agentkernel 0.2.14__tar.gz → 0.3.0__tar.gz

{agentkernel-0.2.14 → agentkernel-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: agentkernel
-Version: 0.2.14
+Version: 0.3.0
 Summary: Agent Kernel - Unified AI Agents Runtime
 Author: Yaala Labs
 Author-email: Yaala Labs <agentkernel@yaalalabs.com>
@@ -38,6 +38,7 @@ Requires-Dist: langchain-community~=0.4.1 ; extra == 'langgraph'
 Requires-Dist: litellm~=1.74.9 ; extra == 'langgraph'
 Requires-Dist: fastmcp>=2.12.4 ; extra == 'mcp'
 Requires-Dist: httpx>=0.27.0 ; extra == 'messenger'
+Requires-Dist: litellm>=1.74.9 ; extra == 'multimodal'
 Requires-Dist: openai-agents>=0.6.5 ; extra == 'openai'
 Requires-Dist: openinference-instrumentation-openai-agents>=1.4.0 ; extra == 'openai'
 Requires-Dist: openai-guardrails>=0.2.1 ; extra == 'openai'
@@ -73,6 +74,7 @@ Provides-Extra: langfuse
 Provides-Extra: langgraph
 Provides-Extra: mcp
 Provides-Extra: messenger
+Provides-Extra: multimodal
 Provides-Extra: openai
 Provides-Extra: openllmetry
 Provides-Extra: redis
@@ -263,6 +265,32 @@ OpenAIModule([assistant])
 handler = Lambda.handler
 ```
 
+**Note that this is just the simple serverless version. A more advanced serverless deployment mode, which uses queues for scalability, is also available. For queue-backed execution modes and response-store configuration, see the [AWS Serverless Deployment](https://github.com/yaalalabs/agent-kernel/tree/develop/docs/docs/deployment/aws-serverless.md) guide.**
+
+The AWS serverless handler accepts both a direct `BaseRunRequest` payload and the normalized `BaseRequest` envelope. If a flat run payload is provided, Agent Kernel generates a `request_id` and normalizes it before processing. 
+
+Accepted payloads:
+
+```json
+{
+  "prompt": "Hello agent",
+  "agent": "assistant",
+  "session_id": "user-123"
+}
+```
+
+```json
+{
+  "request_id": "req-123",
+  "user_id": "user-123",
+  "body": {
+    "prompt": "Hello agent",
+    "agent": "assistant",
+    "session_id": "user-123"
+  }
+}
+```
+
 ### Azure Functions Deployment
 
 Deploy your agents as Azure Functions using the built-in Azure handler.
@@ -284,11 +312,18 @@ handler = AzureFunctions.handler
 
 ```json
 {
-  "prompt": "Hello agent",
-  "agent": "assistant"
+  "request_id": "req-123",
+  "user_id": "user-123",
+  "body": {
+    "prompt": "Hello agent",
+    "agent": "assistant",
+    "session_id": "user-123"
+  }
 }
 ```
 
+Azure Functions also accepts the normalized envelope, and flat run payloads are normalized in the same way before the request reaches the agent runtime.
+
 **Response Format:**
 
 ```json
@@ -369,6 +404,72 @@ Required when `session.type=redis`:
   - **Description**: Key prefix for session storage
   - **Environment Variable**: `AK_SESSION__REDIS__PREFIX`
 
