agenthub-python 0.2.0__py3-none-any.whl → 0.3.1__py3-none-any.whl

agenthub/auto_client.py

@@ -46,22 +46,26 @@ class AutoLLMClient(LLMClient):
     ) -> LLMClient:
         """Create the appropriate client for the given model."""
         client_type = client_type or os.getenv("CLIENT_TYPE", model.lower())
-        if "gemini-3" in client_type:  # e.g., gemini-3-flash-preview
+        if "gemini-3-" in client_type or "gemini-3.1-" in client_type:  # e.g., gemini-3-flash-preview
             from .gemini3 import Gemini3Client
 
             return Gemini3Client(model=model, api_key=api_key, base_url=base_url)
-        elif "claude" in client_type and "4-5" in client_type:  # e.g., claude-sonnet-4-5
-            from .claude4_5 import Claude4_5Client
+        elif "claude" in client_type and "4-6" in client_type:  # e.g., claude-sonnet-4-6
+            from .claude4_6 import Claude4_6Client
 
-            return Claude4_5Client(model=model, api_key=api_key, base_url=base_url)
-        elif "gpt-5.1" in client_type or "gpt-5.2" in client_type:  # e.g., gpt-5.2
-            from .gpt5_2 import GPT5_2Client
+            return Claude4_6Client(model=model, api_key=api_key, base_url=base_url)
+        elif "gpt-5.4" in client_type or "gpt-5.5" in client_type:  # e.g., gpt-5.5
+            from .gpt5_5 import GPT5_5Client
 
-            return GPT5_2Client(model=model, api_key=api_key, base_url=base_url)
-        elif "glm-4.7" in client_type:  # e.g., glm-4.7
-            from .glm4_7 import GLM4_7Client
+            return GPT5_5Client(model=model, api_key=api_key, base_url=base_url)
+        elif "glm-5" in client_type:
+            from .glm5 import GLM5Client
 
-            return GLM4_7Client(model=model, api_key=api_key, base_url=base_url)
+            return GLM5Client(model=model, api_key=api_key, base_url=base_url)
+        elif "kimi-k2.5" in client_type:
+            from .kimi_k2_5 import KimiK2_5Client
+
+            return KimiK2_5Client(model=model, api_key=api_key, base_url=base_url)
         elif "qwen3" in client_type:
             from .qwen3 import Qwen3Client
 
@@ -69,7 +73,7 @@ class AutoLLMClient(LLMClient):
         else:
             raise ValueError(
                 f"{client_type} is not supported. "
-                "Supported client types: gemini-3, claude-4-5, gpt-5.2, glm-4.7, qwen3."
+                "Supported client types: gemini-3, claude-4-6, gpt-5.4, gpt-5.5, glm-5, kimi-k2.5, qwen3."
             )
 
     def transform_uni_config_to_model_config(self, config: UniConfig) -> Any:
@@ -84,6 +88,13 @@ class AutoLLMClient(LLMClient):
         """Delegate to underlying client's transform_model_output_to_uni_event."""
         return self._client.transform_model_output_to_uni_event(model_output)
 
+    async def _streaming_response_internal(
+        self,
+        messages: list[UniMessage],
+        config: UniConfig,
+    ) -> AsyncIterator[UniEvent]:
+        raise NotImplementedError("Please use streaming_response instead.")
+
     async def streaming_response(
         self,
         messages: list[UniMessage],
@@ -115,3 +126,7 @@ class AutoLLMClient(LLMClient):
     def get_history(self) -> list[UniMessage]:
         """Get history from the underlying client."""
         return self._client.get_history()
+
+    def set_history(self, history: list[UniMessage]) -> None:
+        """Set history in the underlying client."""
+        self._client.set_history(history)

agenthub/base_client.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import time
 from abc import ABC, abstractmethod
 from typing import Any, AsyncIterator
 
@@ -84,21 +85,31 @@ class LLMClient(ABC):
         content_items: list[ContentItem] = []
         usage_metadata: UsageMetadata | None = None
         finish_reason: FinishReason | None = None
+        created_at: int | None = None
 
         for event in events:
             # Merge content_items from all events
             for item in event["content_items"]:
                 if item["type"] == "text":
-                    if content_items and content_items[-1]["type"] == "text":
+                    if (
+                        content_items
+                        and content_items[-1]["type"] == "text"
+                        and content_items[-1].get("signature") is None  # no signature yet
+                        and item.get("phase") is None  # no new phase
+                    ):
                         content_items[-1]["text"] += item["text"]
-                        if "signature" in item:  # signature may appear at the last item
+                        if "signature" in item:  # finish the current item if signature is not None
                             content_items[-1]["signature"] = item["signature"]
-                    elif item["text"]:  # omit empty text items
+                    elif item["text"] or item.get("phase") is not None:  # text or new phase starts an item
                         content_items.append(item.copy())
                 elif item["type"] == "thinking":
-                    if content_items and content_items[-1]["type"] == "thinking":
+                    if (
+                        content_items
+                        and content_items[-1]["type"] == "thinking"
+                        and content_items[-1].get("signature") is None  # no signature yet
+                    ):
                         content_items[-1]["thinking"] += item["thinking"]
-                        if "signature" in item:  # signature may appear at the last item
+                        if "signature" in item:  # finish the current item if signature is not None
                             content_items[-1]["signature"] = item["signature"]
                     elif item["thinking"] or item.get("signature"):  # omit empty thinking items
                         content_items.append(item.copy())
@@ -110,15 +121,37 @@ class LLMClient(ABC):
 
