1010 lines
48 KiB
Python
1010 lines
48 KiB
Python
import datetime
|
|
import inspect
|
|
import json
|
|
import traceback
|
|
import uuid
|
|
from typing import List, Optional, Tuple, Union, cast
|
|
|
|
from tqdm import tqdm
|
|
|
|
from memgpt.agent_store.storage import StorageConnector
|
|
from memgpt.constants import (
|
|
CLI_WARNING_PREFIX,
|
|
FIRST_MESSAGE_ATTEMPTS,
|
|
JSON_ENSURE_ASCII,
|
|
JSON_LOADS_STRICT,
|
|
LLM_MAX_TOKENS,
|
|
MESSAGE_SUMMARY_TRUNC_KEEP_N_LAST,
|
|
MESSAGE_SUMMARY_TRUNC_TOKEN_FRAC,
|
|
MESSAGE_SUMMARY_WARNING_FRAC,
|
|
)
|
|
from memgpt.data_types import AgentState, EmbeddingConfig, Message, Passage
|
|
from memgpt.interface import AgentInterface
|
|
from memgpt.llm_api.llm_api_tools import create, is_context_overflow_error
|
|
from memgpt.memory import ArchivalMemory, BaseMemory, RecallMemory, summarize_messages
|
|
from memgpt.metadata import MetadataStore
|
|
from memgpt.models import chat_completion_response
|
|
from memgpt.models.pydantic_models import ToolModel
|
|
from memgpt.persistence_manager import LocalStateManager
|
|
from memgpt.system import (
|
|
get_initial_boot_messages,
|
|
get_login_event,
|
|
package_function_response,
|
|
package_summarize_message,
|
|
)
|
|
from memgpt.utils import (
|
|
count_tokens,
|
|
create_uuid_from_string,
|
|
get_local_time,
|
|
get_tool_call_id,
|
|
get_utc_time,
|
|
is_utc_datetime,
|
|
parse_json,
|
|
printd,
|
|
united_diff,
|
|
validate_function_response,
|
|
verify_first_message_correctness,
|
|
)
|
|
|
|
from .errors import LLMError
|
|
|
|
|
|
def construct_system_with_memory(
|
|
system: str,
|
|
memory: BaseMemory,
|
|
memory_edit_timestamp: str,
|
|
archival_memory: Optional[ArchivalMemory] = None,
|
|
recall_memory: Optional[RecallMemory] = None,
|
|
include_char_count: bool = True,
|
|
):
|
|
# TODO: modify this to be generalized
|
|
full_system_message = "\n".join(
|
|
[
|
|
system,
|
|
"\n",
|
|
f"### Memory [last modified: {memory_edit_timestamp.strip()}]",
|
|
f"{len(recall_memory) if recall_memory else 0} previous messages between you and the user are stored in recall memory (use functions to access them)",
|
|
f"{len(archival_memory) if archival_memory else 0} total memories you created are stored in archival memory (use functions to access them)",
|
|
"\nCore memory shown below (limited in size, additional information stored in archival / recall memory):",
|
|
str(memory),
|
|
# f'<persona characters="{len(memory.persona)}/{memory.persona_char_limit}">' if include_char_count else "<persona>",
|
|
# memory.persona,
|
|
# "</persona>",
|
|
# f'<human characters="{len(memory.human)}/{memory.human_char_limit}">' if include_char_count else "<human>",
|
|
# memory.human,
|
|
# "</human>",
|
|
]
|
|
)
|
|
return full_system_message
|
|
|
|
|
|
def initialize_message_sequence(
|
|
model: str,
|
|
system: str,
|
|
memory: BaseMemory,
|
|
archival_memory: Optional[ArchivalMemory] = None,
|
|
recall_memory: Optional[RecallMemory] = None,
|
|
memory_edit_timestamp: Optional[str] = None,
|
|
include_initial_boot_message: bool = True,
|
|
) -> List[dict]:
|
|
if memory_edit_timestamp is None:
|
|
memory_edit_timestamp = get_local_time()
|
|
|
|
full_system_message = construct_system_with_memory(
|
|
system, memory, memory_edit_timestamp, archival_memory=archival_memory, recall_memory=recall_memory
|
|
)
|
|
first_user_message = get_login_event() # event letting MemGPT know the user just logged in
|
|
|
|
if include_initial_boot_message:
|
|
if model is not None and "gpt-3.5" in model:
|
|
initial_boot_messages = get_initial_boot_messages("startup_with_send_message_gpt35")
|
|
else:
|
|
initial_boot_messages = get_initial_boot_messages("startup_with_send_message")
|
|
messages = (
|
|
[
|
|
{"role": "system", "content": full_system_message},
|
|
]
|
|
+ initial_boot_messages
|
|
+ [
|
|
{"role": "user", "content": first_user_message},
|
|
]
|
|
)
|
|
|
|
else:
|
|
messages = [
|
|
{"role": "system", "content": full_system_message},
|
|
{"role": "user", "content": first_user_message},
|
|
]
|
|
|
|
return messages
|
|
|
|
|
|
class Agent(object):
|
|
def __init__(
|
|
self,
|
|
interface: AgentInterface,
|
|
# agents can be created from providing agent_state
|
|
agent_state: AgentState,
|
|
tools: List[ToolModel],
|
|
# memory: BaseMemory,
|
|
# extras
|
|
messages_total: Optional[int] = None, # TODO remove?
