agenthub-python 0.4.2__tar.gz → 0.4.4__tar.gz

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agenthub-python
-Version: 0.4.2
+Version: 0.4.4
 Summary: AgentHub is the LLM API Hub for the Agent era, built for high-precision autonomous agents.
 Keywords: agent,llm,gemini,claude,gpt
 Author: PrismShadow
@@ -13,6 +13,7 @@ Requires-Dist: httpx>=0.27.0
 Requires-Dist: httpx[socks] ; extra == 'dev'
 Requires-Dist: pytest>=8.4.2 ; extra == 'dev'
 Requires-Dist: pytest-asyncio>=0.23.0 ; extra == 'dev'
+Requires-Dist: pytest-xdist>=3.6.0 ; extra == 'dev'
 Requires-Dist: ruff>=0.14.3 ; extra == 'dev'
 Requires-Dist: pillow>=10.0.0 ; extra == 'dev'
 Requires-Python: >=3.11

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/ant_messages/client.py

@@ -75,6 +75,7 @@ class AntMessagesClient(LLMClient):
             ThinkingLevel.MEDIUM: {"thinking": {"type": "adaptive"}, "output_config": {"effort": "medium"}},
             ThinkingLevel.HIGH: {"thinking": {"type": "adaptive"}, "output_config": {"effort": "high"}},
             ThinkingLevel.XHIGH: {"thinking": {"type": "adaptive"}, "output_config": {"effort": "xhigh"}},
+            ThinkingLevel.MAX: {"thinking": {"type": "adaptive"}, "output_config": {"effort": "max"}},
         }
         return mapping.get(thinking_level)
 
@@ -300,7 +301,9 @@ class AntMessagesClient(LLMClient):
         elif ant_event_type == "message_stop":
             event_type = "stop"
 
-        elif ant_event_type in ["text", "thinking", "signature", "input_json"]:
+        elif ant_event_type in ["text", "thinking", "signature", "input_json", "ping"]:
+            # the SDK drops the "ping" heartbeat at the SSE layer; it reaches here only from
+            # gateways that relabel it onto another event
             event_type = "unused"
 
         else:

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/auto_client.py

@@ -47,10 +47,6 @@ class AutoLLMClient(LLMClient):
     ) -> LLMClient:
         """Create the appropriate client for the given model."""
         client_type = (client_type or os.getenv("CLIENT_TYPE") or model).lower()
-        if client_type == "minimax-m3":
-            from .minimax_m3 import MiniMaxM3Client
-
-            return MiniMaxM3Client(model=model, api_key=api_key, base_url=base_url)
         # every Gemini generation shares the unified client ("gemini-3" also matches the
         # gemini-3.7/gemini-3.6/gemini-3.5-flash-lite client types)
         if any(
@@ -78,6 +74,10 @@ class AutoLLMClient(LLMClient):
             from .kimi_k3 import KimiK3Client
 
             return KimiK3Client(model=model, api_key=api_key, base_url=base_url)
+        elif client_type == "minimax-m3":
+            from .minimax_m3 import MiniMaxM3Client
+
+            return MiniMaxM3Client(model=model, api_key=api_key, base_url=base_url)
         elif "deepseek-v4" in client_type:
             from .deepseek_v4 import DeepSeekV4Client
 

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/claude5/client.py

@@ -115,6 +115,8 @@ class Claude5Client(LLMClient):
                 "thinking": {"type": "adaptive"},
                 "output_config": {"effort": "high" if "4-6" in self._model else "xhigh"},
             },
+            # every model this client serves is 4.6 or later, and max spans that whole range
+            ThinkingLevel.MAX: {"thinking": {"type": "adaptive"}, "output_config": {"effort": "max"}},
         }
         return mapping.get(thinking_level)
 
@@ -349,7 +351,9 @@ class Claude5Client(LLMClient):
         elif claude_event_type == "message_stop":
             event_type = "stop"
 
-        elif claude_event_type in ["text", "thinking", "signature", "input_json"]:
+        elif claude_event_type in ["text", "thinking", "signature", "input_json", "ping"]:
+            # the SDK drops the "ping" heartbeat at the SSE layer; it reaches here only from
+            # gateways that relabel it onto another event
             event_type = "unused"
 
         else:

