebadd 100644
--- a/gui_agents/s3/core/mllm.py
+++ b/gui_agents/s3/core/mllm.py
@@ -39,7 +39,14 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                 elif engine_type == "ollama":
                     # Reuse LMMEngineOpenAI for Ollama, defaulting to localhost if not specified
                     if "base_url" not in engine_params:
-                        engine_params["base_url"] = "http://localhost:11434/v1"
+                        import os
+                        base_url = os.getenv("OLLAMA_HOST")
+                        if base_url:
+                            if not base_url.endswith("/v1"):
+                                base_url = base_url.rstrip("/") + "/v1"
+                            engine_params["base_url"] = base_url
+                        else:
+                            engine_params["base_url"] = "http://localhost:11434/v1"
                     if "api_key" not in engine_params:
                         engine_params["api_key"] = "ollama"
                     self.engine = LMMEngineOpenAI(**engine_params)

From d2271de1704fb042cc42ab1748451e931f7d1056 Mon Sep 17 00:00:00 2001
From: lif <1835304752@qq.com>
Date: Thu, 15 Jan 2026 17:16:13 +0800
Subject: [PATCH 5/8] feat: add DeepSeek/Qwen support and improve Ollama config

---
 gui_agents/s3/core/engine.py | 93 ++++++++++++++++++++++++++++++++++++
 gui_agents/s3/core/mllm.py   | 16 +++++--
 tests/test_providers.py      | 68 ++++++++++++++++++++++++++
 3 files changed, 174 insertions(+), 3 deletions(-)
 create mode 100644 tests/test_providers.py

diff --git a/gui_agents/s3/core/engine.py b/gui_agents/s3/core/engine.py
index 21aa019e..50829eb7 100644
--- a/gui_agents/s3/core/engine.py
+++ b/gui_agents/s3/core/engine.py
@@ -445,3 +445,96 @@ def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
         )
 
 
+class LMMEngineDeepSeek(LMMEngine):
+    def __init__(
+        self, base_url=None, api_key=None, model=None, rate_limit=-1, **kwargs
+    ):
+        assert model is not None, "DeepSeek model id must be provided"
+        self.base_url = base_url
+        self.model = model
+        self.api_key = api_key
+        self.request_interval = 0 if rate_limit == -1 else 60.0 / rate_limit
+        self.llm_client = None
+
+    @backoff.on_exception(
+        backoff.expo, (APIConnectionError, APIError, RateLimitError), max_time=60
+    )
+    def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
+        api_key = self.api_key or os.getenv("DEEPSEEK_API_KEY")
+        if api_key is None:
+            raise ValueError(
+                "A DeepSeek API key needs to be provided in either the api_key parameter or as an environment variable named DEEPSEEK_API_KEY"
+            )
+        base_url = self.base_url or os.getenv("DEEPSEEK_ENDPOINT_URL")
+        if base_url is None:
+            base_url = "https://api.deepseek.com"
+        
+        if not self.llm_client:
+            self.llm_client = OpenAI(
+                base_url=base_url,
+                api_key=api_key,
+            )
+        return (
+            self.llm_client.chat.completions.create(
