feat: add configurable remote sub agents

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -12,7 +12,7 @@ body:
     id: version
     attributes:
       label: What version of eigent are you using?
-      placeholder: E.g., 0.0.90
+      placeholder: E.g., 0.0.91
     validations:
       required: true
 

.github/workflows/build.yml

@@ -347,7 +347,7 @@ jobs:
           files: |
             gh-release-assets/*
 
-      # Extract version from tag (e.g., v0.0.85 -> 0.0.90)
+      # Extract version from tag (e.g., v0.0.85 -> 0.0.91)
       - name: Extract version
         if: startsWith(github.ref, 'refs/tags/')
         id: version

backend/app/agent/factory/browser.py

@@ -20,6 +20,9 @@
 from camel.toolkits import ToolkitMessageIntegration
 
 from app.agent.agent_model import agent_model
+from app.agent.factory.remote_sub_agent import (
+    attach_remote_sub_agent_if_enabled,
+)
 from app.agent.listen_chat_agent import logger
 from app.agent.prompt import BROWSER_SYS_PROMPT
 from app.agent.toolkit.human_toolkit import HumanToolkit
@@ -279,6 +282,15 @@ def browser_agent(options: Chat):
         *search_tools,
         *skill_toolkit.get_tools(),
     ]
+    tool_names = [
+        SearchToolkit.toolkit_name(),
+        HybridBrowserToolkit.toolkit_name(),
+        HumanToolkit.toolkit_name(),
+        NoteTakingToolkit.toolkit_name(),
+        TerminalToolkit.toolkit_name(),
+        ScreenshotToolkit.toolkit_name(),
+        SkillToolkit.toolkit_name(),
+    ]
 
     # Build external browser notice
     external_browser_notice = ""
@@ -300,6 +312,16 @@ def browser_agent(options: Chat):
         now_str=NOW_STR,
         external_browser_notice=external_browser_notice,
     )
+    system_message = attach_remote_sub_agent_if_enabled(
+        options=options,
+        agent_name=Agents.browser_agent,
+        working_directory=working_directory,
+        tools=tools,
+        tool_names=tool_names,
+        system_message=system_message,
+        local_tool_description="local browser, search, or terminal actions",
+        message_integration=message_integration,
+    )
 
     agent = agent_model(
         Agents.browser_agent,
@@ -310,15 +332,7 @@ def browser_agent(options: Chat):
         options,
         tools,
         prune_tool_calls_from_memory=True,
-        tool_names=[
-            SearchToolkit.toolkit_name(),
-            HybridBrowserToolkit.toolkit_name(),
-            HumanToolkit.toolkit_name(),
-            NoteTakingToolkit.toolkit_name(),
-            TerminalToolkit.toolkit_name(),
-            ScreenshotToolkit.toolkit_name(),
-            SkillToolkit.toolkit_name(),
-        ],
+        tool_names=tool_names,
         toolkits_to_register_agent=[
             web_toolkit_for_agent_registration,
             screenshot_toolkit_for_agent_registration,

backend/app/agent/factory/developer.py

@@ -18,6 +18,9 @@
 from camel.toolkits import ToolkitMessageIntegration
 
 from app.agent.agent_model import agent_model
+from app.agent.factory.remote_sub_agent import (
+    attach_remote_sub_agent_if_enabled,
+)
 from app.agent.listen_chat_agent import logger
 from app.agent.prompt import DEVELOPER_SYS_PROMPT
 from app.agent.toolkit.human_toolkit import HumanToolkit
@@ -103,12 +106,32 @@ async def developer_agent(options: Chat):
         *skill_toolkit.get_tools(),
         *search_tools,
     ]
+    tool_names = [
+        HumanToolkit.toolkit_name(),
+        TerminalToolkit.toolkit_name(),
+        NoteTakingToolkit.toolkit_name(),
+        WebDeployToolkit.toolkit_name(),
+        ScreenshotToolkit.toolkit_name(),
+        SkillToolkit.toolkit_name(),
+        SearchToolkit.toolkit_name(),
+    ]
+
     system_message = DEVELOPER_SYS_PROMPT.format(
         platform_system=platform.system(),
         platform_machine=platform.machine(),
         working_directory=working_directory,
         now_str=NOW_STR,
     )
+    system_message = attach_remote_sub_agent_if_enabled(
+        options=options,
+        agent_name=Agents.developer_agent,
+        working_directory=working_directory,
+        tools=tools,
+        tool_names=tool_names,
+        system_message=system_message,
+        local_tool_description="local `shell_exec` or local file writes",
+        message_integration=message_integration,
+    )
 