+#### Execution Configuration
+
+Configure queue-backed and serverless execution behavior.
+
+- **Execution Mode**
+  - **Field**: `execution.mode`
+  - **Options**: `rest_sync`, `rest_async`, `stream`, `async`
+  - **Default**: `None`
+  - **Description**: Selects the Lambda execution mode
+  - **Environment Variable**: `AK_EXECUTION__MODE`
+
+- **Queues**
+  - **Field**: `execution.queues`
+  - **Description**: Queue settings used by serverless execution modes
+
+  - **Input Queue URL**
+    - **Field**: `execution.queues.input.url`
+    - **Default**: `None`
+    - **Environment Variable**: `AK_EXECUTION__QUEUES__INPUT__URL`
+
+  - **Output Queue URL**
+    - **Field**: `execution.queues.output.url`
+    - **Default**: `None`
+    - **Environment Variable**: `AK_EXECUTION__QUEUES__OUTPUT__URL`
+
+  - **Input Queue Max Receive Count**
+    - **Field**: `execution.queues.input.max_receive_count`
+    - **Default**: `3`
+    - **Environment Variable**: `AK_EXECUTION__QUEUES__INPUT__MAX_RECEIVE_COUNT`
+
+  - **Output Queue Max Receive Count**
+    - **Field**: `execution.queues.output.max_receive_count`
+    - **Default**: `3`
+    - **Environment Variable**: `AK_EXECUTION__QUEUES__OUTPUT__MAX_RECEIVE_COUNT`
+
+- **Response Store**
+  - **Field**: `execution.response_store`
+  - **Description**: Response persistence settings used by the serverless response handler
+
+  - **Type**
+    - **Field**: `execution.response_store.type`
+    - **Description**: Response store backend selector configured in `config.yaml`; this value is not exported as an environment variable
+
+  - **Retry Count**
+    - **Field**: `execution.response_store.retry_count`
+    - **Default**: `5`
+    - **Description**: Number of lookup attempts when polling for a response
+    - **Environment Variable**: `AK_EXECUTION__RESPONSE_STORE__RETRY_COUNT`
+
+  - **Delay**
+    - **Field**: `execution.response_store.delay`
+    - **Default**: `5`
+    - **Description**: Delay in seconds between response lookup attempts
+    - **Environment Variable**: `AK_EXECUTION__RESPONSE_STORE__DELAY`
+
+  - **Redis Backend**
+    - **Field**: `execution.response_store.redis`
+    - **Environment Variables**: `AK_EXECUTION__RESPONSE_STORE__REDIS__URL`, `AK_EXECUTION__RESPONSE_STORE__REDIS__PREFIX`, `AK_EXECUTION__RESPONSE_STORE__REDIS__TTL`
+
+  - **DynamoDB Backend**
+    - **Field**: `execution.response_store.dynamodb`
+    - **Environment Variables**: `AK_EXECUTION__RESPONSE_STORE__DYNAMODB__TABLE_NAME`, `AK_EXECUTION__RESPONSE_STORE__DYNAMODB__TTL`
+    - **Description**: DynamoDB-backed response storage with table name and TTL
+
+Use either Redis or DynamoDB for the response store backend. The runtime accepts `BaseRunRequest` payloads directly, normalizes them internally when queueing is required, and uses `request_id` plus optional `user_id` as SQS message attributes.
+
 #### API Configuration
 
 Configure the REST API server (if using the API module).
@@ -862,6 +963,27 @@ session:
     url: redis://localhost:6379
     ttl: 604800
     prefix: "ak:sessions:"
+execution:
+  mode: rest_sync
+  queues:
+    input:
+      url: https://sqs.<region>.amazonaws.com/<accountno>/<queuename>
+      max_receive_count: 3
+    output:
+      url: https://sqs.<region>.amazonaws.com/<accountno>/<queuename>
+      max_receive_count: 3
+  response_store:
+    type: redis
+    retry_count: 5
+    delay: 5
+    redis: # if this is given, then dynamodb response store part cannot be given
+      url: redis://localhost:6379
+      prefix: "ak:responses:"
+      ttl: 3600
+    dynamodb: # if this is given, then redis response store part cannot be given
+      table_name: table-name
+      table_arn: table-arn
+      ttl: 3600
 api:
   host: 0.0.0.0
   port: 8000