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/deepseek_v4/client.py

@@ -54,17 +54,24 @@ class DeepSeekV4Client(LLMClient):
             ThinkingLevel.MEDIUM: {"type": "enabled"},
             ThinkingLevel.HIGH: {"type": "enabled"},
             ThinkingLevel.XHIGH: {"type": "enabled"},
+            ThinkingLevel.MAX: {"type": "enabled"},
         }
         return mapping[thinking_level]
 
     def _convert_reasoning_effort(self, thinking_level: ThinkingLevel) -> str | None:
-        """Convert ThinkingLevel enum to DeepSeek's reasoning_effort."""
+        """Convert ThinkingLevel enum to DeepSeek's reasoning_effort.
+
+        DeepSeek accepts low/high/max and maps medium and xhigh onto high server-side
+        (llmsdk_docs/deepseek_v4/docs/thinking-mode.md), so this sends the value the
+        server would settle on anyway.
+        """
         mapping = {
             ThinkingLevel.NONE: None,
-            ThinkingLevel.LOW: "high",
+            ThinkingLevel.LOW: "low",
             ThinkingLevel.MEDIUM: "high",
             ThinkingLevel.HIGH: "high",
-            ThinkingLevel.XHIGH: "max",
+            ThinkingLevel.XHIGH: "high",
+            ThinkingLevel.MAX: "max",
         }
         return mapping[thinking_level]
 
@@ -206,7 +213,9 @@ class DeepSeekV4Client(LLMClient):
         usage_metadata: UsageMetadata | None = None
         finish_reason: FinishReason | None = None
 
-        if len(model_output.choices) > 0:
+        # gateways inject content-free heartbeat chunks on long generations, whose choices
+        # the SDK leaves as None rather than an empty list
+        if model_output.choices:
             choice = model_output.choices[0]
             delta = choice.delta
 

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/gemini3_7/client.py

@@ -142,6 +142,8 @@ class Gemini3_7Client(LLMClient):
             ThinkingLevel.MEDIUM: types.ThinkingLevel.MEDIUM,
             ThinkingLevel.HIGH: types.ThinkingLevel.HIGH,
             ThinkingLevel.XHIGH: types.ThinkingLevel.HIGH,
+            # Gemini stops at "high", so both top levels land there before per-model clamping
+            ThinkingLevel.MAX: types.ThinkingLevel.HIGH,
         }
         level = mapping.get(thinking_level)
         if level is None:
@@ -201,6 +203,8 @@ class Gemini3_7Client(LLMClient):
                 "sampling parameters starting with the 3.6 generation.",
             )
 
+        # include_thoughts asks for thought summaries, but whether generateContent returns any
+        # is model-dependent (llmsdk_docs/gemini3_7/docs/thinking.md)
         thinking_summary = config.get("thinking_summary")
         thinking_level = config.get("thinking_level")
         if thinking_summary is not None or thinking_level is not None:
@@ -436,6 +440,11 @@ class Gemini3_7Client(LLMClient):
                 "response_tokens": model_output.usage_metadata.candidates_token_count,
             }
 
+        if not content_items and usage_metadata is None and finish_reason is None:
+            # nothing was read out of the chunk, so there is nothing to emit: a gateway
+            # heartbeat looks like this, and so does any other chunk we take no value from
+            event_type = "unused"
+
         return {
             "role": "assistant",
             "event_type": event_type,
@@ -511,6 +520,9 @@ class Gemini3_7Client(LLMClient):
         )
         async for chunk in response_stream:
             event = self.transform_model_output_to_uni_event(chunk)
+            if event["event_type"] == "unused":
+                continue
+
             for item in event["content_items"]:
                 if item["type"] == "tool_call":
                     # gemini 3.7 does not support partial tool call, mock a partial tool call event

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/glm5_3/client.py

@@ -81,6 +81,7 @@ class GLM5_3Client(LLMClient):
                 ThinkingLevel.MEDIUM: "high",
                 ThinkingLevel.HIGH: "high",
                 ThinkingLevel.XHIGH: "max",
+                ThinkingLevel.MAX: "max",
             }
             return mapping.get(thinking_level)
         if "glm-5.2" in model:
@@ -90,6 +91,7 @@ class GLM5_3Client(LLMClient):
                 ThinkingLevel.MEDIUM: "medium",
                 ThinkingLevel.HIGH: "high",
                 ThinkingLevel.XHIGH: "xhigh",
+                ThinkingLevel.MAX: "max",
             }
             return mapping.get(thinking_level)
         return None
@@ -239,7 +241,9 @@ class GLM5_3Client(LLMClient):
         usage_metadata: UsageMetadata | None = None
         finish_reason: FinishReason | None = None
 