     return agent_model(
         Agents.developer_agent,
@@ -118,15 +141,7 @@ async def developer_agent(options: Chat):
         ),
         options,
         tools,
-        tool_names=[
-            HumanToolkit.toolkit_name(),
-            TerminalToolkit.toolkit_name(),
-            NoteTakingToolkit.toolkit_name(),
-            WebDeployToolkit.toolkit_name(),
-            ScreenshotToolkit.toolkit_name(),
-            SkillToolkit.toolkit_name(),
-            SearchToolkit.toolkit_name(),
-        ],
+        tool_names=tool_names,
         toolkits_to_register_agent=[
             screenshot_toolkit_for_agent_registration,
         ],

backend/app/agent/factory/document.py

@@ -17,6 +17,9 @@
 from camel.toolkits import ToolkitMessageIntegration
 
 from app.agent.agent_model import agent_model
+from app.agent.factory.remote_sub_agent import (
+    attach_remote_sub_agent_if_enabled,
+)
 from app.agent.listen_chat_agent import logger
 from app.agent.prompt import DOCUMENT_SYS_PROMPT
 from app.agent.toolkit.excel_toolkit import ExcelToolkit
@@ -126,12 +129,37 @@ async def document_agent(options: Chat):
         *skill_toolkit.get_tools(),
         *search_tools,
     ]
+    tool_names = [
+        FileToolkit.toolkit_name(),
+        PPTXToolkit.toolkit_name(),
+        HumanToolkit.toolkit_name(),
+        MarkItDownToolkit.toolkit_name(),
+        ExcelToolkit.toolkit_name(),
+        NoteTakingToolkit.toolkit_name(),
+        TerminalToolkit.toolkit_name(),
+        ScreenshotToolkit.toolkit_name(),
+        GoogleDriveMCPToolkit.toolkit_name(),
+        SkillToolkit.toolkit_name(),
+        SearchToolkit.toolkit_name(),
+    ]
     system_message = DOCUMENT_SYS_PROMPT.format(
         platform_system=platform.system(),
         platform_machine=platform.machine(),
         working_directory=working_directory,
         now_str=NOW_STR,
     )
+    system_message = attach_remote_sub_agent_if_enabled(
+        options=options,
+        agent_name=Agents.document_agent,
+        working_directory=working_directory,
+        tools=tools,
+        tool_names=tool_names,
+        system_message=system_message,
+        local_tool_description=(
+            "local document, file, terminal, or search tools"
+        ),
+        message_integration=message_integration,
+    )
 
     return agent_model(
         Agents.document_agent,
@@ -141,19 +169,7 @@ async def document_agent(options: Chat):
         ),
         options,
         tools,
-        tool_names=[
-            FileToolkit.toolkit_name(),
-            PPTXToolkit.toolkit_name(),
-            HumanToolkit.toolkit_name(),
-            MarkItDownToolkit.toolkit_name(),
-            ExcelToolkit.toolkit_name(),
-            NoteTakingToolkit.toolkit_name(),
-            TerminalToolkit.toolkit_name(),
-            ScreenshotToolkit.toolkit_name(),
-            GoogleDriveMCPToolkit.toolkit_name(),
-            SkillToolkit.toolkit_name(),
-            SearchToolkit.toolkit_name(),
-        ],
+        tool_names=tool_names,
         toolkits_to_register_agent=[
             screenshot_toolkit_for_agent_registration,
         ],

backend/app/agent/factory/mcp.py

@@ -16,25 +16,38 @@
 import uuid
 
 from camel.models import ModelFactory
+from camel.toolkits import ToolkitMessageIntegration
 from camel.types import ModelPlatformType
 
+from app.agent.factory.remote_sub_agent import (
+    attach_remote_sub_agent_if_enabled,
+)
 from app.agent.listen_chat_agent import ListenChatAgent, logger
 from app.agent.prompt import MCP_SYS_PROMPT
+from app.agent.toolkit.human_toolkit import HumanToolkit
 from app.agent.toolkit.mcp_search_toolkit import McpSearchToolkit
 from app.agent.tools import get_mcp_tools
 from app.model.chat import Chat
 from app.model.model_platform import patch_bedrock_cloud_config
 from app.service.task import ActionCreateAgentData, Agents, get_task_lock
+from app.utils.file_utils import get_working_directory
 