{agentkernel-0.2.14 → agentkernel-0.3.0}/README.md

@@ -178,6 +178,32 @@ OpenAIModule([assistant])
 handler = Lambda.handler
 ```
 
+**Note that this is just the simple serverless version. A more advanced serverless deployment mode, which uses queues for scalability, is also available. For queue-backed execution modes and response-store configuration, see the [AWS Serverless Deployment](https://github.com/yaalalabs/agent-kernel/tree/develop/docs/docs/deployment/aws-serverless.md) guide.**
+
+The AWS serverless handler accepts both a direct `BaseRunRequest` payload and the normalized `BaseRequest` envelope. If a flat run payload is provided, Agent Kernel generates a `request_id` and normalizes it before processing. 
+
+Accepted payloads:
+
+```json
+{
+  "prompt": "Hello agent",
+  "agent": "assistant",
+  "session_id": "user-123"
+}
+```
+
+```json
+{
+  "request_id": "req-123",
+  "user_id": "user-123",
+  "body": {
+    "prompt": "Hello agent",
+    "agent": "assistant",
+    "session_id": "user-123"
+  }
+}
+```
+
 ### Azure Functions Deployment
 
 Deploy your agents as Azure Functions using the built-in Azure handler.
@@ -199,11 +225,18 @@ handler = AzureFunctions.handler
 
 ```json
 {
-  "prompt": "Hello agent",
-  "agent": "assistant"
+  "request_id": "req-123",
+  "user_id": "user-123",
+  "body": {
+    "prompt": "Hello agent",
+    "agent": "assistant",
+    "session_id": "user-123"
+  }
 }
 ```
 
+Azure Functions also accepts the normalized envelope, and flat run payloads are normalized in the same way before the request reaches the agent runtime.
+
 **Response Format:**
 
 ```json
@@ -284,6 +317,72 @@ Required when `session.type=redis`:
   - **Description**: Key prefix for session storage
   - **Environment Variable**: `AK_SESSION__REDIS__PREFIX`
 
+#### Execution Configuration
+
+Configure queue-backed and serverless execution behavior.
+
+- **Execution Mode**
+  - **Field**: `execution.mode`
+  - **Options**: `rest_sync`, `rest_async`, `stream`, `async`
+  - **Default**: `None`
+  - **Description**: Selects the Lambda execution mode
+  - **Environment Variable**: `AK_EXECUTION__MODE`
+
+- **Queues**
+  - **Field**: `execution.queues`
+  - **Description**: Queue settings used by serverless execution modes
+
+  - **Input Queue URL**
+    - **Field**: `execution.queues.input.url`
+    - **Default**: `None`
+    - **Environment Variable**: `AK_EXECUTION__QUEUES__INPUT__URL`
+
+  - **Output Queue URL**
+    - **Field**: `execution.queues.output.url`
+    - **Default**: `None`
+    - **Environment Variable**: `AK_EXECUTION__QUEUES__OUTPUT__URL`
+
+  - **Input Queue Max Receive Count**
+    - **Field**: `execution.queues.input.max_receive_count`
+    - **Default**: `3`
+    - **Environment Variable**: `AK_EXECUTION__QUEUES__INPUT__MAX_RECEIVE_COUNT`
+
+  - **Output Queue Max Receive Count**
+    - **Field**: `execution.queues.output.max_receive_count`
+    - **Default**: `3`
+    - **Environment Variable**: `AK_EXECUTION__QUEUES__OUTPUT__MAX_RECEIVE_COUNT`
+
+- **Response Store**
+  - **Field**: `execution.response_store`
+  - **Description**: Response persistence settings used by the serverless response handler
+
+  - **Type**
+    - **Field**: `execution.response_store.type`
+    - **Description**: Response store backend selector configured in `config.yaml`; this value is not exported as an environment variable
+
+  - **Retry Count**
+    - **Field**: `execution.response_store.retry_count`
+    - **Default**: `5`
+    - **Description**: Number of lookup attempts when polling for a response
+    - **Environment Variable**: `AK_EXECUTION__RESPONSE_STORE__RETRY_COUNT`
+
+  - **Delay**
+    - **Field**: `execution.response_store.delay`
+    - **Default**: `5`
+    - **Description**: Delay in seconds between response lookup attempts
+    - **Environment Variable**: `AK_EXECUTION__RESPONSE_STORE__DELAY`
+
+  - **Redis Backend**
+    - **Field**: `execution.response_store.redis`
+    - **Environment Variables**: `AK_EXECUTION__RESPONSE_STORE__REDIS__URL`, `AK_EXECUTION__RESPONSE_STORE__REDIS__PREFIX`, `AK_EXECUTION__RESPONSE_STORE__REDIS__TTL`
+
+  - **DynamoDB Backend**
+    - **Field**: `execution.response_store.dynamodb`
+    - **Environment Variables**: `AK_EXECUTION__RESPONSE_STORE__DYNAMODB__TABLE_NAME`, `AK_EXECUTION__RESPONSE_STORE__DYNAMODB__TTL`
+    - **Description**: DynamoDB-backed response storage with table name and TTL
+
+Use either Redis or DynamoDB for the response store backend. The runtime accepts `BaseRunRequest` payloads directly, normalizes them internally when queueing is required, and uses `request_id` plus optional `user_id` as SQS message attributes.
+
 #### API Configuration
 