+                model=self.model,
+                messages=messages,
+                max_tokens=max_new_tokens if max_new_tokens else 4096,
+                temperature=temperature,
+                **kwargs,
+            )
+            .choices[0]
+            .message.content
+        )
+
+
+class LMMEngineQwen(LMMEngine):
+    def __init__(
+        self, base_url=None, api_key=None, model=None, rate_limit=-1, **kwargs
+    ):
+        assert model is not None, "Qwen model id must be provided"
+        self.base_url = base_url
+        self.model = model
+        self.api_key = api_key
+        self.request_interval = 0 if rate_limit == -1 else 60.0 / rate_limit
+        self.llm_client = None
+
+    @backoff.on_exception(
+        backoff.expo, (APIConnectionError, APIError, RateLimitError), max_time=60
+    )
+    def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
+        api_key = self.api_key or os.getenv("QWEN_API_KEY") # Or DASHSCOPE_API_KEY
+        if api_key is None:
+            raise ValueError(
+                "A Qwen API key needs to be provided in either the api_key parameter or as an environment variable named QWEN_API_KEY"
+            )
+        base_url = self.base_url or os.getenv("QWEN_ENDPOINT_URL")
+        if base_url is None:
+             # Alibaba Qwen often uses DashScope, but for compatible APIs let's assume standard or user provided
+             # If strictly Qwen (via DashScope compatible), valid URL is needed.
+             # defaulting to DashScope compatible endpoint as placeholder or rely on user.
+             # For this strict implementation, ensuring we have a URL is better if known,
+             # but generic "Qwen" usually implies usage via compatible interface (like vLLM serving Qwen or DashScope).
+             # Let's require it or default to a common one if reasonable.
+             # Given the other engines, let's enforce user providing it or env var if we don't have a single canonical one (DashScope is common).
+             # Let's default to DashScope's openai compatible endpoint if none provided?
+             # https://dashscope.aliyuncs.com/compatible-mode/v1
+             base_url = "https://dashscope.aliyuncs.com/compatible-mode/v1"
+
+        if not self.llm_client:
+            self.llm_client = OpenAI(
+                base_url=base_url,
+                api_key=api_key,
+            )
+        return (
+            self.llm_client.chat.completions.create(
+                model=self.model,
+                messages=messages,
+                max_tokens=max_new_tokens if max_new_tokens else 4096,
+                temperature=temperature,
+                **kwargs,
+            )
+            .choices[0]
+            .message.content
+        )
+
+
diff --git a/gui_agents/s3/core/mllm.py b/gui_agents/s3/core/mllm.py
index d74ebadd..90c1c222 100644
--- a/gui_agents/s3/core/mllm.py
+++ b/gui_agents/s3/core/mllm.py
@@ -12,6 +12,8 @@
     LMMEngineParasail,
     LMMEnginevLLM,
     LMMEngineGemini,
+    LMMEngineDeepSeek,
+    LMMEngineQwen,
 )
 