|
|
first_message_verify_mono: bool = True, # TODO move to config?
|
|
):
|
|
# tools
|
|
for tool in tools:
|
|
assert tool, f"Tool is None - must be error in querying tool from DB"
|
|
assert tool.name in agent_state.tools, f"Tool {tool} not found in agent_state.tools"
|
|
for tool_name in agent_state.tools:
|
|
assert tool_name in [tool.name for tool in tools], f"Tool name {tool_name} not included in agent tool list"
|
|
# Store the functions schemas (this is passed as an argument to ChatCompletion)
|
|
self.functions = []
|
|
self.functions_python = {}
|
|
env = {}
|
|
env.update(globals())
|
|
for tool in tools:
|
|
# WARNING: name may not be consistent?
|
|
if tool.module: # execute the whole module
|
|
exec(tool.module, env)
|
|
else:
|
|
exec(tool.source_code, env)
|
|
self.functions_python[tool.name] = env[tool.name]
|
|
self.functions.append(tool.json_schema)
|
|
assert all([callable(f) for k, f in self.functions_python.items()]), self.functions_python
|
|
|
|
# Hold a copy of the state that was used to init the agent
|
|
self.agent_state = agent_state
|
|
|
|
# gpt-4, gpt-3.5-turbo, ...
|
|
self.model = self.agent_state.llm_config.model
|
|
|
|
# Store the system instructions (used to rebuild memory)
|
|
self.system = self.agent_state.system
|
|
|
|
# Initialize the memory object
|
|
self.memory = BaseMemory.load(self.agent_state.state["memory"])
|
|
printd("Initialized memory object", self.memory)
|
|
|
|
# Interface must implement:
|
|
# - internal_monologue
|
|
# - assistant_message
|
|
# - function_message
|
|
# ...
|
|
# Different interfaces can handle events differently
|
|
# e.g., print in CLI vs send a discord message with a discord bot
|
|
self.interface = interface
|
|
|
|
# Create the persistence manager object based on the AgentState info
|
|
self.persistence_manager = LocalStateManager(agent_state=self.agent_state)
|
|
|
|
# State needed for heartbeat pausing
|
|
self.pause_heartbeats_start = None
|
|
self.pause_heartbeats_minutes = 0
|
|
|
|
self.first_message_verify_mono = first_message_verify_mono
|
|
|
|
# Controls if the convo memory pressure warning is triggered
|
|
# When an alert is sent in the message queue, set this to True (to avoid repeat alerts)
|
|
# When the summarizer is run, set this back to False (to reset)
|
|
self.agent_alerted_about_memory_pressure = False
|
|
|
|
self._messages: List[Message] = []
|
|
|
|
# Once the memory object is initialized, use it to "bake" the system message
|
|
if "messages" in self.agent_state.state and self.agent_state.state["messages"] is not None:
|
|
# print(f"Agent.__init__ :: loading, state={agent_state.state['messages']}")
|
|
if not isinstance(self.agent_state.state["messages"], list):
|
|
raise ValueError(f"'messages' in AgentState was bad type: {type(self.agent_state.state['messages'])}")
|
|
assert all([isinstance(msg, str) for msg in self.agent_state.state["messages"]])
|
|
|
|
# Convert to IDs, and pull from the database
|
|
raw_messages = [
|
|
self.persistence_manager.recall_memory.storage.get(id=uuid.UUID(msg_id)) for msg_id in self.agent_state.state["messages"]
|
|
]
|
|
assert all([isinstance(msg, Message) for msg in raw_messages]), (raw_messages, self.agent_state.state["messages"])
|
|
self._messages.extend([cast(Message, msg) for msg in raw_messages if msg is not None])
|
|
|
|
for m in self._messages:
|
|
# assert is_utc_datetime(m.created_at), f"created_at on message for agent {self.agent_state.name} isn't UTC:\n{vars(m)}"
|