 
 async def mcp_agent(options: Chat):
+    working_directory = get_working_directory(options)
     logger.info(
         f"Creating MCP agent for project: {options.project_id} "
         f"with {len(options.installed_mcp['mcpServers'])} MCP servers"
     )
+    message_integration = ToolkitMessageIntegration(
+        message_handler=HumanToolkit(
+            options.project_id, Agents.mcp_agent
+        ).send_message_to_user
+    )
     tools = [
         *McpSearchToolkit(options.project_id).get_tools(),
     ]
+    tool_names = [McpSearchToolkit.toolkit_name()]
     if len(options.installed_mcp["mcpServers"]) > 0:
         try:
             mcp_tools = await get_mcp_tools(options.installed_mcp)
@@ -43,15 +56,16 @@ async def mcp_agent(options: Chat):
                 f"for task {options.project_id}"
             )
             if mcp_tools:
-                tool_names = [
+                mcp_tool_names = [
                     (
                         tool.get_function_name()
                         if hasattr(tool, "get_function_name")
                         else str(tool)
                     )
                     for tool in mcp_tools
                 ]
-                logger.debug(f"MCP tools: {tool_names}")
+                logger.debug(f"MCP tools: {mcp_tool_names}")
+                tool_names.extend(mcp_tool_names)
             tools = [*tools, *mcp_tools]
         except Exception as e:
             logger.debug(repr(e))
@@ -92,6 +106,17 @@ async def mcp_agent(options: Chat):
         extra_params = dict(extra_params)
         extra_params.setdefault("api_version", "2024-12-01-preview")
 
+    system_message = attach_remote_sub_agent_if_enabled(
+        options=options,
+        agent_name=Agents.mcp_agent,
+        working_directory=working_directory,
+        tools=tools,
+        tool_names=tool_names,
+        system_message=MCP_SYS_PROMPT,
+        local_tool_description="local MCP or search tools",
+        message_integration=message_integration,
+    )
+
     # Build model_config_dict with prompt caching
     model_config_dict = {}
     if options.is_cloud():
@@ -116,7 +141,7 @@ async def mcp_agent(options: Chat):
     return ListenChatAgent(
         options.project_id,
         Agents.mcp_agent,
-        system_message=MCP_SYS_PROMPT,
+        system_message=system_message,
         model=ModelFactory.create(
             model_platform=options.model_platform,
             model_type=options.model_type,

backend/app/agent/factory/multi_modal.py

@@ -19,6 +19,9 @@
 from camel.types import ModelPlatformType
 
 from app.agent.agent_model import agent_model
+from app.agent.factory.remote_sub_agent import (
+    attach_remote_sub_agent_if_enabled,
+)
 from app.agent.listen_chat_agent import logger
 from app.agent.prompt import MULTI_MODAL_SYS_PROMPT
 from app.agent.toolkit.audio_analysis_toolkit import AudioAnalysisToolkit
@@ -149,12 +152,35 @@ def multi_modal_agent(options: Chat):
         )
         tools.extend(audio_analysis_toolkit.get_tools())
 
+    tool_names = [
+        VideoDownloaderToolkit.toolkit_name(),
+        AudioAnalysisToolkit.toolkit_name(),
+        ScreenshotToolkit.toolkit_name(),
+        OpenAIImageToolkit.toolkit_name(),
+        HumanToolkit.toolkit_name(),
+        TerminalToolkit.toolkit_name(),
+        NoteTakingToolkit.toolkit_name(),
+        SearchToolkit.toolkit_name(),
+        SkillToolkit.toolkit_name(),
+    ]
     system_message = MULTI_MODAL_SYS_PROMPT.format(
         platform_system=platform.system(),
         platform_machine=platform.machine(),
         working_directory=working_directory,
         now_str=NOW_STR,
     )
+    system_message = attach_remote_sub_agent_if_enabled(
+        options=options,
+        agent_name=Agents.multi_modal_agent,
+        working_directory=working_directory,
+        tools=tools,
+        tool_names=tool_names,
+        system_message=system_message,
+        local_tool_description=(
+            "local media, terminal, file, or search tools"
+        ),
+        message_integration=message_integration,
+    )
 
     return agent_model(
         Agents.multi_modal_agent,
@@ -164,17 +190,7 @@ def multi_modal_agent(options: Chat):
         ),
         options,
         tools,
-        tool_names=[
-            VideoDownloaderToolkit.toolkit_name(),
-            AudioAnalysisToolkit.toolkit_name(),
-            ScreenshotToolkit.toolkit_name(),
-            OpenAIImageToolkit.toolkit_name(),
-            HumanToolkit.toolkit_name(),
-            TerminalToolkit.toolkit_name(),
-            NoteTakingToolkit.toolkit_name(),
-            SearchToolkit.toolkit_name(),
-            SkillToolkit.toolkit_name(),
-        ],
+        tool_names=tool_names,
         toolkits_to_register_agent=[
             screenshot_toolkit_for_agent_registration,
         ],