 
@@ -37,7 +39,7 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                 elif engine_type == "parasail":
                     self.engine = LMMEngineParasail(**engine_params)
                 elif engine_type == "ollama":
-                    # Reuse LMMEngineOpenAI for Ollama, defaulting to localhost if not specified
+                    # Reuse LMMEngineOpenAI for Ollama
                     if "base_url" not in engine_params:
                         import os
                         base_url = os.getenv("OLLAMA_HOST")
@@ -46,10 +48,17 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                                 base_url = base_url.rstrip("/") + "/v1"
                             engine_params["base_url"] = base_url
                         else:
-                            engine_params["base_url"] = "http://localhost:11434/v1"
+                            # RAISE ERROR instead of default
+                            raise ValueError(
+                                "Ollama endpoint must be provided via 'base_url' parameter or 'OLLAMA_HOST' environment variable."
+                            )
                     if "api_key" not in engine_params:
                         engine_params["api_key"] = "ollama"
                     self.engine = LMMEngineOpenAI(**engine_params)
+                elif engine_type == "deepseek":
+                    self.engine = LMMEngineDeepSeek(**engine_params)
+                elif engine_type == "qwen":
+                    self.engine = LMMEngineQwen(**engine_params)
                 else:
                     raise ValueError(f"engine_type '{engine_type}' is not supported")
             else:
@@ -144,7 +153,8 @@ def add_message(
                 LMMEngineGemini,
                 LMMEngineOpenRouter,
                 LMMEngineParasail,
-
+                LMMEngineDeepSeek,
+                LMMEngineQwen,
             ),
         ):
             # infer role from previous message
diff --git a/tests/test_providers.py b/tests/test_providers.py
new file mode 100644
index 00000000..95432663
--- /dev/null
+++ b/tests/test_providers.py
@@ -0,0 +1,68 @@
+
+import os
+import unittest
+from unittest.mock import patch, MagicMock
+from gui_agents.s3.core.mllm import LMMAgent
+from gui_agents.s3.core.engine import LMMEngineOpenAI, LMMEngineDeepSeek, LMMEngineQwen
+
+class TestProviders(unittest.TestCase):
+    def setUp(self):
+        # Clear env vars before each test
+        if "OLLAMA_HOST" in os.environ:
+            del os.environ["OLLAMA_HOST"]
+        if "DEEPSEEK_API_KEY" in os.environ:
+            del os.environ["DEEPSEEK_API_KEY"]
+        if "QWEN_API_KEY" in os.environ:
+            del os.environ["QWEN_API_KEY"]
+
+    def test_ollama_missing_config(self):
+        """Test that Ollama raises ValueError if no endpoint is provided"""
+        with self.assertRaises(ValueError) as cm:
+            LMMAgent(engine_params={"engine_type": "ollama", "model": "llama3"})
+        self.assertIn("Ollama endpoint must be provided", str(cm.exception))
+
+    def test_ollama_valid_config_param(self):
+        """Test Ollama init with base_url param"""
+        agent = LMMAgent(engine_params={
+            "engine_type": "ollama", 
+            "model": "llama3",
+            "base_url": "http://example.com/v1"
+        })
+        self.assertIsInstance(agent.engine, LMMEngineOpenAI)
+        self.assertEqual(agent.engine.base_url, "http://example.com/v1")
+
+    def test_ollama_valid_config_env(self):
+        """Test Ollama init with OLLAMA_HOST env var"""
+        with patch.dict(os.environ, {"OLLAMA_HOST": "http://env-host:11434"}):
+            agent = LMMAgent(engine_params={
+                "engine_type": "ollama", 
+                "model": "llama3"
+            })
+            self.assertIsInstance(agent.engine, LMMEngineOpenAI)
+            # Check for /v1 addition
+            self.assertEqual(agent.engine.base_url, "http://env-host:11434/v1")
+
+    def test_deepseek_init(self):
+        """Test DeepSeek initialization"""
+        with patch.dict(os.environ, {"DEEPSEEK_API_KEY": "sk-test"}):
+            agent = LMMAgent(engine_params={
+                "engine_type": "deepseek",
+                "model": "deepseek-coder"
+            })
+            self.assertIsInstance(agent.engine, LMMEngineDeepSeek)
+            # Default URL
+            self.assertEqual(agent.engine.base_url, None) 
+            # (Note: engine.py logic resolves default at generate() time or if client created, 
+            # but init just stores what's passed. Let's verify prompt generation to ensure it doesn't crash on init)
+    
+    def test_qwen_init(self):
+        """Test Qwen initialization"""
+        with patch.dict(os.environ, {"QWEN_API_KEY": "sk-qwen"}):
+            agent = LMMAgent(engine_params={
+                "engine_type": "qwen",
+                "model": "qwen-max"
+            })
+            self.assertIsInstance(agent.engine, LMMEngineQwen)
+
+if __name__ == "__main__":
+    unittest.main()