 Configure the REST API server (if using the API module).
@@ -777,6 +876,27 @@ session:
     url: redis://localhost:6379
     ttl: 604800
     prefix: "ak:sessions:"
+execution:
+  mode: rest_sync
+  queues:
+    input:
+      url: https://sqs.<region>.amazonaws.com/<accountno>/<queuename>
+      max_receive_count: 3
+    output:
+      url: https://sqs.<region>.amazonaws.com/<accountno>/<queuename>
+      max_receive_count: 3
+  response_store:
+    type: redis
+    retry_count: 5
+    delay: 5
+    redis: # if this is given, then dynamodb response store part cannot be given
+      url: redis://localhost:6379
+      prefix: "ak:responses:"
+      ttl: 3600
+    dynamodb: # if this is given, then redis response store part cannot be given
+      table_name: table-name
+      table_arn: table-arn
+      ttl: 3600
 api:
   host: 0.0.0.0
   port: 8000

{agentkernel-0.2.14 → agentkernel-0.3.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "uv_build"
 
 [project]
 name = "agentkernel"
-version = "0.2.14"
+version = "0.3.0"
 description = "Agent Kernel - Unified AI Agents Runtime"
 readme = "README.md"
 requires-python = ">=3.12,<3.14"
@@ -115,6 +115,9 @@ a2a = [
 mcp = [
     "fastmcp>=2.12.4",
 ]
+multimodal = [
+    "litellm>=1.74.9",
+]
 
 [project.scripts]
 ak = "agentkernel.cli.ak:main"

{agentkernel-0.2.14 → agentkernel-0.3.0}/src/agentkernel/api/handler.py

@@ -15,6 +15,7 @@ from agentkernel.core.model import (
     AgentRequestFile,
     AgentRequestImage,
     AgentRequestText,
+    BaseRunRequest,
 )
 
 from ..core import AgentService, Config
@@ -67,12 +68,9 @@ class AgentRESTRequestHandler(RESTRequestHandler):
         name: str
         mime_type: Optional[str] = None
 
-    class RunRequest(BaseModel):
+    class RunRequest(BaseRunRequest):
         model_config = ConfigDict(extra="allow")
 
-        prompt: str
-        agent: Optional[str] = None
-        session_id: Optional[str] = None
         files: Optional[List["AgentRESTRequestHandler.FileData"]] = None
         images: Optional[List["AgentRESTRequestHandler.ImageData"]] = None
 

agentkernel-0.3.0/src/agentkernel/auth.py

@@ -0,0 +1 @@
+from .auth import *

{agentkernel-0.2.14 → agentkernel-0.3.0}/src/agentkernel/core/__init__.py

@@ -24,7 +24,7 @@ from .model import (
 )
 from .config import AKConfig as Config
 from .module import Module
-from .runtime import Runtime, AuxiliaryCache
+from .runtime import Runtime
 from .service import AgentService
 from .hooks import PreHook, PostHook
 from .tool import ToolContext, ToolBuilder

{agentkernel-0.2.14 → agentkernel-0.3.0}/src/agentkernel/core/base.py

@@ -6,8 +6,6 @@ from collections.abc import Iterator
 from enum import Enum
 from typing import Any, ClassVar, Self, cast
 