|
# TODO eventually do casting via an edit_message function
|
|
if not is_utc_datetime(m.created_at):
|
|
printd(f"Warning - created_at on message for agent {self.agent_state.name} isn't UTC (text='{m.text}')")
|
|
m.created_at = m.created_at.replace(tzinfo=datetime.timezone.utc)
|
|
|
|
else:
|
|
printd(f"Agent.__init__ :: creating, state={agent_state.state['messages']}")
|
|
init_messages = initialize_message_sequence(
|
|
self.model,
|
|
self.system,
|
|
self.memory,
|
|
)
|
|
init_messages_objs = []
|
|
for msg in init_messages:
|
|
init_messages_objs.append(
|
|
Message.dict_to_message(
|
|
agent_id=self.agent_state.id, user_id=self.agent_state.user_id, model=self.model, openai_message_dict=msg
|
|
)
|
|
)
|
|
assert all([isinstance(msg, Message) for msg in init_messages_objs]), (init_messages_objs, init_messages)
|
|
self.messages_total = 0
|
|
self._append_to_messages(added_messages=[cast(Message, msg) for msg in init_messages_objs if msg is not None])
|
|
|
|
for m in self._messages:
|
|
assert is_utc_datetime(m.created_at), f"created_at on message for agent {self.agent_state.name} isn't UTC:\n{vars(m)}"
|
|
# TODO eventually do casting via an edit_message function
|
|
if not is_utc_datetime(m.created_at):
|
|
printd(f"Warning - created_at on message for agent {self.agent_state.name} isn't UTC (text='{m.text}')")
|
|
m.created_at = m.created_at.replace(tzinfo=datetime.timezone.utc)
|
|
|
|
# Keep track of the total number of messages throughout all time
|
|
self.messages_total = messages_total if messages_total is not None else (len(self._messages) - 1) # (-system)
|
|
self.messages_total_init = len(self._messages) - 1
|
|
printd(f"Agent initialized, self.messages_total={self.messages_total}")
|
|
|
|
# Create the agent in the DB
|
|
self.update_state()
|
|
|
|
@property
|
|
def messages(self) -> List[dict]:
|
|
"""Getter method that converts the internal Message list into OpenAI-style dicts"""
|
|
return [msg.to_openai_dict() for msg in self._messages]
|
|
|
|
@messages.setter
|
|
def messages(self, value):
|
|
raise Exception("Modifying message list directly not allowed")
|
|
|
|
def _trim_messages(self, num):
|
|
"""Trim messages from the front, not including the system message"""
|
|
self.persistence_manager.trim_messages(num)
|
|
|
|
new_messages = [self._messages[0]] + self._messages[num:]
|
|
self._messages = new_messages
|
|
|
|
def _prepend_to_messages(self, added_messages: List[Message]):
|
|
"""Wrapper around self.messages.prepend to allow additional calls to a state/persistence manager"""
|
|
assert all([isinstance(msg, Message) for msg in added_messages])
|
|
|
|
self.persistence_manager.prepend_to_messages(added_messages)
|
|
|
|
new_messages = [self._messages[0]] + added_messages + self._messages[1:] # prepend (no system)
|
|
self._messages = new_messages
|
|
self.messages_total += len(added_messages) # still should increment the message counter (summaries are additions too)
|
|
|
|
def _append_to_messages(self, added_messages: List[Message]):
|
|
"""Wrapper around self.messages.append to allow additional calls to a state/persistence manager"""
|
|
assert all([isinstance(msg, Message) for msg in added_messages])
|
|
|
|
self.persistence_manager.append_to_messages(added_messages)
|
|
|
|
# strip extra metadata if it exists
|
|
# for msg in added_messages:
|
|
# msg.pop("api_response", None)
|
|
# msg.pop("api_args", None)
|
|
