realtimex-agent-a2a-agent 0.4.2__py3-none-any.whl

realtimex_agent_a2a_agent/agent.py

@@ -0,0 +1,755 @@
+from any_agent import AgentConfig, AnyAgent
+from any_agent.config import MCPStdio
+import json
+import os
+from typing import Callable, Dict, Literal, Optional, Union, Any
+
+from .tools import send_email
+from .callbacks.tool_execution import ShowToolCalling
+from any_agent.callbacks import get_default_callbacks
+from mem0 import Memory
+from any_agent.tools import a2a_tool_async
+from any_agent.serving import A2AServingConfig
+
+
+MEMORIES_GLOBAL_VIRTUAL_ROOT = "/memories/global/"
+MEMORIES_WORKSPACE_VIRTUAL_ROOT_PREFIX = "/memories/workspace/"
+SKILLS_GLOBAL_VIRTUAL_ROOT = "/skills/global/"
+SKILLS_WORKSPACE_VIRTUAL_ROOT_PREFIX = "/skills/workspace/"
+
+
+def get_user_dir():
+    return os.path.join(os.path.expanduser("~"),".realtimex.ai")
+
+def get_base_user_dir():
+    return os.path.join(os.path.expanduser("~"))
+
+def get_uvx_executable():
+    unix_realtimex_uvx_path = os.path.join(get_user_dir(),"Resources","envs","bin","uvx")
+    if os.path.exists(unix_realtimex_uvx_path):
+        return unix_realtimex_uvx_path
+    win_realtimex_uvx_path = os.path.join(get_user_dir(),"Resources","envs","Scripts","uvx.exe")
+    if os.path.exists(win_realtimex_uvx_path):
+        return win_realtimex_uvx_path
+    return "uvx"
+
+def get_nvm_dir():
+    path = os.path.join(get_base_user_dir(),".nvm")
+    if os.path.exists(path):
+        return path
+    path = os.path.join('c:', os.sep, "nvm")
+    if os.path.exists(path):
+        return path
+    return ""
+
+
+def get_nvm_inc():
+    # /Users/phuongnguyen/.nvm/versions/node/v22.16.0/include/node
+    path = os.path.join(get_nvm_dir(),"versions","node","v22.16.0","include","node")
+    if os.path.exists(path):
+        return path
+    path = os.path.join('c:', os.sep, "nvm")
+    if os.path.exists(path):
+        return path
+    return ""
+
+def get_nvm_bin():
+    # /Users/phuongnguyen/.nvm/versions/node/v22.16.0/include/node
+    path = os.path.join(get_nvm_dir(),"versions","node","v22.16.0","bin")
+    if os.path.exists(path):
+        return path
+    path = os.path.join('c:', os.sep, "nvm")
+    if os.path.exists(path):
+        return path
+    return ""
+
+def get_npx_executable():
+    unix_realtimex_npx_path = os.path.join(get_base_user_dir(),".nvm","versions","node","v22.16.0","bin","npx")
+    if os.path.exists(unix_realtimex_npx_path):
+        return unix_realtimex_npx_path
+    win_realtimex_npx_path = os.path.join('c:', os.sep, "nvm", "v22.16.0", "npx.cmd")
+    if os.path.exists(win_realtimex_npx_path):
+        return win_realtimex_npx_path
+    return "npx"
+
+
+def get_deepagents_agent_path(scope, agent_id, workspace_slug=None):
+    """
+    Get the agent.md path for DeepAgents.
+    
+    Args:
+        scope: 'global' or 'workspace'
+        agent_id: The agent identifier
+        workspace_slug: Required when scope is 'workspace'
+    
+    Returns:
+        Path string if file exists, None otherwise
+    """
+    if scope == "global":
+        path = os.path.join(
+            get_user_dir(), "Resources", "agent-skills", "global", agent_id, "agent.md"
+        )
+    elif scope == "workspace" and workspace_slug:
+        path = os.path.join(
+            get_user_dir(), "Resources", "agent-skills", "workspaces", workspace_slug, agent_id, "agent.md"
+        )
+    else:
+        return None
+    
+    return path if os.path.isfile(path) else None
+
+
+def get_deepagents_skills_dir(scope, agent_id, workspace_slug=None):
+    """
+    Get the skills directory path for DeepAgents.
