refactor: improve BedrockClient docs and export from __init__

- Add comprehensive docstring with model ID formats and auth notes
- Document region_name behavior with inference profiles
- Export BedrockClient from llm_clients package

tradingagents/llm_clients/__init__.py

@@ -1,4 +1,5 @@
 from .base_client import BaseLLMClient
 from .factory import create_llm_client
+from .bedrock_client import BedrockClient
 
-__all__ = ["BaseLLMClient", "create_llm_client"]
+__all__ = ["BaseLLMClient", "create_llm_client", "BedrockClient"]

tradingagents/llm_clients/bedrock_client.py

@@ -3,30 +3,54 @@ from typing import Any, Optional
 from langchain_aws import ChatBedrockConverse
 
 from .base_client import BaseLLMClient
-from .validators import validate_model
 
 
 class BedrockClient(BaseLLMClient):
-    """Client for Amazon Bedrock models (Claude, Kimi, Qwen, GLM, etc.)."""
+    """Client for Amazon Bedrock models.
+
+    Supports any model available on Bedrock via IAM Role (no API key needed),
+    including Claude, Amazon Nova, Kimi, Qwen, GLM, DeepSeek, MiniMax, and more.
+
+    Authentication:
+        Uses boto3 default credential chain: IAM Role (EC2/Lambda), environment
+        variables (AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY), or ~/.aws/credentials.
+
+    Model ID formats:
+        - Cross-region inference profile (recommended):
+            ``us.anthropic.claude-haiku-4-5-20251001-v1:0``
+            ``eu.anthropic.claude-3-5-sonnet-20240620-v1:0``
+        - Direct on-demand (us-east-1 default region only):
+            ``amazon.nova-lite-v1:0``
+            ``moonshotai.kimi-k2.5``
+            ``qwen.qwen3-32b-v1:0``
+            ``zai.glm-4.7-flash``
+            ``deepseek.v3.2``
+
+    Note:
+        When specifying a non-default ``region_name``, use region-specific
+        inference profile IDs (e.g. ``us-west-2.anthropic.claude-...``),
+        as direct model IDs only support on-demand throughput in us-east-1.
+
+    Example::
+
+        config["llm_provider"]    = "bedrock"
+        config["deep_think_llm"]  = "us.anthropic.claude-haiku-4-5-20251001-v1:0"
+        config["quick_think_llm"] = "amazon.nova-micro-v1:0"
+    """
 
     def __init__(self, model: str, base_url: Optional[str] = None, **kwargs):
         super().__init__(model, base_url, **kwargs)
 
     def get_llm(self) -> Any:
         """Return configured ChatBedrockConverse instance."""
-        llm_kwargs = {"model_id": self.model}
+        llm_kwargs: dict = {"model_id": self.model}
 
-        if "region_name" in self.kwargs:
+        for key in ("region_name", "max_tokens", "callbacks", "timeout"):
-            llm_kwargs["region_name"] = self.kwargs["region_name"]
+            if key in self.kwargs:
-        if "max_tokens" in self.kwargs:
+                llm_kwargs[key] = self.kwargs[key]
-            llm_kwargs["max_tokens"] = self.kwargs["max_tokens"]
-        if "callbacks" in self.kwargs:
-            llm_kwargs["callbacks"] = self.kwargs["callbacks"]
-        if "timeout" in self.kwargs:
-            llm_kwargs["timeout"] = self.kwargs["timeout"]
 
         return ChatBedrockConverse(**llm_kwargs)
 
     def validate_model(self) -> bool:
-        """Validate model for Bedrock (pass-through, model IDs are flexible)."""
+        """Bedrock model IDs are dynamic; skip static validation."""
         return True