new_messages = self._messages + added_messages # append
|
|
|
|
self._messages = new_messages
|
|
self.messages_total += len(added_messages)
|
|
|
|
def append_to_messages(self, added_messages: List[dict]):
|
|
"""An external-facing message append, where dict-like messages are first converted to Message objects"""
|
|
added_messages_objs = [
|
|
Message.dict_to_message(
|
|
agent_id=self.agent_state.id,
|
|
user_id=self.agent_state.user_id,
|
|
model=self.model,
|
|
openai_message_dict=msg,
|
|
)
|
|
for msg in added_messages
|
|
]
|
|
self._append_to_messages(added_messages_objs)
|
|
|
|
def _swap_system_message(self, new_system_message: Message):
|
|
assert isinstance(new_system_message, Message)
|
|
assert new_system_message.role == "system", new_system_message
|
|
assert self._messages[0].role == "system", self._messages
|
|
|
|
self.persistence_manager.swap_system_message(new_system_message)
|
|
|
|
new_messages = [new_system_message] + self._messages[1:] # swap index 0 (system)
|
|
self._messages = new_messages
|
|
|
|
def _get_ai_reply(
|
|
self,
|
|
message_sequence: List[Message],
|
|
function_call: str = "auto",
|
|
first_message: bool = False, # hint
|
|
stream: bool = False, # TODO move to config?
|
|
) -> chat_completion_response.ChatCompletionResponse:
|
|
"""Get response from LLM API"""
|
|
try:
|
|
response = create(
|
|
# agent_state=self.agent_state,
|
|
llm_config=self.agent_state.llm_config,
|
|
user_id=self.agent_state.user_id,
|
|
messages=message_sequence,
|
|
functions=self.functions,
|
|
functions_python=self.functions_python,
|
|
function_call=function_call,
|
|
# hint
|
|
first_message=first_message,
|
|
# streaming
|
|
stream=stream,
|
|
stream_inferface=self.interface,
|
|
)
|
|
# special case for 'length'
|
|
if response.choices[0].finish_reason == "length":
|
|
raise Exception("Finish reason was length (maximum context length)")
|
|
|
|
# catches for soft errors
|
|
if response.choices[0].finish_reason not in ["stop", "function_call", "tool_calls"]:
|
|
raise Exception(f"API call finish with bad finish reason: {response}")
|
|
|
|
# unpack with response.choices[0].message.content
|
|
return response
|
|
except Exception as e:
|
|
raise e
|
|
|
|
def _handle_ai_response(
|
|
self, response_message: chat_completion_response.Message, override_tool_call_id: bool = True
|
|
) -> Tuple[List[Message], bool, bool]:
|
|
"""Handles parsing and function execution"""
|
|
|
|
messages = [] # append these to the history when done
|
|
|
|
# Step 2: check if LLM wanted to call a function
|
|
if response_message.function_call or (response_message.tool_calls is not None and len(response_message.tool_calls) > 0):
|
|
if response_message.function_call:
|
|
raise DeprecationWarning(response_message)
|
|
if response_message.tool_calls is not None and len(response_message.tool_calls) > 1:
|
|
# raise NotImplementedError(f">1 tool call not supported")
|
|
# TODO eventually support sequential tool calling
|
|
printd(f">1 tool call not supported, using index=0 only\n{response_message.tool_calls}")
|
|
response_message.tool_calls = [response_message.tool_calls[0]]
|
|
assert response_message.tool_calls is not None and len(response_message.tool_calls) > 0
|
|
|
|
# generate UUID for tool call
|
|
if override_tool_call_id or response_message.function_call:
|
|
tool_call_id = get_tool_call_id() # needs to be a string for JSON
|
|
response_message.tool_calls[0].id = tool_call_id
|
|
else:
|
|
tool_call_id = response_message.tool_calls[0].id
|
|
assert tool_call_id is not None # should be defined
|
|
|
|
# only necessary to add the tool_cal_id to a function call (antipattern)
|
|
# response_message_dict = response_message.model_dump()
|
|
# response_message_dict["tool_call_id"] = tool_call_id
|
|
|
|
# role: assistant (requesting tool call, set tool call ID)
|
|
messages.append(
|
|
# NOTE: we're recreating the message here
|
|
# TODO should probably just overwrite the fields?
|
|
Message.dict_to_message(
|
|
agent_id=self.agent_state.id,
|
|
user_id=self.agent_state.user_id,
|
|
model=self.model,
|
|
openai_message_dict=response_message.model_dump(),
|
|
)
|
|
) # extend conversation with assistant's reply
|
|
printd(f"Function call message: {messages[-1]}")
|
|
|
|
# The content if then internal monologue, not chat
|
|
self.interface.internal_monologue(response_message.content, msg_obj=messages[-1])
|
|
|
|
# Step 3: call the function
|
|
# Note: the JSON response may not always be valid; be sure to handle errors
|
|
|
|
# Failure case 1: function name is wrong
|
|
function_call = (
|
|
response_message.function_call if response_message.function_call is not None else response_message.tool_calls[0].function
|
|
)
|
|
function_name = function_call.name
|
|
printd(f"Request to call function {function_name} with tool_call_id: {tool_call_id}")
|
|
try:
|
|
function_to_call = self.functions_python[function_name]
|
|
except KeyError as e:
|
|
error_msg = f"No function named {function_name}"
|
|
function_response = package_function_response(False, error_msg)
|
|
messages.append(
|
|
Message.dict_to_message(
|
|
agent_id=self.agent_state.id,
|
|
user_id=self.agent_state.user_id,
|
|
model=self.model,
|
|
openai_message_dict={
|
|
"role": "tool",
|
|
"name": function_name,
|
|
"content": function_response,
|
|
"tool_call_id": tool_call_id,
|
|
},
|
|
)
|
|
) # extend conversation with function response
|
|
self.interface.function_message(f"Error: {error_msg}", msg_obj=messages[-1])
|
|
return messages, False, True # force a heartbeat to allow agent to handle error
|
|
|
|
# Failure case 2: function name is OK, but function args are bad JSON
|
|
try:
|
|
raw_function_args = function_call.arguments
|
|
function_args = parse_json(raw_function_args)
|
|
except Exception as e:
|
|
error_msg = f"Error parsing JSON for function '{function_name}' arguments: {function_call.arguments}"
|
|
function_response = package_function_response(False, error_msg)
|
|
messages.append(
|
|
Message.dict_to_message(
|
|
agent_id=self.agent_state.id,
|
|
user_id=self.agent_state.user_id,
|
|
model=self.model,
|
|
openai_message_dict={
|
|
"role": "tool",
|
|
"name": function_name,
|
|
"content": function_response,
|
|
"tool_call_id": tool_call_id,
|
|
},
|
|
)
|
|
) # extend conversation with function response
|
|
self.interface.function_message(f"Error: {error_msg}", msg_obj=messages[-1])
|
|
return messages, False, True # force a heartbeat to allow agent to handle error
|
|
|
|
# (Still parsing function args)
|
|
# Handle requests for immediate heartbeat
|
|
heartbeat_request = function_args.pop("request_heartbeat", None)
|
|
if not (isinstance(heartbeat_request, bool) or heartbeat_request is None):
|
|
printd(
|
|
f"{CLI_WARNING_PREFIX}'request_heartbeat' arg parsed was not a bool or None, type={type(heartbeat_request)}, value={heartbeat_request}"
|
|
)
|
|
heartbeat_request = False
|
|
|
|
# Failure case 3: function failed during execution
|
|
# NOTE: the msg_obj associated with the "Running " message is the prior assistant message, not the function/tool role message
|
|
# this is because the function/tool role message is only created once the function/tool has executed/returned
|
|
self.interface.function_message(f"Running {function_name}({function_args})", msg_obj=messages[-1])
|
|
try:
|
|
spec = inspect.getfullargspec(function_to_call).annotations
|
|
|
|
for name, arg in function_args.items():
|
|
if isinstance(function_args[name], dict):
|
|