From c8f306efd7911def2f0231f5fb6fa59e1d62fb93 Mon Sep 17 00:00:00 2001
From: lif <1835304752@qq.com>
Date: Mon, 19 Jan 2026 18:06:24 +0800
Subject: [PATCH 6/8] refactor: reuse LMMEngineOpenAI for DeepSeek and Qwen
 providers

---
 gui_agents/s3/core/engine.py | 95 ------------------------------------
 gui_agents/s3/core/mllm.py   | 46 ++++++++++++++---
 tests/test_providers.py      | 48 +++++++++---------
 3 files changed, 63 insertions(+), 126 deletions(-)

diff --git a/gui_agents/s3/core/engine.py b/gui_agents/s3/core/engine.py
index 50829eb7..7bf90f14 100644
--- a/gui_agents/s3/core/engine.py
+++ b/gui_agents/s3/core/engine.py
@@ -443,98 +443,3 @@ def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
             .choices[0]
             .message.content
         )
-
-
-class LMMEngineDeepSeek(LMMEngine):
-    def __init__(
-        self, base_url=None, api_key=None, model=None, rate_limit=-1, **kwargs
-    ):
-        assert model is not None, "DeepSeek model id must be provided"
-        self.base_url = base_url
-        self.model = model
-        self.api_key = api_key
-        self.request_interval = 0 if rate_limit == -1 else 60.0 / rate_limit
-        self.llm_client = None
-
-    @backoff.on_exception(
-        backoff.expo, (APIConnectionError, APIError, RateLimitError), max_time=60
-    )
-    def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
-        api_key = self.api_key or os.getenv("DEEPSEEK_API_KEY")
-        if api_key is None:
-            raise ValueError(
-                "A DeepSeek API key needs to be provided in either the api_key parameter or as an environment variable named DEEPSEEK_API_KEY"
-            )
-        base_url = self.base_url or os.getenv("DEEPSEEK_ENDPOINT_URL")
-        if base_url is None:
-            base_url = "https://api.deepseek.com"
-        
-        if not self.llm_client:
-            self.llm_client = OpenAI(
-                base_url=base_url,
-                api_key=api_key,
-            )
-        return (
-            self.llm_client.chat.completions.create(
-                model=self.model,
-                messages=messages,
-                max_tokens=max_new_tokens if max_new_tokens else 4096,
-                temperature=temperature,
-                **kwargs,
-            )
-            .choices[0]
-            .message.content
-        )
-
-
-class LMMEngineQwen(LMMEngine):
-    def __init__(
-        self, base_url=None, api_key=None, model=None, rate_limit=-1, **kwargs
-    ):
-        assert model is not None, "Qwen model id must be provided"
-        self.base_url = base_url
-        self.model = model
-        self.api_key = api_key
-        self.request_interval = 0 if rate_limit == -1 else 60.0 / rate_limit
-        self.llm_client = None
-
-    @backoff.on_exception(
-        backoff.expo, (APIConnectionError, APIError, RateLimitError), max_time=60
-    )
-    def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
-        api_key = self.api_key or os.getenv("QWEN_API_KEY") # Or DASHSCOPE_API_KEY
-        if api_key is None:
-            raise ValueError(
-                "A Qwen API key needs to be provided in either the api_key parameter or as an environment variable named QWEN_API_KEY"
-            )
-        base_url = self.base_url or os.getenv("QWEN_ENDPOINT_URL")
-        if base_url is None:
-             # Alibaba Qwen often uses DashScope, but for compatible APIs let's assume standard or user provided
-             # If strictly Qwen (via DashScope compatible), valid URL is needed.
-             # defaulting to DashScope compatible endpoint as placeholder or rely on user.
-             # For this strict implementation, ensuring we have a URL is better if known,
-             # but generic "Qwen" usually implies usage via compatible interface (like vLLM serving Qwen or DashScope).
-             # Let's require it or default to a common one if reasonable.
-             # Given the other engines, let's enforce user providing it or env var if we don't have a single canonical one (DashScope is common).
-             # Let's default to DashScope's openai compatible endpoint if none provided?
-             # https://dashscope.aliyuncs.com/compatible-mode/v1
-             base_url = "https://dashscope.aliyuncs.com/compatible-mode/v1"
-
-        if not self.llm_client:
-            self.llm_client = OpenAI(
-                base_url=base_url,
-                api_key=api_key,
-            )
-        return (
-            self.llm_client.chat.completions.create(
-                model=self.model,
-                messages=messages,
-                max_tokens=max_new_tokens if max_new_tokens else 4096,
-                temperature=temperature,
-                **kwargs,
-            )
-            .choices[0]
-            .message.content
-        )
-
-
diff --git a/gui_agents/s3/core/mllm.py b/gui_agents/s3/core/mllm.py
index 90c1c222..c908da4f 100644
--- a/gui_agents/s3/core/mllm.py
+++ b/gui_agents/s3/core/mllm.py
@@ -6,14 +6,11 @@
     LMMEngineAnthropic,
     LMMEngineAzureOpenAI,
     LMMEngineHuggingFace,
-
     LMMEngineOpenAI,
     LMMEngineOpenRouter,
     LMMEngineParasail,
     LMMEnginevLLM,
     LMMEngineGemini,
-    LMMEngineDeepSeek,
-    LMMEngineQwen,
 )
 
