AgentOps Runtime Logging Implementation (#2682)

* add agentops req

* track conversable agents with agentops

* track tool usage

* track message sending

* remove record from parent

* remove record

* simple example

* notebook example

* remove spacing change

* optional dependency

* documentation

* remove extra import

* optional import

* record if agentops

* if agentops

* wrap function auto name

* install agentops before notebook test

* documentation fixes

* notebook metadata

* notebook metadata

* pre-commit hook changes

* doc link fixes

* git lfs

* autogen tag

* bump agentops version

* log tool events

* notebook fixes

* docs

* formatting

* Updated ecosystem manual

* Update notebook for clarity

* cleaned up notebook

* updated precommit recommendations

* Fixed links to screenshots and examples

* removed unused files

* changed notebook hyperlink

* update docusaurus link path

* reverted setup.py

* change setup again

* undo changes

* revert conversable agent

* removed file not in branch

* Updated notebook to look nicer

* change letter

* revert setup

* revert setup again

* change ref link

* change reflink

* remove optional dependency

* removed duplicated section

* Addressed clarity commetns from howard

* minor updates to wording

* formatting and pr fixes

* added info markdown cell

* better docs

* notebook

* observability docs

* pre-commit fixes

* example images in notebook

* example images in docs

* example images in docs

* delete agentops ong

* doc updates

* docs updates

* docs updates

* use agent as extra_kwarg

* add logging tests

* pass function properly

* create table

* dummy function name

* log chat completion source name

* safe serialize

* test fixes

* formatting

* type checks

---------

Co-authored-by: reibs <[email protected]>
Co-authored-by: Chi Wang <[email protected]>
Co-authored-by: Eric Zhu <[email protected]>
Co-authored-by: Howard Gil <[email protected]>
Co-authored-by: Alex Reibman <[email protected]>

Author: 5 people

autogen/agentchat/conversable_agent.py

@@ -31,7 +31,7 @@
 from ..function_utils import get_function_schema, load_basemodels_if_needed, serialize_to_str
 from ..io.base import IOStream
 from ..oai.client import ModelClient, OpenAIWrapper
-from ..runtime_logging import log_event, log_new_agent, logging_enabled
+from ..runtime_logging import log_event, log_function_use, log_new_agent, logging_enabled
 from .agent import Agent, LLMAgent
 from .chat import ChatResult, a_initiate_chats, initiate_chats
 from .utils import consolidate_chat_info, gather_usage_summary
@@ -1357,9 +1357,7 @@ def _generate_oai_reply_from_client(self, llm_client, messages, cache) -> Union[
 
         # TODO: #1143 handle token limit exceeded error
         response = llm_client.create(
-            context=messages[-1].pop("context", None),
-            messages=all_messages,
-            cache=cache,
+            context=messages[-1].pop("context", None), messages=all_messages, cache=cache, agent=self
         )
         extracted_response = llm_client.extract_text_or_completion_object(response)[0]
 
@@ -2528,13 +2526,14 @@ def _wrap_function(self, func: F) -> F:
         @functools.wraps(func)
         def _wrapped_func(*args, **kwargs):
             retval = func(*args, **kwargs)
-
+            log_function_use(self, func, kwargs, retval)
             return serialize_to_str(retval)
 
         @load_basemodels_if_needed
         @functools.wraps(func)
         async def _a_wrapped_func(*args, **kwargs):
             retval = await func(*args, **kwargs)
+            log_function_use(self, func, kwargs, retval)
             return serialize_to_str(retval)
 
         wrapped_func = _a_wrapped_func if inspect.iscoroutinefunction(func) else _wrapped_func

autogen/logger/base_logger.py

@@ -3,14 +3,15 @@
 import sqlite3
 import uuid
 from abc import ABC, abstractmethod
-from typing import TYPE_CHECKING, Any, Dict, List, Union
+from typing import TYPE_CHECKING, Any, Callable, Dict, List, TypeVar, Union
 
 from openai import AzureOpenAI, OpenAI
 from openai.types.chat import ChatCompletion
 
 if TYPE_CHECKING:
     from autogen import Agent, ConversableAgent, OpenAIWrapper
 
+F = TypeVar("F", bound=Callable[..., Any])
 ConfigItem = Dict[str, Union[str, List[str]]]
 LLMConfig = Dict[str, Union[None, float, int, ConfigItem, List[ConfigItem]]]
 
