diff --git a/changelog/4219.added.md b/changelog/4219.added.md
new file mode 100644
index 000000000..ddf4bc0fa
--- /dev/null
+++ b/changelog/4219.added.md
@@ -0,0 +1 @@
+- Added `enable_prompt_caching` setting to `AWSBedrockLLMService` for Bedrock ConverseStream prompt caching.
diff --git a/src/pipecat/services/aws/llm.py b/src/pipecat/services/aws/llm.py
index 722072592..52c95070d 100644
--- a/src/pipecat/services/aws/llm.py
+++ b/src/pipecat/services/aws/llm.py
@@ -36,6 +36,7 @@ from pipecat.frames.frames import (
     FunctionCallInProgressFrame,
     FunctionCallResultFrame,
     LLMContextFrame,
+    LLMEnablePromptCachingFrame,
     LLMFullResponseEndFrame,
     LLMFullResponseStartFrame,
     UserImageRawFrame,
@@ -66,11 +67,16 @@ class AWSBedrockLLMSettings(LLMSettings):
     Parameters:
         stop_sequences: List of strings that stop generation.
         latency: Performance mode - "standard" or "optimized".
+        enable_prompt_caching: Whether to enable prompt caching by adding cachePoint
+            markers to system prompts and tool definitions. Can reduce TTFT by up to
+            85% for multi-turn conversations. See:
+            https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html
         additional_model_request_fields: Additional model-specific parameters.
     """
 
     stop_sequences: List[str] | None | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
     latency: str | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
+    enable_prompt_caching: bool | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
     additional_model_request_fields: Dict[str, Any] | _NotGiven = field(
         default_factory=lambda: NOT_GIVEN
     )
@@ -174,6 +180,7 @@ class AWSBedrockLLMService(LLMService):
             user_turn_completion_config=None,
             stop_sequences=None,
             latency=None,
+            enable_prompt_caching=False,
             additional_model_request_fields={},
         )
 
@@ -455,6 +462,24 @@ class AWSBedrockLLMService(LLMService):
             if self._settings.latency in ["standard", "optimized"]:
                 request_params["performanceConfig"] = {"latency": self._settings.latency}
 
+            # Add cache checkpoints to system prompts and tool definitions.
+            # This enables prompt caching for providers that support it (e.g.
+            # Anthropic Claude on Bedrock), reducing TTFT by up to 85% on
+            # multi-turn conversations where the system prompt stays constant.
+            if self._settings.enable_prompt_caching:
+                if "system" in request_params and request_params["system"]:
+                    system_list = request_params["system"]
+                    if not any("cachePoint" in item for item in system_list):
+                        system_list.append({"cachePoint": {"type": "default"}})
+                if (
+                    "toolConfig" in request_params
+                    and "tools" in request_params["toolConfig"]
+                    and request_params["toolConfig"]["tools"]
+                ):
+                    tools_list = request_params["toolConfig"]["tools"]
+                    if not any("cachePoint" in t for t in tools_list):
+                        tools_list.append({"cachePoint": {"type": "default"}})
+
             # Log request params with messages redacted for logging
             adapter = self.get_llm_adapter()
             messages_for_logging = adapter.get_messages_for_logging(context)
@@ -566,6 +591,9 @@ class AWSBedrockLLMService(LLMService):
 
         if isinstance(frame, LLMContextFrame):
             await self._process_context(frame.context)
+        elif isinstance(frame, LLMEnablePromptCachingFrame):
+            logger.debug(f"Setting enable prompt caching to: [{frame.enable}]")
+            self._settings.enable_prompt_caching = frame.enable
         else:
             await self.push_frame(frame, direction)