             usage_metadata = event.get("usage_metadata")  # usage_metadata is taken from the last event
             finish_reason = event.get("finish_reason")  # finish_reason is taken from the last event
+            created_at = event.get("created_at")  # created_at is taken from the last event
 
         return {
             "role": "assistant",
             "content_items": content_items,
             "usage_metadata": usage_metadata,
             "finish_reason": finish_reason,
+            "created_at": created_at,
         }
 
     @abstractmethod
+    async def _streaming_response_internal(
+        self,
+        messages: list[UniMessage],
+        config: UniConfig,
+    ) -> AsyncIterator[UniEvent]:
+        """
+        Internal method to handle streaming response.
+
+        This method should be implemented by each model client to handle
+        the actual streaming request and yield model-specific events.
+
+        Args:
+            messages: List of universal message dictionaries
+            config: Universal configuration dict
+
+        Yields:
+            Model-specific events from the streaming response
+        """
+        pass
+
     async def streaming_response(
         self,
         messages: list[UniMessage],
@@ -138,7 +171,28 @@ class LLMClient(ABC):
         Yields:
             Universal events from the streaming response
         """
-        pass
+        # Stamp any messages that don't yet have a created_at timestamp
+        for msg in messages:
+            if "created_at" not in msg:
+                msg["created_at"] = int(time.time() * 1000)
+
+        last_event: UniEvent | None = None
+        events = []
+        async for event in self._streaming_response_internal(messages, config):
+            event["created_at"] = int(time.time() * 1000)
+            last_event = event
+            events.append(event)
+            yield event
+
+        self._validate_last_event(last_event)
+
+        # Save history to file if trace_id is specified
+        if config.get("trace_id") and events:
+            from .integration.tracer import Tracer
+
+            assistant_message = self.concat_uni_events_to_uni_message(events)
+            tracer = Tracer()
+            tracer.save_history(self._model, messages + [assistant_message], config["trace_id"], config)
 
     async def streaming_response_stateful(
         self,
@@ -159,26 +213,43 @@ class LLMClient(ABC):
         Yields:
             Universal events from the streaming response
         """
-        # Add user message to history
-        self._history.append(message)
+        # Build a temporary messages list for inference without mutating history yet
+        temp_messages = self._history + [message]
 
         # Collect all events for history
         events = []
-        async for event in self.streaming_response(messages=self._history, config=config):
+        async for event in self.streaming_response(messages=temp_messages, config=config):
             events.append(event)
             yield event
 
-        # Convert events to message and add to history
+        # Only update history after successful inference
+        # temp_messages[-1] is the user message, now stamped with created_at by streaming_response
         if events:
             assistant_message = self.concat_uni_events_to_uni_message(events)
+            self._history.append(temp_messages[-1])
             self._history.append(assistant_message)
 
-        # Save history to file if trace_id is specified
-        if config.get("trace_id"):
-            from .integration.tracer import Tracer
+    @staticmethod
+    def _validate_last_event(last_event: UniEvent | None) -> None:
+        """Validate that the last event has usage_metadata and finish_reason.
 
-            tracer = Tracer()
-            tracer.save_history(self._model, self._history, config["trace_id"], config)
+        This validation guards against servers that silently terminate streaming
+        output partway through without sending a proper final event.
+
+        Args:
+            last_event: The last event yielded by streaming_response
+
+        Raises:
+            ValueError: If last_event is None or missing usage_metadata/finish_reason
+        """
+        if last_event is None:
+            raise ValueError("Streaming response yielded no events")
+
+        if last_event["usage_metadata"] is None:
+            raise ValueError(f"Last event must carry usage_metadata, got: {last_event}")
+
+        if last_event["finish_reason"] is None:
+            raise ValueError(f"Last event must carry finish_reason, got: {last_event}")
 
     def clear_history(self) -> None:
         """Clear the message history."""
@@ -187,3 +258,11 @@ class LLMClient(ABC):
     def get_history(self) -> list[UniMessage]:
         """Get the current message history."""
         return self._history.copy()
+
+    def set_history(self, history: list[UniMessage]) -> None:
+        """Replace the message history with a copy of the provided history.
+
+        Args:
+            history: List of universal message dictionaries to set as the new history
+        """
+        self._history = list(history)

agenthub/{claude4_5 → claude4_6}/__init__.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from .client import Claude4_5Client
+from .client import Claude4_6Client
 
 
-__all__ = ["Claude4_5Client"]
+__all__ = ["Claude4_6Client"]