+    
+    Args:
+        scope: 'global' or 'workspace'
+        agent_id: The agent identifier
+        workspace_slug: Required when scope is 'workspace'
+    
+    Returns:
+        Path string if directory exists, None otherwise
+    """
+    if scope == "global":
+        path = os.path.join(
+            get_user_dir(), "Resources", "agent-skills", "global", agent_id, "skills"
+        )
+    elif scope == "workspace" and workspace_slug:
+        path = os.path.join(
+            get_user_dir(), "Resources", "agent-skills", "workspaces", workspace_slug, agent_id, "skills"
+        )
+    else:
+        return None
+    
+    return path if os.path.isdir(path) else None
+
+class RealTimeXAgent():
+    def __init__(
+        self, current_session_id
+    ):
+        self.agent_id: str = None
+        self.agent_data: Dict = None
+        self.agent: AnyAgent = None
+        self.agent_name: str = None
+        self.system_prompt: str = None
+        self.agent_framework: str = None
+        self.agent_description: str = None
+        self.default_model:str = None
+        self.recommended_agent_flows = None
+        self.recommended_aci_mcp_apps = None
+        self.recommended_local_mcp_apps = None
+        self.recommended_team_members = None
+        self.memory:Memory = None
+
+        self.current_session_id = current_session_id
+
+    async def prepare_llm(self, provider_name,model_name,api_base,api_key):
+        def get_provider_name(provider_name):
+            if provider_name == "realtimexai":
+                return "openai"
+            elif provider_name == "litellm":
+                return "openai"
+            elif provider_name == "openai":
+                return "openai"
+            elif provider_name == "ollama":
+                return "ollama"
+            elif provider_name == "gemini":
+                return "google"
+            elif provider_name == "google":
+                return "google"
+        
+        def get_model_name(model_name):
+            return model_name
+
+        provider_name = get_provider_name(provider_name)
+        model_name = get_model_name(model_name)
+
+        # if provider_name == "openai":
+        #     os.environ['OPENAI_BASE_URL'] = api_base
+        #     os.environ['OPENAI_API_KEY'] = api_key
+
+        # return {
+        #     "api_base": api_base,
+        #     "api_key": api_key,
+        #     "model_id": f"{provider_name}:{model_name}"
+        # }
+
+        return {
+            "api_base": api_base,
+            "api_key": api_key,
+            "model_id": f"{provider_name}/{model_name}"
+        }
+
+        # return {
+        #     "api_base": api_base,
+        #     "api_key": api_key,
+        #     "model_id": model_name
+        # }
+
+    async def create_subagents(self,instructions=None,tools=[],llm_config=None):
+        from openai import OpenAI
+        client = OpenAI(
+            api_key=llm_config["api_key"],
+            base_url=llm_config["api_base"],
+        )
+
+        system_prompt = self.system_prompt
+        if not instructions:
+            system_prompt = instructions
+        
+        schema = {
+            "title": "subagents",
+            "description": "The list of subagents to do the task well and effectively.",
+            "required": [
+                "subagents",
+            ],
+            "type": "object",
+            "properties": {
+                "subagents":{
+                    "type": "array",
+                    "description": "The list of subagents to do the task well and effectively.",
+                    "items": {
+                        "type": "object",
+                        "properties": {
+                            "name": {
+                                "type": "string",
+                                "description": "The name of the sub-agent"
+                            },
+                            "description": {
+                                "type": "string",
+                                "description": "A description of the sub-agent"
+                            },
+                            "prompt": {
+                                "type": "string",
+                                "description": "The prompt used by the sub-agent"
+                            },
+                            # "tools": {
+                            #     "type": "array",
+                            #     "description": "Optional list of tools name the sub-agent can use",
+                            #     "items": {
+                            #         "type": "string"
+                            #     }
+                            # }
+                        },
+                        "required": ["name", "description", "prompt"],
+                        "additionalProperties": False
+                    }
+                }
+            },
+            "additionalProperties": False
+        }
+
+        # print("schema", json.dumps(schema))
+
+        response_format = { "type": "json_schema", "json_schema": {"strict": True, "name": schema["title"], "schema": schema}}
+
+#         tools_str = ""
+#         for tool in tools:
+#             tools_str = f"""{tools_str}
+#   - {tool.__name__}: {tool.__doc__}"""
+
+        completion = client.beta.chat.completions.parse(
+            model=llm_config["model_id"],
+            messages=[{"role": "system", "content": 
+f"""You are tasked with designing a small team of specialized subagents to work together under the guidance of the main agent.