 
@@ -42,6 +39,7 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                     # Reuse LMMEngineOpenAI for Ollama
                     if "base_url" not in engine_params:
                         import os
+
                         base_url = os.getenv("OLLAMA_HOST")
                         if base_url:
                             if not base_url.endswith("/v1"):
@@ -56,9 +54,44 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                         engine_params["api_key"] = "ollama"
                     self.engine = LMMEngineOpenAI(**engine_params)
                 elif engine_type == "deepseek":
-                    self.engine = LMMEngineDeepSeek(**engine_params)
+                    if "base_url" not in engine_params:
+                        import os
+
+                        base_url = os.getenv("DEEPSEEK_ENDPOINT_URL")
+                        if not base_url:
+                            base_url = "https://api.deepseek.com"
+                        engine_params["base_url"] = base_url
+                    if "api_key" not in engine_params:
+                        import os
+
+                        api_key = os.getenv("DEEPSEEK_API_KEY")
+                        if not api_key:
+                            raise ValueError(
+                                "DeepSeek API key must be provided via 'api_key' parameter or 'DEEPSEEK_API_KEY' environment variable."
+                            )
+                        engine_params["api_key"] = api_key
+
+                    self.engine = LMMEngineOpenAI(**engine_params)
                 elif engine_type == "qwen":
-                    self.engine = LMMEngineQwen(**engine_params)
+                    if "base_url" not in engine_params:
+                        import os
+
+                        base_url = os.getenv("QWEN_ENDPOINT_URL")
+                        if not base_url:
+                            base_url = (
+                                "https://dashscope.aliyuncs.com/compatible-mode/v1"
+                            )
+                        engine_params["base_url"] = base_url
+                    if "api_key" not in engine_params:
+                        import os
+
+                        api_key = os.getenv("QWEN_API_KEY")
+                        if not api_key:
+                            raise ValueError(
+                                "Qwen API key must be provided via 'api_key' parameter or 'QWEN_API_KEY' environment variable."
+                            )
+                        engine_params["api_key"] = api_key
+                    self.engine = LMMEngineOpenAI(**engine_params)
                 else:
                     raise ValueError(f"engine_type '{engine_type}' is not supported")
             else:
@@ -153,8 +186,7 @@ def add_message(
                 LMMEngineGemini,
                 LMMEngineOpenRouter,
                 LMMEngineParasail,
-                LMMEngineDeepSeek,
-                LMMEngineQwen,
+                LMMEngineParasail,
             ),
         ):
             # infer role from previous message
diff --git a/tests/test_providers.py b/tests/test_providers.py
index 95432663..ff52a2d4 100644
--- a/tests/test_providers.py
+++ b/tests/test_providers.py
@@ -1,9 +1,9 @@
-
 import os
 import unittest
 from unittest.mock import patch, MagicMock
 from gui_agents.s3.core.mllm import LMMAgent
-from gui_agents.s3.core.engine import LMMEngineOpenAI, LMMEngineDeepSeek, LMMEngineQwen
+from gui_agents.s3.core.engine import LMMEngineOpenAI
+
 
 class TestProviders(unittest.TestCase):
     def setUp(self):
@@ -23,21 +23,20 @@ def test_ollama_missing_config(self):
 
     def test_ollama_valid_config_param(self):
         """Test Ollama init with base_url param"""
-        agent = LMMAgent(engine_params={
-            "engine_type": "ollama", 
-            "model": "llama3",
-            "base_url": "http://example.com/v1"
-        })
+        agent = LMMAgent(
+            engine_params={
+                "engine_type": "ollama",
+                "model": "llama3",
+                "base_url": "http://example.com/v1",
+            }
+        )
         self.assertIsInstance(agent.engine, LMMEngineOpenAI)
         self.assertEqual(agent.engine.base_url, "http://example.com/v1")
 
