refactor(agent): Agent modular refactoring (eosphoros-ai#1487)

dbgpt/agent/__init__.py

@@ -1,6 +1,6 @@
 """DB-GPT Multi-Agents Module."""
 
-from .actions.action import Action, ActionOutput  # noqa: F401
+from .core.action import *  # noqa: F401, F403
 from .core.agent import (  # noqa: F401
     Agent,
     AgentContext,
@@ -13,12 +13,15 @@
     initialize_agent,
 )
 from .core.base_agent import ConversableAgent  # noqa: F401
-from .core.llm.llm import LLMConfig  # noqa: F401
+from .core.memory import *  # noqa: F401, F403
+from .core.memory.gpts.gpts_memory import GptsMemory  # noqa: F401
+from .core.plan import *  # noqa: F401, F403
+from .core.profile import *  # noqa: F401, F403
 from .core.schema import PluginStorageType  # noqa: F401
 from .core.user_proxy_agent import UserProxyAgent  # noqa: F401
-from .memory.gpts_memory import GptsMemory  # noqa: F401
 from .resource.resource_api import AgentResource, ResourceType  # noqa: F401
 from .resource.resource_loader import ResourceLoader  # noqa: F401
+from .util.llm.llm import LLMConfig  # noqa: F401
 
 __ALL__ = [
     "Agent",

dbgpt/agent/core/__init__.py

@@ -1 +1,22 @@
-"""Core Module for the Agent."""
+"""Core Module for the Agent.
+
+There are four modules in DB-GPT agent core according the paper
+`A survey on large language model based autonomous agents
+<https://link.springer.com/article/10.1007/s11704-024-40231-1>`
+by `Lei Wang, Chen Ma, Xueyang Feng, et al.`:
+
+1. Profiling Module: The profiling module aims to indicate the profiles of the agent
+roles.
+
+2. Memory Module: It stores information perceived from the environment and leverages
+the recorded memories to facilitate future actions.
+
+3. Planning Module: When faced with a complex task, humans tend to deconstruct it into
+simpler subtasks and solve them individually. The planning module aims to empower the
+agents with such human capability, which is expected to make the agent behave more
+reasonably, powerfully, and reliably
+
+4. Action Module: The action module is responsible for translating the agent’s
+decisions into specific outcomes. This module is located at the most downstream
+position and directly interacts with the environment.
+"""

dbgpt/agent/core/action/__init__.py

@@ -0,0 +1,20 @@
+"""Action Module.
+
+The action module is responsible for translating the agent’s decisions into specific
+outcomes. This module is located at the most downstream position and directly interacts
+with the environment. It is influenced by the profile, memory, and planning modules.
+
+
+The Goal Of The Action Module:
+--------
+1. Task Completion: Complete specific tasks, write a function in software development,
+and make an iron pick in the game.
+
+2. Communication: Communicate with other agents.
+
+3. Environment exploration: Explore unfamiliar environments to expand its perception
+and strike a balance between exploring and exploiting.
+"""
+
+from .base import Action, ActionOutput  # noqa: F401
+from .blank_action import BlankAction  # noqa: F401

dbgpt/agent/actions/action.py → dbgpt/agent/core/action/base.py

@@ -1,4 +1,5 @@
 """Base Action class for defining agent actions."""
+
 import json
 from abc import ABC, abstractmethod
 from typing import (
@@ -21,12 +22,13 @@
     field_description,
     model_fields,
     model_to_dict,
+    model_validator,
 )
 from dbgpt.util.json_utils import find_json_objects
+from dbgpt.vis.base import Vis
 
-from ...vis.base import Vis
-from ..resource.resource_api import AgentResource, ResourceType
-from ..resource.resource_loader import ResourceLoader
+from ...resource.resource_api import AgentResource, ResourceType
+from ...resource.resource_loader import ResourceLoader
 
 T = TypeVar("T", bound=Union[BaseModel, List[BaseModel], None])
 
@@ -41,6 +43,20 @@ class ActionOutput(BaseModel):
     view: Optional[str] = None
     resource_type: Optional[str] = None
     resource_value: Optional[Any] = None
+    action: Optional[str] = None
+    thoughts: Optional[str] = None
+    observations: Optional[str] = None
+
+    @model_validator(mode="before")
+    @classmethod
+    def pre_fill(cls, values: Any) -> Any:
+        """Pre-fill the values."""
+        if not isinstance(values, dict):
+            return values
+        is_exe_success = values.get("is_exe_success", True)
+        if not is_exe_success and "observations" not in values:
+            values["observations"] = values.get("content")
+        return values
 