+
+* The main agent’s role and purpose is defined by:
+  {system_prompt}
+
+Your job is to create **no more than 5 subagents**. Each subagent must include:
+
+1. Name: lowercase, short, clear, and distinct, only alphabets and underscore allowed.
+2. Description: what this subagent is specialized at and how it contributes to the team.
+3. System Prompt: clear instructions that define the subagent’s behavior, style, and responsibilities.
+
+Guidelines:
+
+* Each subagent should have a well-defined, non-overlapping role.
+* The team should collectively cover all major aspects required for the main agent’s purpose.
+* Avoid redundancy—each subagent must bring unique value.
+* Keep the team **small (≤5 subagents)** but **effective**.
+
+Finally, return the result as a list of subagent objects in JSON format."""},
+                {"role": "user", "content": "Create a team of subagents to do task: {}"},
+            ],
+
+            response_format=response_format,
+        )
+
+
+
+        result = json.loads(completion.choices[0].message.content)
+
+        return result["subagents"]
+            
+        
+    async def prepare_realtimex_agent(self, agent_id, agent_data):
+        # directus_client = DirectusClient(server_url = directus_server_url,access_token = directus_access_token)
+
+        # d_agent = directus_client.get_directus_item_by_id("realtimex_agents",agent_id)
+
+        agent_name = agent_data["name"]
+        agent_description = agent_data["description"]
+        system_prompt = agent_data["instructions"]
+        agent_framework = agent_data["execution_config"]["framework"]
+        default_model = agent_data["execution_config"]["models"]["default_model"]
+
+        recommended_agent_flows = None
+        if "recommended_agent_flows" in agent_data:
+            recommended_agent_flows = agent_data["recommended_agent_flows"]
+
+        recommended_aci_mcp_apps = None
+        if "recommended_aci_mcp_apps" in agent_data:
+            recommended_aci_mcp_apps = agent_data["recommended_aci_mcp_apps"]
+
+        recommended_local_mcp_apps = None
+        if "recommended_local_mcp_apps" in agent_data:
+            recommended_local_mcp_apps = agent_data["recommended_local_mcp_apps"]
+            
+        recommended_team_members = None
+        if "recommended_team_members" in agent_data:
+            recommended_team_members = agent_data["recommended_team_members"]
+
+        self.agent_name = agent_name
+        self.system_prompt = system_prompt
+        self.agent_framework = agent_framework
+        self.agent_description = agent_description
+        self.default_model = default_model
+        self.agent_id = agent_id
+        self.agent_data = agent_data
+        self.recommended_agent_flows = recommended_agent_flows
+        self.recommended_aci_mcp_apps = recommended_aci_mcp_apps
+        self.recommended_local_mcp_apps = recommended_local_mcp_apps
+        self.recommended_team_members = recommended_team_members
+
+    async def prepare_memory(self,memory_id, memory_path, litellm_base_url, litellm_api_key):
+        pass
+        # config_dict = {
+        #     "version": "v1.1",
+        #     "vector_store": {
+        #         "provider": "chroma",
+        #         "config": {
+        #             "collection_name": memory_id,
+        #             "path": memory_path,
+        #         }
+        #     },
+        #     "llm": {"provider": "openai", "config": {"api_key": litellm_api_key, "openai_base_url": litellm_base_url, "temperature": 0.2, "model": "gpt-4o-mini"}},
+        #     "embedder": {"provider": "openai", "config": {"api_key": litellm_api_key, "openai_base_url": litellm_base_url, "model": "text-embedding-3-small"}},
+        #     "history_db_path": "",
+        # }
+        # print("config_dict",config_dict)
+        # memory = Memory.from_config(config_dict=config_dict)
+        # self.memory = memory
+
+    async def load_default_callbacks(self):
+        return [ShowToolCalling(),*get_default_callbacks()]
+
+    async def load_default_tools(self):
+        return []
+
+    async def load_aci_mcp_tools(self, linked_account_owner_id, aci_api_key):
+        from any_agent.config import MCPStdio
+        mcp_apps = []
+        for app in self.recommended_aci_mcp_apps:
+            if "realtimex_mcp_server_id" in app:
+                mcp_apps.append(app["realtimex_mcp_server_id"]["name"])
+            else:
+                mcp_apps.append(app["name"])
+        # mcp_apps = [app["realtimex_mcp_server_id"]["name"] for app in self.recommended_aci_mcp_apps]
+        if len(mcp_apps)>0:
+            mcp = MCPStdio(
+                # command=get_uvx_executable(),
+                # args=["aci-mcp@latest", 'apps-server', '--apps', ','.join(mcp_apps) , '--linked-account-owner-id', linked_account_owner_id],
+                command="aci-mcp",
+                args=['apps-server', '--apps', ','.join(mcp_apps) , '--linked-account-owner-id', linked_account_owner_id],
+                client_session_timeout_seconds=300,
+                env={
+                    "ACI_SERVER_URL":"https://mcp.realtimex.ai/v1/",
+                    "ACI_API_KEY":aci_api_key
+                }
+            )
+            return mcp
+        return None
+
+    async def load_local_mcp_tools(self, workspace_slug, thread_id):
+        from any_agent.config import MCPStdio
+        mcp_apps = [app["config"] for app in self.recommended_local_mcp_apps]
+        # mcp_apps = [
+        #     {"command":"uvx","args":["mcp-shell-server"],"env":{"ALLOW_COMMANDS":"ls,cat,pwd,grep,wc,touch,find"}}
+        # ]
+        mcps = []
+        default_env = {
+            'NVM_INC': get_nvm_inc(),
+            'NVM_CD_FLAGS': '-q',
+            'NVM_DIR': get_nvm_dir(),
+            'PATH': f'{os.environ.copy()["PATH"]}{os.pathsep}{get_nvm_bin()}',
+            'NVM_BIN': get_nvm_bin(),
+            "SESSION_ID": self.current_session_id,
+            "WORKSPACE_SLUG": workspace_slug,
+            "THREAD_ID": thread_id,
+            "AGENT_ID": self.agent_id,
+        }
+        for mcp_app in mcp_apps:
+            if mcp_app["command"] == "uvx":
+                mcp_app["command"] = get_uvx_executable()
+            if mcp_app["command"] == "npx":
+                mcp_app["command"] = get_npx_executable()
+            if "env" not in mcp_app:
+                mcp_app["env"] = {}
+            mcp_app["env"] = {**mcp_app["env"],**default_env}
+            # print("mcp_app",mcp_app)
+            mcp = MCPStdio(
+                **mcp_app,
+                client_session_timeout_seconds=300,
+            )
+            mcps.append(mcp)
+        return mcps
+
+    async def load_mcp_agent_flow_tools(self, linked_account_owner_id, aci_api_key, litellm_base_url, litellm_api_key, realtimex_access_token, workspace_slug, thread_id):
+        from any_agent.config import MCPStdio
+        agent_flow_ids = [flow["realtimex_agent_flows_id"]["id"] for flow in self.recommended_agent_flows]
+        if len(agent_flow_ids)>0:
+            mcp = MCPStdio(
+                # command=get_uvx_executable(),
+                # args=["--from", "git+https://oauth2:5yTHSE9k34jbWgzXmsxQ@rtgit.rta.vn/rtlab/rtwebteam/mcp-servers/realtimex-ai-agent-flows@main","agent-flows-mcp-server",'--flows',','.join(agent_flow_ids)],
+                command="agent-flows-mcp-server",
+                args=['--flows',','.join(agent_flow_ids)],
+                client_session_timeout_seconds=300,
+                env={
+                    "AGENT_FLOWS_API_KEY": realtimex_access_token,
+                    "LITELLM_API_KEY": litellm_api_key,
+                    "LITELLM_API_BASE": litellm_base_url,
+                    "MCP_ACI_API_KEY": aci_api_key,
+                    "MCP_ACI_LINKED_ACCOUNT_OWNER_ID": linked_account_owner_id,
+                    "SESSION_ID": self.current_session_id,
+                    "WORKSPACE_SLUG": workspace_slug,
+                    "THREAD_ID": thread_id,
+                    "AGENT_ID": self.agent_id,
+                    # "AGENT_FLOWS_BASE_URL": "https://your-custom-instance.com"  # Optional
+                }
+            )
+            return mcp
+        return None
+
+    async def load_a2a_agents_as_tools(self):
+        def get_free_port():
+            import socket
+            sock = socket.socket()
+            sock.bind(('', 0))
+            return sock.getsockname()[1]
+
+    
+        tools = []
+        for a2a_agent in self.recommended_team_members:
+            agent_id = a2a_agent["agent_id"]
+            agent_data = a2a_agent["agent_data"]
+            a2a_port = get_free_port()
+            # Create agent
+            agent = RealTimeXAgent()
+