-from deprecated import deprecated
-
 from .hooks import PostHook, PreHook
 from .model import AgentReply, AgentRequest
 from .util.key_value_cache import KeyValueCache
@@ -42,12 +40,6 @@ class Session:
         VOLATILE_CACHE = "v_cache"
         NON_VOLATILE_CACHE = "nv_cache"
 
-    VOLATILE_CACHE_KEY: str = Keys.VOLATILE_CACHE.value
-    """Deprecated since 0.2.12, use Session.Keys.VOLATILE_CACHE.value instead."""
-
-    NON_VOLATILE_CACHE_KEY: str = Keys.NON_VOLATILE_CACHE.value
-    """Deprecated since 0.2.12, use Session.Keys.NON_VOLATILE_CACHE.value instead."""
-
     current_session: ClassVar[contextvars.ContextVar[Self | None]] = contextvars.ContextVar("current_session", default=None)
 
     @classmethod
@@ -58,16 +50,6 @@ class Session:
         """
         return cls.current_session.get()
 
-    @classmethod
-    @deprecated(version="0.2.12", reason="Use Session.current() instead.")
-    def get_current_session_id(cls) -> str:
-        """
-        Returns the current session identifier from the context variable.
-        :return: The current session identifier.
-        """
-        session = cls.current()
-        return session.id if session else ""
-
     def __init__(self, id: str):
         """
         Initializes a Session instance.
@@ -150,14 +132,6 @@ class Session:
         self._log.debug(f"Retrieved session data object for key {key}: {result}")
         return result
 
-    @deprecated(version="0.2.12", reason="Use get_all() instead.")
-    def get_all_keys(self):
-        """
-        Returns a list of all keys in the session data.
-        :return: A list of all keys in the session data.
-        """
-        return self._data.keys()
-
     def get_all(self, durable: bool = True, volatile: bool = True) -> Iterator[tuple[str, Any]]:
         """
         Returns all session data objects.
@@ -214,22 +188,6 @@ class Session:
         self.get_volatile_cache().clear()
         self.get_non_volatile_cache().clear()
 
-    @deprecated(version="0.2.12", reason="Use async with on the session to acquire it.")
-    def set_context(self):
-        """
-        Sets the current session context variable to this session.
-        """
-        self._token = Session.current_session.set(self)
-
-    @deprecated(version="0.2.12", reason="Use async with on the session to acquire it.")
-    def reset_context(self):
-        """
-        Resets the current session context variable to the previous value.
-        """
-        if self._token:
-            Session.current_session.reset(self._token)
-            self._token = None
-
 
 class Runner(ABC):
     """
@@ -330,42 +288,32 @@ class Agent(ABC):
         """
         return self._post_hooks
 
-    @deprecated(
-        version="0.2.12",
-        reason="Use Agent.pre_hooks.extend() instead. Note that unlike attach_pre_hooks(), extend() does not perform duplicate checking.",
-    )
-    def attach_pre_hooks(self, hooks: list[PreHook]):
+    @abstractmethod
+    def get_description(self) -> str:
         """
-        Attaches pre-execution hooks to the agent.
-        Duplicate hook objects are ignored to prevent multiple registrations
-        of the same hook.
-        :param hooks: List of pre-execution hooks to attach.
+        Returns the description of the agent.
         """
-        for hook in hooks:
-            if hook not in self._pre_hooks:
-                self._pre_hooks.append(hook)
+        pass
 
-    @deprecated(
-        version="0.2.12",
-        reason="Use Agent.post_hooks.extend() instead. Note that unlike attach_post_hooks(), extend() does not perform duplicate checking.",
-    )
-    def attach_post_hooks(self, hooks: list[PostHook]):
+    @abstractmethod
+    def override_system_prompt(self, prompt: str) -> None:
         """
-        Attaches post-execution hooks to the agent.
-        Duplicate hook objects are ignored to prevent multiple registrations
-        of the same hook.
-        :param hooks: List of post-execution hooks to attach.
+        Appends additional instructions to the agent's system prompt.
+
+        Called by ``_setup_system_prompt()`` at init time to inject system-level
+        tool instructions (e.g., multimodal attachment analysis guidance).
+
+        :param prompt: The instruction text to append.
         """
-        for hook in hooks:
-            if hook not in self._post_hooks:
-                self._post_hooks.append(hook)
+        raise NotImplementedError()
 