-        if len(model_output.choices) > 0:
+        # gateways inject content-free heartbeat chunks on long generations, whose choices
+        # the SDK leaves as None rather than an empty list
+        if model_output.choices:
             choice = model_output.choices[0]
             delta = choice.delta
 

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/gpt5_6/client.py

@@ -54,6 +54,7 @@ class GPT5_6Client(LLMClient):
             ThinkingLevel.MEDIUM: "medium",
             ThinkingLevel.HIGH: "high",
             ThinkingLevel.XHIGH: "xhigh",
+            ThinkingLevel.MAX: "max",
         }
         return mapping.get(thinking_level)
 
@@ -301,6 +302,7 @@ class GPT5_6Client(LLMClient):
             "response.reasoning_text.done",
             "response.content_part.added",
             "response.content_part.done",
+            "keepalive",  # gateway heartbeat on long generations; carries no content
         ]:
             event_type = "unused"
 

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/integration/playground.py

@@ -286,7 +286,8 @@ def create_chat_app() -> Flask:
                             <button type="button" role="option" aria-selected="false" class="w-full px-3 py-2 text-left hover:bg-gray-50 focus:bg-gray-50 focus:outline-none" data-combobox-option data-value="low" data-label="Low" data-description="Low thinking budget" onclick="selectComboboxOption('thinkingLevelCombobox', this)">Low</button>
                             <button type="button" role="option" aria-selected="false" class="w-full px-3 py-2 text-left hover:bg-gray-50 focus:bg-gray-50 focus:outline-none" data-combobox-option data-value="medium" data-label="Medium" data-description="Medium thinking budget" onclick="selectComboboxOption('thinkingLevelCombobox', this)">Medium</button>
                             <button type="button" role="option" aria-selected="false" class="w-full px-3 py-2 text-left hover:bg-gray-50 focus:bg-gray-50 focus:outline-none" data-combobox-option data-value="high" data-label="High" data-description="High thinking budget" onclick="selectComboboxOption('thinkingLevelCombobox', this)">High</button>
-                            <button type="button" role="option" aria-selected="false" class="w-full px-3 py-2 text-left hover:bg-gray-50 focus:bg-gray-50 focus:outline-none" data-combobox-option data-value="xhigh" data-label="XHigh" data-description="Maximum thinking budget" onclick="selectComboboxOption('thinkingLevelCombobox', this)">XHigh</button>
+                            <button type="button" role="option" aria-selected="false" class="w-full px-3 py-2 text-left hover:bg-gray-50 focus:bg-gray-50 focus:outline-none" data-combobox-option data-value="xhigh" data-label="XHigh" data-description="Extended thinking budget" onclick="selectComboboxOption('thinkingLevelCombobox', this)">XHigh</button>
+                            <button type="button" role="option" aria-selected="false" class="w-full px-3 py-2 text-left hover:bg-gray-50 focus:bg-gray-50 focus:outline-none" data-combobox-option data-value="max" data-label="Max" data-description="Maximum thinking budget" onclick="selectComboboxOption('thinkingLevelCombobox', this)">Max</button>
                         </div>
                     </div>
                 </div>

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/kimi_k3/client.py

@@ -78,6 +78,7 @@ class KimiK3Client(LLMClient):
             ThinkingLevel.MEDIUM: {"type": "enabled", "keep": "all"},
             ThinkingLevel.HIGH: {"type": "enabled", "keep": "all"},
             ThinkingLevel.XHIGH: {"type": "enabled", "keep": "all"},
+            ThinkingLevel.MAX: {"type": "enabled", "keep": "all"},
         }
         return mapping.get(thinking_level)
 
@@ -92,6 +93,7 @@ class KimiK3Client(LLMClient):
             ThinkingLevel.MEDIUM: "high",
             ThinkingLevel.HIGH: "high",
             ThinkingLevel.XHIGH: "max",
+            ThinkingLevel.MAX: "max",
         }
         return mapping.get(thinking_level)
 