@@ -32,6 +33,7 @@ def log_chat_completion(
         invocation_id: uuid.UUID,
         client_id: int,
         wrapper_id: int,
+        source: Union[str, Agent],
         request: Dict[str, Union[float, str, List[Dict[str, str]]]],
         response: Union[str, ChatCompletion],
         is_cached: int,
@@ -49,9 +51,10 @@ def log_chat_completion(
             invocation_id (uuid):               A unique identifier for the invocation to the OpenAIWrapper.create method call
             client_id (int):                    A unique identifier for the underlying OpenAI client instance
             wrapper_id (int):                   A unique identifier for the OpenAIWrapper instance
-            request (dict):                     A dictionary representing the the request or call to the OpenAI client endpoint
+            source (str or Agent):              The source/creator of the event as a string name or an Agent instance
+            request (dict):                     A dictionary representing the request or call to the OpenAI client endpoint
             response (str or ChatCompletion):   The response from OpenAI
-            is_chached (int):                   1 if the response was a cache hit, 0 otherwise
+            is_cached (int):                    1 if the response was a cache hit, 0 otherwise
             cost(float):                        The cost for OpenAI response
             start_time (str):                   A string representing the moment the request was initiated
         """
@@ -104,6 +107,18 @@ def log_new_client(
         """
         ...
 
+    @abstractmethod
+    def log_function_use(self, source: Union[str, Agent], function: F, args: Dict[str, Any], returns: Any) -> None:
+        """
+        Log the use of a registered function (could be a tool)
+
+        Args:
+            source (str or Agent):      The source/creator of the event as a string name or an Agent instance
+            function (F):               The function information
+            args (dict):                The function args to log
+            returns (any):              The return
+        """
+
     @abstractmethod
     def stop(self) -> None:
         """

autogen/logger/file_logger.py

@@ -5,7 +5,7 @@
 import os
 import threading
 import uuid
-from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Union
+from typing import TYPE_CHECKING, Any, Callable, Dict, List, Optional, Tuple, TypeVar, Union
 
 from openai import AzureOpenAI, OpenAI
 from openai.types.chat import ChatCompletion
@@ -21,9 +21,21 @@
 
 logger = logging.getLogger(__name__)
 
+F = TypeVar("F", bound=Callable[..., Any])
+
 __all__ = ("FileLogger",)
 
 
+def safe_serialize(obj: Any) -> str:
+    def default(o: Any) -> str:
+        if hasattr(o, "to_json"):
+            return str(o.to_json())
+        else:
+            return f"<<non-serializable: {type(o).__qualname__}>>"
+
+    return json.dumps(obj, default=default)
+
+
 class FileLogger(BaseLogger):
     def __init__(self, config: Dict[str, Any]):
         self.config = config
@@ -59,6 +71,7 @@ def log_chat_completion(
         invocation_id: uuid.UUID,
         client_id: int,
         wrapper_id: int,
+        source: Union[str, Agent],
         request: Dict[str, Union[float, str, List[Dict[str, str]]]],
         response: Union[str, ChatCompletion],
         is_cached: int,
@@ -69,6 +82,11 @@ def log_chat_completion(
         Log a chat completion.
         """
         thread_id = threading.get_ident()
+        source_name = None
+        if isinstance(source, str):
+            source_name = source
+        else:
+            source_name = source.name
         try:
             log_data = json.dumps(
                 {
@@ -82,6 +100,7 @@ def log_chat_completion(
                     "start_time": start_time,
                     "end_time": get_current_ts(),
                     "thread_id": thread_id,
+                    "source_name": source_name,
                 }
             )
 
@@ -204,6 +223,29 @@ def log_new_client(
         except Exception as e:
             self.logger.error(f"[file_logger] Failed to log event {e}")
 