     @abstractmethod
-    def get_description(self) -> str:
+    def attach_tool(self, tool: Any) -> None:
         """
-        Returns the description of the agent.
+        Attaches a tool to the agent.
+        :param tool: The tool to attach.
         """
-        pass
+        raise NotImplementedError()
 
     @abstractmethod
     def get_a2a_card(self) -> Any:
@@ -373,3 +321,22 @@ class Agent(ABC):
         Returns the A2A AgentCard associated with the agent.
         """
         pass
+
+    def _setup_system_prompt(self) -> None:
+        """
+        Appends system-defined instructions to the agent's prompt during initialization.
+        Should be invoked by subclasses in their initialization process after configuration.
+        """
+        from agentkernel.core.tool import SystemToolFactory
+
+        suffix = SystemToolFactory.get_system_prompt_suffix()
+        self.override_system_prompt(prompt=suffix)
+
+    def _attach_system_tools(self) -> None:
+        """
+        Attaches system-level tools to the agent during initialization.
+        """
+        from agentkernel.core.tool import SystemToolFactory
+
+        for tool in SystemToolFactory.get_all():
+            self.attach_tool(tool.func)

{agentkernel-0.2.14 → agentkernel-0.3.0}/src/agentkernel/core/config.py

@@ -3,6 +3,7 @@ from typing import List, Optional
 
 from pydantic import BaseModel, Field
 
+from .model import ExecutionMode
 from .util.config_yaml_util import YamlBaseSettingsModified
 
 
@@ -62,7 +63,7 @@ class _APIConfig(BaseModel):
     port: int = Field(default=8000, description="API port")
     enabled_routes: _RoutesConfig = Field(description="API route flags", default_factory=_RoutesConfig)
     custom_router_prefix: str = Field(default="/custom", description="Custom router prefix")
-    max_file_size: int = Field(default=2097152, description="Maximum file size in bytes (default: 2 MB)")
+    max_file_size: int = Field(default=20971520, description="Maximum file size in bytes (default: 20 MB)")
 
 
 class _A2AConfig(BaseModel):
@@ -131,6 +132,43 @@ class _GmailConfig(BaseModel):
     label_filter: str = Field(default="INBOX", description="Gmail label to monitor (e.g., INBOX, UNREAD)")
 
 
+class _MultimodalStorageRedisConfig(BaseModel):
+    url: str = Field(default="redis://localhost:6379", description="Redis connection URL")
+    ttl: int = Field(default=604800, description="Attachment TTL in seconds")
+    prefix: str = Field(default="ak:attachments:", description="Key prefix for attachment keys")
+
+
+class _MultimodalStorageDynamoDBConfig(BaseModel):
+    table_name: str = Field(default="ak-attachments", description="DynamoDB table name for attachment storage")
+    ttl: int = Field(default=604800, description="Attachment TTL in seconds (0 disables)")
+
+
+class _MultimodalConfig(BaseModel):
+    """Configuration for multimodal attachment memory."""
+
+    enabled: bool = Field(
+        default=False,
+        description="Enable multimodal memory for images and files.",
+    )
+    storage_type: str = Field(
+        default="in_memory",
+        pattern="^(session_cache|in_memory|redis|dynamodb)$",
+        description="Storage backend for multimodal attachments. Options: session_cache, in_memory, redis, dynamodb",
+    )
+    max_attachments: int = Field(default=20, description="Maximum number of attachments to keep per session")
+    description_max_length: int = Field(default=200, description="Maximum length of attachment description text")
+    description_model: str = Field(
+        default="gpt-4o",
+        description="LiteLLM model used to generate brief descriptions when an attachment is first received (called by the pre-hook)",
+    )
+    analysis_model: str = Field(
+        default="gpt-4o",
+        description="LiteLLM model used by the analyze_attachments tool when the agent requests a full analysis of an attachment",
+    )
+    redis: Optional[_MultimodalStorageRedisConfig] = None
+    dynamodb: Optional[_MultimodalStorageDynamoDBConfig] = None
+
+
 class _TraceConfig(BaseModel):
     enabled: bool = Field(default=False, description="Enable tracing")
     type: str = Field(default="langfuse", pattern="^(langfuse|openllmetry)$")
@@ -162,6 +200,57 @@ class _GuardrailConfig(BaseModel):
     output: _GuardrailParamConfig = Field(description="Output Guardrail configuration", default_factory=_GuardrailParamConfig)
 