@@ -267,7 +269,9 @@ class KimiK3Client(LLMClient):
         usage_metadata: UsageMetadata | None = None
         finish_reason: FinishReason | None = None
 
-        if len(model_output.choices) > 0:
+        # gateways inject content-free heartbeat chunks on long generations, whose choices
+        # the SDK leaves as None rather than an empty list
+        if model_output.choices:
             choice = model_output.choices[0]
             delta = choice.delta
 

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/minimax_m3/client.py

@@ -63,6 +63,8 @@ class MiniMaxM3Client(LLMClient):
             ThinkingLevel.MEDIUM: "medium",
             ThinkingLevel.HIGH: "high",
             ThinkingLevel.XHIGH: "high",
+            # MiniMax stops at "high"
+            ThinkingLevel.MAX: "high",
         }
         return mapping[thinking_level]
 
@@ -248,6 +250,7 @@ class MiniMaxM3Client(LLMClient):
             "response.output_item.done",
             "response.content_part.added",
             "response.content_part.done",
+            "keepalive",  # gateway heartbeat on long generations; carries no content
         ):
             raise ValueError(f"Unknown output: {model_output}")
 

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/openai_chat/client.py

@@ -220,7 +220,9 @@ class OpenaiChatClient(LLMClient):
         usage_metadata: UsageMetadata | None = None
         finish_reason: FinishReason | None = None
 
-        if len(model_output.choices) > 0:
+        # gateways inject content-free heartbeat chunks on long generations, whose choices
+        # the SDK leaves as None rather than an empty list
+        if model_output.choices:
             choice = model_output.choices[0]
             delta = choice.delta
 

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/openai_responses/client.py

@@ -54,6 +54,7 @@ class OpenaiResponsesClient(LLMClient):
             ThinkingLevel.MEDIUM: "medium",
             ThinkingLevel.HIGH: "high",
             ThinkingLevel.XHIGH: "xhigh",
+            ThinkingLevel.MAX: "max",
         }
         return mapping.get(thinking_level)
 
@@ -298,6 +299,7 @@ class OpenaiResponsesClient(LLMClient):
             "response.reasoning_summary_text.done",
             "response.content_part.added",
             "response.content_part.done",
+            "keepalive",  # gateway heartbeat on long generations; carries no content
         ):
             event_type = "unused"
 

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/agenthub/types.py

@@ -24,6 +24,7 @@ class ThinkingLevel(StrEnum):
     MEDIUM = "medium"
     HIGH = "high"
     XHIGH = "xhigh"
+    MAX = "max"
 
 
 class PromptCaching(StrEnum):

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agenthub-python"
-version = "0.4.2"
+version = "0.4.4"
 description = "AgentHub is the LLM API Hub for the Agent era, built for high-precision autonomous agents."
 keywords = [
     "agent",
@@ -33,6 +33,7 @@ dev = [
     "httpx[socks]",
     "pytest>=8.4.2",
     "pytest-asyncio>=0.23.0",
+    "pytest-xdist>=3.6.0",
     "ruff>=0.14.3",
     "pillow>=10.0.0",
 ]

{agenthub_python-0.4.2 → agenthub_python-0.4.4}/pyproject.toml.orig

@@ -1,6 +1,6 @@
 [project]
 name = "agenthub-python"
-version = "0.4.2"
+version = "0.4.4"
 description = "AgentHub is the LLM API Hub for the Agent era, built for high-precision autonomous agents."
 keywords = ["agent", "llm", "gemini", "claude", "gpt"]
 readme = "README.md"
@@ -17,7 +17,7 @@ Repository = "https://github.com/Prism-Shadow/agenthub"
 Issues = "https://github.com/Prism-Shadow/agenthub/issues"
 
 [project.optional-dependencies]
-dev = ["httpx[socks]", "pytest>=8.4.2", "pytest-asyncio>=0.23.0", "ruff>=0.14.3", "pillow>=10.0.0"]
+dev = ["httpx[socks]", "pytest>=8.4.2", "pytest-asyncio>=0.23.0", "pytest-xdist>=3.6.0", "ruff>=0.14.3", "pillow>=10.0.0"]
 
 [build-system]
 requires = ["uv_build>=0.9.16,<0.10.0"]