+    def log_function_use(self, source: Union[str, Agent], function: F, args: Dict[str, Any], returns: Any) -> None:
+        """
+        Log a registered function(can be a tool) use from an agent or a string source.
+        """
+        thread_id = threading.get_ident()
+
+        try:
+            log_data = json.dumps(
+                {
+                    "source_id": id(source),
+                    "source_name": str(source.name) if hasattr(source, "name") else source,
+                    "agent_module": source.__module__,
+                    "agent_class": source.__class__.__name__,
+                    "timestamp": get_current_ts(),
+                    "thread_id": thread_id,
+                    "input_args": safe_serialize(args),
+                    "returns": safe_serialize(returns),
+                }
+            )
+            self.logger.info(log_data)
+        except Exception as e:
+            self.logger.error(f"[file_logger] Failed to log event {e}")
+
     def get_connection(self) -> None:
         """Method is intentionally left blank because there is no specific connection needed for the FileLogger."""
         pass

autogen/logger/sqlite_logger.py

@@ -6,7 +6,7 @@
 import sqlite3
 import threading
 import uuid
-from typing import TYPE_CHECKING, Any, Dict, List, Tuple, Union
+from typing import TYPE_CHECKING, Any, Callable, Dict, List, Tuple, TypeVar, Union
 
 from openai import AzureOpenAI, OpenAI
 from openai.types.chat import ChatCompletion
@@ -25,6 +25,18 @@
 
 __all__ = ("SqliteLogger",)
 
+F = TypeVar("F", bound=Callable[..., Any])
+
+
+def safe_serialize(obj: Any) -> str:
+    def default(o: Any) -> str:
+        if hasattr(o, "to_json"):
+            return str(o.to_json())
+        else:
+            return f"<<non-serializable: {type(o).__qualname__}>>"
+
+    return json.dumps(obj, default=default)
+
 
 class SqliteLogger(BaseLogger):
     schema_version = 1
@@ -49,6 +61,7 @@ def start(self) -> str:
                     client_id INTEGER,
                     wrapper_id INTEGER,
                     session_id TEXT,
+                    source_name TEXT,
                     request TEXT,
                     response TEXT,
                     is_cached INEGER,
@@ -118,6 +131,18 @@ class TEXT,                                         -- type or class name of cli
             """
             self._run_query(query=query)
 
+            query = """
+                        CREATE TABLE IF NOT EXISTS function_calls (
+                            source_id INTEGER,
+                            source_name TEXT,
+                            function_name TEXT,
+                            args TEXT DEFAULT NULL,
+                            returns TEXT DEFAULT NULL,
+                            timestamp DATETIME DEFAULT CURRENT_TIMESTAMP
+                        );
+                        """
+            self._run_query(query=query)
+
             current_verion = self._get_current_db_version()
             if current_verion is None:
                 self._run_query(
@@ -192,6 +217,7 @@ def log_chat_completion(
         invocation_id: uuid.UUID,
         client_id: int,
         wrapper_id: int,
+        source: Union[str, Agent],
         request: Dict[str, Union[float, str, List[Dict[str, str]]]],
         response: Union[str, ChatCompletion],
         is_cached: int,
@@ -208,10 +234,16 @@ def log_chat_completion(
         else:
             response_messages = json.dumps(to_dict(response), indent=4)
 
+        source_name = None
+        if isinstance(source, str):
+            source_name = source
+        else:
+            source_name = source.name
+
         query = """
             INSERT INTO chat_completions (
-                invocation_id, client_id, wrapper_id, session_id, request, response, is_cached, cost, start_time, end_time
-            ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+                invocation_id, client_id, wrapper_id, session_id, request, response, is_cached, cost, start_time, end_time, source_name
+            ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
         """
         args = (
             invocation_id,
@@ -224,6 +256,7 @@ def log_chat_completion(
             cost,
             start_time,
             end_time,
+            source_name,
         )
 
         self._run_query(query=query, args=args)
@@ -335,6 +368,24 @@ def log_new_wrapper(self, wrapper: OpenAIWrapper, init_args: Dict[str, Union[LLM
         )
         self._run_query(query=query, args=args)
 
+    def log_function_use(self, source: Union[str, Agent], function: F, args: Dict[str, Any], returns: Any) -> None:
+
+        if self.con is None:
+            return
+
+        query = """
+        INSERT INTO function_calls (source_id, source_name, function_name, args, returns, timestamp) VALUES (?, ?, ?, ?, ?, ?)
+        """
+        query_args: Tuple[Any, ...] = (
+            id(source),
+            source.name if hasattr(source, "name") else source,
+            function.__name__,
+            safe_serialize(args),
+            safe_serialize(returns),
+            get_current_ts(),
+        )
+        self._run_query(query=query, args=query_args)
+
     def log_new_client(
         self, client: Union[AzureOpenAI, OpenAI, GeminiClient], wrapper: OpenAIWrapper, init_args: Dict[str, Any]
     ) -> None:

autogen/oai/client.py

@@ -319,6 +319,7 @@ class OpenAIWrapper:
     """A wrapper class for openai client."""
 