+            await agent.load_default_agent(agent_id, agent_data, payload=a2a_agent)
+            
+            agent_server_url = await agent.serve_as_a2a(
+                a2a_serving_config={"port":a2a_port,"stream_tool_usage":True},
+            )
+
+            # print(agent_server_url)
+
+            tools.append(
+                await a2a_tool_async(
+                    agent_server_url,
+                    http_kwargs={
+                        "timeout": 300
+                    },  # This gives the time agent up to 30 seconds to respond to each request
+                )
+            )
+
+
+        return tools
+
+    async def load_knowledges(self,query, user_id, workspace_slug, thread_id, knowledges=["thread"]):
+        memory_session_id = None
+        if "user" in knowledges:
+            memory_session_id = None
+        elif "workspace" in knowledges:
+            memory_session_id = workspace_slug
+        elif "thread" in knowledges:
+            memory_session_id = f"{workspace_slug}_{thread_id}"
+        # print("memory_session_id",memory_session_id)
+        history_memories = self.memory.search(query=query, user_id=user_id, run_id=memory_session_id,limit=5)
+        # history_memories = self.memory.get_all(user_id=user_id, run_id=memory_session_id,limit=20)
+        # print("history_memories",history_memories)
+        
+        memories_str = "\n".join(f"- {entry['memory']}" for entry in history_memories["results"])
+
+        knowledges_str = ""
+        all_knowledge_memories = []
+        for knowledge_id in knowledges:
+            if knowledge_id in ["account","workspace","thread"]:
+                continue
+            knowledge_memories = self.memory.search(query=query, user_id=user_id, run_id=knowledge_id, limit=5)
+            all_knowledge_memories = [*all_knowledge_memories,*knowledge_memories["results"]]
+
+        knowledges_str = "\n".join(f"- {entry['memory']}" for entry in all_knowledge_memories)
+        return memories_str, knowledges_str
+
+
+    async def create_agent(self, agent_framework="tinyagent",agent_config=None,tools=[],callbacks=[], memories_str=None, knowledges_str=None):
+        default_agent_config = {
+            "name": self.agent_name,
+            "model_id": self.default_model,
+            "description": self.agent_description,
+            "instructions": self.system_prompt,
+            "tools": tools,
+            "callbacks": callbacks
+        }
+
+        default_agent_framework = self.agent_framework
+
+        if agent_config:
+            default_agent_config.update(agent_config)
+
+        if agent_framework:
+            default_agent_framework = agent_framework
+        
+        # print("default_agent_config", default_agent_config)
+        # print("agent_config", agent_config)
+
+        if memories_str:
+            default_agent_config["instructions"] = default_agent_config["instructions"].replace("##MEMORIES##",memories_str)
+        if knowledges_str:
+            default_agent_config["instructions"] = default_agent_config["instructions"].replace("##KNOWLEDGES##",knowledges_str)
+        # print(default_agent_framework)
+        # print(default_agent_config)
+
+        self.agent = await AnyAgent.create_async(
+            default_agent_framework,  # See all options in https://mozilla-ai.github.io/any-agent/
+            AgentConfig(
+                **default_agent_config,
+                current_session_id=self.current_session_id,
+                # agent_args={
+                #     "interrupt_after":["get_user_info"]
+                # }
+            ),
+        )
+        
+        return self.agent
+    
+    async def serve_as_a2a(self, a2a_serving_config):
+        handle = await self.agent.serve_async(A2AServingConfig(**a2a_serving_config))
+        server_port = handle.port
+        server_url = f"http://localhost:{server_port}"
+
+        return server_url
+
+    async def load_default_agent(self, agent_id, agent_data, payload):
+        system_prompt = None
+        agent_framework = None
+        agent_description = None
+        agent_name = None
+        default_model = None
+        provider_name = None
+        llm_setting = None
+
+        user_id = payload["user_id"]
+        workspace_slug = payload["workspace_slug"]
+        thread_id = payload["thread_id"]