     def test_ollama_valid_config_env(self):
         """Test Ollama init with OLLAMA_HOST env var"""
         with patch.dict(os.environ, {"OLLAMA_HOST": "http://env-host:11434"}):
-            agent = LMMAgent(engine_params={
-                "engine_type": "ollama", 
-                "model": "llama3"
-            })
+            agent = LMMAgent(engine_params={"engine_type": "ollama", "model": "llama3"})
             self.assertIsInstance(agent.engine, LMMEngineOpenAI)
             # Check for /v1 addition
             self.assertEqual(agent.engine.base_url, "http://env-host:11434/v1")
@@ -45,24 +44,25 @@ def test_ollama_valid_config_env(self):
     def test_deepseek_init(self):
         """Test DeepSeek initialization"""
         with patch.dict(os.environ, {"DEEPSEEK_API_KEY": "sk-test"}):
-            agent = LMMAgent(engine_params={
-                "engine_type": "deepseek",
-                "model": "deepseek-coder"
-            })
-            self.assertIsInstance(agent.engine, LMMEngineDeepSeek)
+            agent = LMMAgent(
+                engine_params={"engine_type": "deepseek", "model": "deepseek-coder"}
+            )
+            self.assertIsInstance(agent.engine, LMMEngineOpenAI)
             # Default URL
-            self.assertEqual(agent.engine.base_url, None) 
-            # (Note: engine.py logic resolves default at generate() time or if client created, 
+            self.assertEqual(agent.engine.base_url, "https://api.deepseek.com")
+            # (Note: engine.py logic resolves default at generate() time or if client created,
             # but init just stores what's passed. Let's verify prompt generation to ensure it doesn't crash on init)
-    
+
     def test_qwen_init(self):
         """Test Qwen initialization"""
         with patch.dict(os.environ, {"QWEN_API_KEY": "sk-qwen"}):
-            agent = LMMAgent(engine_params={
-                "engine_type": "qwen",
-                "model": "qwen-max"
-            })
-            self.assertIsInstance(agent.engine, LMMEngineQwen)
+            agent = LMMAgent(engine_params={"engine_type": "qwen", "model": "qwen-max"})
+            self.assertIsInstance(agent.engine, LMMEngineOpenAI)
+            self.assertEqual(
+                agent.engine.base_url,
+                "https://dashscope.aliyuncs.com/compatible-mode/v1",
+            )
+
 
 if __name__ == "__main__":
     unittest.main()

From b6601890d6f4b0eba7b7e1d2837e826d2862948a Mon Sep 17 00:00:00 2001
From: lif <1835304752@qq.com>
Date: Mon, 19 Jan 2026 18:16:07 +0800
Subject: [PATCH 7/8] fix: address review comments (robust params, url
 normalization, cleanup)

---
 gui_agents/s3/core/mllm.py | 13 +++++++++----
 tests/test_providers.py    |  2 +-
 2 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/gui_agents/s3/core/mllm.py b/gui_agents/s3/core/mllm.py
index c908da4f..efea62aa 100644
--- a/gui_agents/s3/core/mllm.py
+++ b/gui_agents/s3/core/mllm.py
@@ -60,8 +60,11 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                         base_url = os.getenv("DEEPSEEK_ENDPOINT_URL")
                         if not base_url:
                             base_url = "https://api.deepseek.com"
+                        if not base_url.endswith("/v1"):
+                            base_url = base_url.rstrip("/") + "/v1"
                         engine_params["base_url"] = base_url
-                    if "api_key" not in engine_params:
+
+                    if not engine_params.get("api_key"):
                         import os
 
                         api_key = os.getenv("DEEPSEEK_API_KEY")
@@ -73,7 +76,7 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
 
                     self.engine = LMMEngineOpenAI(**engine_params)
                 elif engine_type == "qwen":
-                    if "base_url" not in engine_params:
+                    if not engine_params.get("base_url"):
                         import os
 