 
+class _ResponseStoreRedisConfig(_RedisConfig):
+    prefix: str = Field(default="ak:responses:", description="Key prefix for Redis response storage")
+
+
+class _ResponseStoreDynamoDBConfig(_DynamoDBConfig):
+    table_name: Optional[str] = Field(
+        default=None,
+        description="DynamoDB table name for session storage.",
+    )
+
+
+class _ResponseStoreConfig(BaseModel):
+    type: str = Field(default=None, pattern="^(redis|dynamodb)$")
+    retry_count: int = Field(default=5, description="Number of retry attempts for response store reads")
+    delay: float = Field(default=5, description="Delay in seconds between response store reads retry attempts")
+    redis: Optional[_ResponseStoreRedisConfig] = None
+    dynamodb: Optional[_ResponseStoreDynamoDBConfig] = None
+
+
+class _InputQueueConfig(BaseModel):
+    url: str = Field(default="", description="Input SQS queue URL for async execution mode")
+    max_receive_count: int = Field(
+        default=3, description="Maximum number of times a message can be received from input queue before being treated as permanently failed"
+    )
+
+
+class _OutputQueueConfig(BaseModel):
+    url: str = Field(default="", description="Output SQS queue URL for async execution mode")
+    max_receive_count: int = Field(
+        default=3, description="Maximum number of times a message can be received from output queue before being treated as permanently failed"
+    )
+
+
+class _QueuesConfig(BaseModel):
+    input: _InputQueueConfig = Field(default_factory=_InputQueueConfig, description="Input SQS queue configuration for async execution mode")
+    output: _OutputQueueConfig = Field(default_factory=_OutputQueueConfig, description="Output SQS queue configuration for async execution mode")
+
+
+class _ExecutionConfig(BaseModel):
+    mode: Optional[ExecutionMode] = Field(
+        default=None,
+        description="Execution mode: rest_sync for synchronous REST, rest_async for asynchronous REST",
+    )
+    queues: Optional[_QueuesConfig] = Field(default_factory=_QueuesConfig, description="Queue URLs for async execution mode")
+    # websocket_connection_table: Optional[str] = Field(default=None, description="DynamoDB table name for storing WebSocket connections")
+    response_store: Optional[_ResponseStoreConfig] = Field(
+        default=None,
+        description="Response storage configuration for async execution mode",
+    )
+
+
 class AKConfig(YamlBaseSettingsModified):
     debug: bool = Field(default=False, description="Enable debug mode")
     session: _SessionStoreConfig = Field(
@@ -180,10 +269,12 @@ class AKConfig(YamlBaseSettingsModified):
     instagram: _InstagramConfig = Field(description="Instagram Business API related configurations", default_factory=_InstagramConfig)
     telegram: _TelegramConfig = Field(description="Telegram Bot related configurations", default_factory=_TelegramConfig)
     gmail: _GmailConfig = Field(description="Gmail related configurations", default_factory=_GmailConfig)
+    multimodal: _MultimodalConfig = Field(description="Multimodal attachment memory configurations", default_factory=_MultimodalConfig)
 
     trace: _TraceConfig = Field(description="Tracing related configurations", default_factory=_TraceConfig)
     test: _TestConfig = Field(description="Test related configurations", default_factory=_TestConfig)
     guardrail: _GuardrailConfig = Field(description="Guardrail related configurations", default_factory=_GuardrailConfig)
+    execution: _ExecutionConfig = Field(description="Execution mode and queue related configurations", default_factory=_ExecutionConfig)
     library_version: str = Field(default=_get_ak_version(), description="Library version")
 
     @classmethod