     extra_kwargs = {
+        "agent",
         "cache",
         "cache_seed",
         "filter_func",
@@ -542,6 +543,7 @@ def create(self, **config: Any) -> ModelClient.ModelClientResponseProtocol:
                 Note that the cache argument overrides the legacy cache_seed argument: if this argument is provided,
                 then the cache_seed argument is ignored. If this argument is not provided or None,
                 then the cache_seed argument is used.
+            - agent (AbstractAgent | None): The object responsible for creating a completion if an agent.
             - (Legacy) cache_seed (int | None) for using the DiskCache. Default to 41.
                 An integer cache_seed is useful when implementing "controlled randomness" for the completion.
                 None for no caching.
@@ -589,6 +591,7 @@ def yes_or_no_filter(context, response):
             cache = extra_kwargs.get("cache")
             filter_func = extra_kwargs.get("filter_func")
             context = extra_kwargs.get("context")
+            agent = extra_kwargs.get("agent")
 
             total_usage = None
             actual_usage = None
@@ -626,6 +629,7 @@ def yes_or_no_filter(context, response):
                                 invocation_id=invocation_id,
                                 client_id=id(client),
                                 wrapper_id=id(self),
+                                agent=agent,
                                 request=params,
                                 response=response,
                                 is_cached=1,
@@ -658,6 +662,7 @@ def yes_or_no_filter(context, response):
                         invocation_id=invocation_id,
                         client_id=id(client),
                         wrapper_id=id(self),
+                        agent=agent,
                         request=params,
                         response=f"error_code:{error_code}, config {i} failed",
                         is_cached=0,
@@ -688,6 +693,7 @@ def yes_or_no_filter(context, response):
                         invocation_id=invocation_id,
                         client_id=id(client),
                         wrapper_id=id(self),
+                        agent=agent,
                         request=params,
                         response=response,
                         is_cached=0,

autogen/runtime_logging.py

@@ -3,7 +3,7 @@
 import logging
 import sqlite3
 import uuid
-from typing import TYPE_CHECKING, Any, Dict, List, Literal, Optional, Union
+from typing import TYPE_CHECKING, Any, Callable, Dict, List, Literal, Optional, TypeVar, Union
 
 from openai import AzureOpenAI, OpenAI
 from openai.types.chat import ChatCompletion
@@ -20,6 +20,8 @@
 autogen_logger = None
 is_logging = False
 
+F = TypeVar("F", bound=Callable[..., Any])
+
 
 def start(
     logger: Optional[BaseLogger] = None,
@@ -56,6 +58,7 @@ def log_chat_completion(
     invocation_id: uuid.UUID,
     client_id: int,
     wrapper_id: int,
+    agent: Union[str, Agent],
     request: Dict[str, Union[float, str, List[Dict[str, str]]]],
     response: Union[str, ChatCompletion],
     is_cached: int,
@@ -67,7 +70,7 @@ def log_chat_completion(
         return
 
     autogen_logger.log_chat_completion(
-        invocation_id, client_id, wrapper_id, request, response, is_cached, cost, start_time
+        invocation_id, client_id, wrapper_id, agent, request, response, is_cached, cost, start_time
     )
 
 
@@ -87,6 +90,14 @@ def log_event(source: Union[str, Agent], name: str, **kwargs: Dict[str, Any]) ->
     autogen_logger.log_event(source, name, **kwargs)
 
 
+def log_function_use(agent: Union[str, Agent], function: F, args: Dict[str, Any], returns: any):
+    if autogen_logger is None:
+        logger.error("[runtime logging] log_function_use: autogen logger is None")
+        return
+
+    autogen_logger.log_function_use(agent, function, args, returns)
+
+
 def log_new_wrapper(wrapper: OpenAIWrapper, init_args: Dict[str, Union[LLMConfig, List[LLMConfig]]]) -> None:
     if autogen_logger is None:
         logger.error("[runtime logging] log_new_wrapper: autogen logger is None")