+        knowledges = payload["knowledges"]
+        memory_id = payload["memory_id"]
+        memory_path = payload["memory_path"]
+        execution_id = payload["session_id"]
+        aci_linked_account_owner_id = payload["aci_linked_account_owner_id"]
+        aci_agent_first_api_key = payload["aci_api_key"]
+        realtimex_access_token = payload["realtimex_access_token"]
+
+        agent_description = agent_data["description"]
+        agent_name = agent_data["name"]
+        agent_framework = agent_data["execution_config"]["framework"]
+
+        if "agent_description" in payload:
+            agent_description = payload["agent_description"]
+        if "agent_name" in payload:
+            agent_name = payload["agent_name"]
+        if "agent_framework" in payload:
+            agent_framework = payload["agent_framework"]
+        if "system_prompt" in payload:
+            system_prompt = payload["system_prompt"]
+        if "llm_setting" in payload:
+            llm_setting = payload["llm_setting"]            
+            
+
+        default_openai_base_url = payload["litellm_api_base"]
+        default_openai_api_key = payload["litellm_api_key"]
+
+        # Load MCP tools
+        
+        # # Create agent
+        # agent = RealTimeXAgent()
+
+        # print("agent_data")
+
+        await self.prepare_realtimex_agent(
+            agent_id=agent_id,
+            agent_data=agent_data
+        )
+        
+        # await self.prepare_memory(memory_id=memory_id, memory_path=memory_path, litellm_base_url=default_openai_base_url, litellm_api_key=default_openai_api_key)
+
+        default_callbacks = await self.load_default_callbacks()
+        
+        default_tools = await self.load_default_tools()
+        all_tools = [*default_tools]
+        
+        if self.recommended_aci_mcp_apps:
+            aci_mcp = await self.load_aci_mcp_tools(
+                linked_account_owner_id=aci_linked_account_owner_id,
+                aci_api_key=aci_agent_first_api_key
+            )
+            if aci_mcp:
+                all_tools = [*all_tools,aci_mcp]
+
+        if self.recommended_local_mcp_apps:
+            local_mcps = await self.load_local_mcp_tools(
+                workspace_slug=workspace_slug,
+                thread_id=thread_id
+            )
+            # print("local_mcps",local_mcps)
+            all_tools = [*all_tools,*local_mcps]
+
+        if self.recommended_agent_flows:
+            aci_mcp_agent_flow = await self.load_mcp_agent_flow_tools(
+                linked_account_owner_id=aci_linked_account_owner_id,
+                aci_api_key=aci_agent_first_api_key,
+                realtimex_access_token=realtimex_access_token,
+                litellm_base_url=default_openai_base_url,
+                litellm_api_key=default_openai_api_key,
+                workspace_slug=workspace_slug,
+                thread_id=thread_id
+            )
+            if aci_mcp_agent_flow:
+                all_tools = [*all_tools,aci_mcp_agent_flow]
+
+        if self.recommended_team_members:
+            
+            team_members = await self.load_a2a_agents_as_tools()
+            # print(team_members)
+            if team_members:
+                all_tools = [*all_tools,*team_members]
+
+        
+
+        agent_config = {
+            "api_base": default_openai_base_url,
+            "api_key": default_openai_api_key,
+        }
+
+        # print("agent_framework",agent_framework)
+
+        if agent_description:
+            agent_config["description"] = agent_description
+        if system_prompt:
+            agent_config["instructions"] = system_prompt
+
+
+        if llm_setting:
+            llm_config = await self.prepare_llm(**llm_setting["default"])
+            # print("llm_config",llm_config)
+            agent_config.update(llm_config)
+
+
+        memories_str = ""
+        knowledges_str = ""
+        # if knowledges:
+        #     memories_str, knowledges_str = await self.load_knowledges(message, user_id, workspace_slug, thread_id, knowledges)
+
+        if agent_framework == "deepagents":
+            from deepagents import create_realtimex_deep_agent
+            from deepagents.backends import CompositeBackend
+            from deepagents.backends.filesystem import FilesystemBackend
+            agent_framework = "langchain"