                         base_url = os.getenv("QWEN_ENDPOINT_URL")
@@ -81,8 +84,11 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                             base_url = (
                                 "https://dashscope.aliyuncs.com/compatible-mode/v1"
                             )
+                        if not base_url.endswith("/v1"):
+                            base_url = base_url.rstrip("/") + "/v1"
                         engine_params["base_url"] = base_url
-                    if "api_key" not in engine_params:
+
+                    if not engine_params.get("api_key"):
                         import os
 
                         api_key = os.getenv("QWEN_API_KEY")
@@ -186,7 +192,6 @@ def add_message(
                 LMMEngineGemini,
                 LMMEngineOpenRouter,
                 LMMEngineParasail,
-                LMMEngineParasail,
             ),
         ):
             # infer role from previous message
diff --git a/tests/test_providers.py b/tests/test_providers.py
index ff52a2d4..5eb8399b 100644
--- a/tests/test_providers.py
+++ b/tests/test_providers.py
@@ -49,7 +49,7 @@ def test_deepseek_init(self):
             )
             self.assertIsInstance(agent.engine, LMMEngineOpenAI)
             # Default URL
-            self.assertEqual(agent.engine.base_url, "https://api.deepseek.com")
+            self.assertEqual(agent.engine.base_url, "https://api.deepseek.com/v1")
             # (Note: engine.py logic resolves default at generate() time or if client created,
             # but init just stores what's passed. Let's verify prompt generation to ensure it doesn't crash on init)
 

From 000dd3700646a78db7069a074ebbcc19739b6ab6 Mon Sep 17 00:00:00 2001
From: lif <1835304752@qq.com>
Date: Mon, 19 Jan 2026 19:26:51 +0800
Subject: [PATCH 8/8] fix: clean up test env and normalize ollama param checks

---
 gui_agents/s3/core/mllm.py | 4 ++--
 tests/test_providers.py    | 4 ++++
 2 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/gui_agents/s3/core/mllm.py b/gui_agents/s3/core/mllm.py
index efea62aa..58beb93c 100644
--- a/gui_agents/s3/core/mllm.py
+++ b/gui_agents/s3/core/mllm.py
@@ -37,7 +37,7 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                     self.engine = LMMEngineParasail(**engine_params)
                 elif engine_type == "ollama":
                     # Reuse LMMEngineOpenAI for Ollama
-                    if "base_url" not in engine_params:
+                    if not engine_params.get("base_url"):
                         import os
 
                         base_url = os.getenv("OLLAMA_HOST")
@@ -50,7 +50,7 @@ def __init__(self, engine_params=None, system_prompt=None, engine=None):
                             raise ValueError(
                                 "Ollama endpoint must be provided via 'base_url' parameter or 'OLLAMA_HOST' environment variable."
                             )
-                    if "api_key" not in engine_params:
+                    if not engine_params.get("api_key"):
                         engine_params["api_key"] = "ollama"
                     self.engine = LMMEngineOpenAI(**engine_params)
                 elif engine_type == "deepseek":
diff --git a/tests/test_providers.py b/tests/test_providers.py
index 5eb8399b..2e1b24d2 100644
--- a/tests/test_providers.py
+++ b/tests/test_providers.py
@@ -14,6 +14,10 @@ def setUp(self):
             del os.environ["DEEPSEEK_API_KEY"]
         if "QWEN_API_KEY" in os.environ:
             del os.environ["QWEN_API_KEY"]
+        if "DEEPSEEK_ENDPOINT_URL" in os.environ:
+            del os.environ["DEEPSEEK_ENDPOINT_URL"]
+        if "QWEN_ENDPOINT_URL" in os.environ:
+            del os.environ["QWEN_ENDPOINT_URL"]
 
     def test_ollama_missing_config(self):
         """Test that Ollama raises ValueError if no endpoint is provided"""