     @classmethod
     def from_dict(

dbgpt/agent/actions/blank_action.py → dbgpt/agent/core/action/blank_action.py

@@ -3,8 +3,8 @@
 import logging
 from typing import Optional
 
-from ..resource.resource_api import AgentResource
-from .action import Action, ActionOutput
+from ...resource.resource_api import AgentResource
+from .base import Action, ActionOutput
 
 logger = logging.getLogger(__name__)
 
@@ -33,4 +33,8 @@ async def run(
 
         Just return the AI message.
         """
-        return ActionOutput(is_exe_success=True, content=ai_message, view=ai_message)
+        return ActionOutput(
+            is_exe_success=True,
+            content=ai_message,
+            view=ai_message,
+        )

dbgpt/agent/core/agent.py

@@ -9,9 +9,9 @@
 from dbgpt.core import LLMClient
 from dbgpt.util.annotations import PublicAPI
 
-from ..actions.action import ActionOutput
-from ..memory.gpts_memory import GptsMemory
 from ..resource.resource_loader import ResourceLoader
+from .action.base import ActionOutput
+from .memory.agent_memory import AgentMemory
 
 
 class Agent(ABC):
@@ -160,17 +160,20 @@ async def verify(
                 verification result.
         """
 
+    @property
     @abstractmethod
-    def get_name(self) -> str:
-        """Return name of the agent."""
+    def name(self) -> str:
+        """Return the name of the agent."""
 
+    @property
     @abstractmethod
-    def get_profile(self) -> str:
-        """Return profile of the agent."""
+    def role(self) -> str:
+        """Return the role of the agent."""
 
+    @property
     @abstractmethod
-    def get_describe(self) -> str:
-        """Return describe of the agent."""
+    def desc(self) -> Optional[str]:
+        """Return the description of the agent."""
 
 
 @dataclasses.dataclass
@@ -204,7 +207,7 @@ class AgentGenerateContext:
     rely_messages: List[AgentMessage] = dataclasses.field(default_factory=list)
     final: Optional[bool] = True
 
-    memory: Optional[GptsMemory] = None
+    memory: Optional[AgentMemory] = None
     agent_context: Optional[AgentContext] = None
     resource_loader: Optional[ResourceLoader] = None
     llm_client: Optional[LLMClient] = None
@@ -302,3 +305,9 @@ def copy(self) -> "AgentMessage":
             role=self.role,
             success=self.success,
         )
+
+    def get_dict_context(self) -> Dict[str, Any]:
+        """Return the context as a dictionary."""
+        if isinstance(self.context, dict):
+            return self.context
+        return {}

dbgpt/agent/core/agent_manage.py

@@ -18,7 +18,7 @@ def participant_roles(agents: List[Agent]) -> str:
     # Default to all agents registered
     roles = []
     for agent in agents:
-        roles.append(f"{agent.get_name()}: {agent.get_describe()}")
+        roles.append(f"{agent.name}: {agent.desc}")
     return "\n".join(roles)
 
 
@@ -34,13 +34,13 @@ def mentioned_agents(message_content: str, agents: List[Agent]) -> Dict:
     mentions = dict()
     for agent in agents:
         regex = (
-            r"(?<=\W)" + re.escape(agent.get_name()) + r"(?=\W)"
+            r"(?<=\W)" + re.escape(agent.name) + r"(?=\W)"
         )  # Finds agent mentions, taking word boundaries into account
         count = len(
             re.findall(regex, " " + message_content + " ")
         )  # Pad the message to help with matching
         if count > 0:
-            mentions[agent.get_name()] = count
+            mentions[agent.name] = count
     return mentions
 
 
@@ -84,7 +84,7 @@ def register_agent(
     ) -> str:
         """Register an agent."""
         inst = cls()
-        profile = inst.get_profile()
+        profile = inst.role
         if profile in self._agents and (
             profile in self._core_agents or not ignore_duplicate
         ):
@@ -110,13 +110,13 @@ def get_by_name(self, name: str) -> Type[ConversableAgent]:
 
     def get_describe_by_name(self, name: str) -> str:
         """Return the description of an agent by name."""
-        return self._agents[name][1].desc
+        return self._agents[name][1].desc or ""
 
     def all_agents(self) -> Dict[str, str]:
         """Return a dictionary of all registered agents and their descriptions."""
         result = {}
         for name, value in self._agents.items():
-            result[name] = value[1].desc
+            result[name] = value[1].desc or ""
         return result
 
     def list_agents(self):
@@ -125,7 +125,7 @@ def list_agents(self):
         for name, value in self._agents.items():
             result.append(
                 {
-                    "name": value[1].profile,
+                    "name": value[1].role,
                     "desc": value[1].goal,
                 }
             )