+            agent_config["agent_type"] = create_realtimex_deep_agent
+
+            routes = {}
+            agent_config["agent_args"] = {"assistant_id": agent_id}
+
+            # Resolve memory paths (global and workspace)
+            global_agent_path = get_deepagents_agent_path("global", agent_id)
+            workspace_agent_path = get_deepagents_agent_path("workspace", agent_id, workspace_slug)
+
+            # Enable memory if either global or workspace config exists
+            agent_config["agent_args"]["enable_memory"] = bool(global_agent_path or workspace_agent_path)
+            if global_agent_path:
+                global_agent_root = os.path.dirname(global_agent_path)
+                routes[MEMORIES_GLOBAL_VIRTUAL_ROOT] = FilesystemBackend(
+                    root_dir=global_agent_root,
+                    virtual_mode=True,
+                )
+                agent_config["agent_args"]["global_agent_path"] = f"{MEMORIES_GLOBAL_VIRTUAL_ROOT}agent.md"
+            if workspace_agent_path:
+                workspace_agent_root = os.path.dirname(workspace_agent_path)
+                workspace_memories_root = f"{MEMORIES_WORKSPACE_VIRTUAL_ROOT_PREFIX}{workspace_slug}/"
+                routes[workspace_memories_root] = FilesystemBackend(
+                    root_dir=workspace_agent_root,
+                    virtual_mode=True,
+                )
+                agent_config["agent_args"]["workspace_agent_path"] = f"{workspace_memories_root}agent.md"
+
+            # Resolve skills paths (global and workspace)
+            global_skills_dir = get_deepagents_skills_dir("global", agent_id)
+            workspace_skills_dir = get_deepagents_skills_dir("workspace", agent_id, workspace_slug)
+
+            # Enable skills if either global or workspace config exists
+            agent_config["agent_args"]["enable_skills"] = bool(global_skills_dir or workspace_skills_dir)
+            if global_skills_dir:
+                routes[SKILLS_GLOBAL_VIRTUAL_ROOT] = FilesystemBackend(
+                    root_dir=global_skills_dir,
+                    virtual_mode=True,
+                )
+                agent_config["agent_args"]["global_skills_dir"] = SKILLS_GLOBAL_VIRTUAL_ROOT
+            if workspace_skills_dir:
+                workspace_skills_root = f"{SKILLS_WORKSPACE_VIRTUAL_ROOT_PREFIX}{workspace_slug}/"
+                routes[workspace_skills_root] = FilesystemBackend(
+                    root_dir=workspace_skills_dir,
+                    virtual_mode=True,
+                )
+                agent_config["agent_args"]["workspace_skills_dir"] = workspace_skills_root
+
+            # Backend is always required for deepagents
+            agent_config["agent_args"]["backend"] = lambda runtime: CompositeBackend(
+                default=FilesystemBackend(),
+                routes=routes,
+            )
+
+            if "subagents" in agent_data["execution_config"]:
+                if "_auto" in agent_data["execution_config"]["subagents"]:
+                    try:
+                        subagents = await self.create_subagents(
+                            instructions=system_prompt,
+                            tools=[],
+                            llm_config={
+                                "api_base": agent_config["api_base"],
+                                "api_key": agent_config["api_key"],
+                                "model_id": agent_config["model_id"]
+                            }
+                        )
+                        if subagents:
+                            # Merge subagents into existing agent_args if present
+                            if "agent_args" not in agent_config:
+                                agent_config["agent_args"] = {}
+                            agent_config["agent_args"]["subagents"] = subagents
+                    except Exception as e:
+                        print(e)
+                        
+                    # print(agent_config["agent_args"])
+
+        await self.create_agent(
+            agent_framework=agent_framework,  # See all options in https://mozilla-ai.github.io/any-agent/
+            agent_config=agent_config,
+            tools=all_tools,
+            callbacks=[*default_callbacks],
+            memories_str=memories_str,
+            knowledges_str=knowledges_str
+        )
+
+        # return agent