function_args[name] = spec[name](**function_args[name])
|
|
|
|
function_args["self"] = self # need to attach self to arg since it's dynamically linked
|
|
|
|
function_response = function_to_call(**function_args)
|
|
if function_name in ["conversation_search", "conversation_search_date", "archival_memory_search"]:
|
|
# with certain functions we rely on the paging mechanism to handle overflow
|
|
truncate = False
|
|
else:
|
|
# but by default, we add a truncation safeguard to prevent bad functions from
|
|
# overflow the agent context window
|
|
truncate = True
|
|
function_response_string = validate_function_response(function_response, truncate=truncate)
|
|
function_args.pop("self", None)
|
|
function_response = package_function_response(True, function_response_string)
|
|
function_failed = False
|
|
except Exception as e:
|
|
function_args.pop("self", None)
|
|
# error_msg = f"Error calling function {function_name} with args {function_args}: {str(e)}"
|
|
# Less detailed - don't provide full args, idea is that it should be in recent context so no need (just adds noise)
|
|
error_msg = f"Error calling function {function_name}: {str(e)}"
|
|
error_msg_user = f"{error_msg}\n{traceback.format_exc()}"
|
|
printd(error_msg_user)
|
|
function_response = package_function_response(False, error_msg)
|
|
messages.append(
|
|
Message.dict_to_message(
|
|
agent_id=self.agent_state.id,
|
|
user_id=self.agent_state.user_id,
|
|
model=self.model,
|
|
openai_message_dict={
|
|
"role": "tool",
|
|
"name": function_name,
|
|
"content": function_response,
|
|
"tool_call_id": tool_call_id,
|
|
},
|
|
)
|
|
) # extend conversation with function response
|
|
self.interface.function_message(f"Ran {function_name}({function_args})", msg_obj=messages[-1])
|
|
self.interface.function_message(f"Error: {error_msg}", msg_obj=messages[-1])
|
|
return messages, False, True # force a heartbeat to allow agent to handle error
|
|
|
|
# If no failures happened along the way: ...
|
|
# Step 4: send the info on the function call and function response to GPT
|
|
messages.append(
|
|
Message.dict_to_message(
|
|
agent_id=self.agent_state.id,
|
|
user_id=self.agent_state.user_id,
|
|
model=self.model,
|
|
openai_message_dict={
|
|
"role": "tool",
|
|
"name": function_name,
|
|
"content": function_response,
|
|
"tool_call_id": tool_call_id,
|
|
},
|
|
)
|
|
) # extend conversation with function response
|
|
self.interface.function_message(f"Ran {function_name}({function_args})", msg_obj=messages[-1])
|
|
self.interface.function_message(f"Success: {function_response_string}", msg_obj=messages[-1])
|
|
|
|
else:
|
|
# Standard non-function reply
|
|
messages.append(
|
|
Message.dict_to_message(
|
|
agent_id=self.agent_state.id,
|
|
user_id=self.agent_state.user_id,
|
|
model=self.model,
|
|
openai_message_dict=response_message.model_dump(),
|
|
)
|
|
) # extend conversation with assistant's reply
|
|
self.interface.internal_monologue(response_message.content, msg_obj=messages[-1])
|
|
heartbeat_request = False
|
|
function_failed = False
|
|
|
|
# rebuild memory
|
|
# TODO: @charles please check this
|
|
self.rebuild_memory()
|
|
|
|
return messages, heartbeat_request, function_failed
|
|
|
|
def step(
|
|
self,
|
|
user_message: Union[Message, str], # NOTE: should be json.dump(dict)
|
|
first_message: bool = False,
|
|
first_message_retry_limit: int = FIRST_MESSAGE_ATTEMPTS,
|
|
skip_verify: bool = False,
|
|
return_dicts: bool = True, # if True, return dicts, if False, return Message objects
|
|
recreate_message_timestamp: bool = True, # if True, when input is a Message type, recreated the 'created_at' field
|
|
stream: bool = False, # TODO move to config?
|
|
timestamp: Optional[datetime.datetime] = None,
|
|
) -> Tuple[List[Union[dict, Message]], bool, bool, bool]:
|
|
"""Top-level event message handler for the MemGPT agent"""
|
|
|
|
def strip_name_field_from_user_message(user_message_text: str) -> Tuple[str, Optional[str]]:
|
|
"""If 'name' exists in the JSON string, remove it and return the cleaned text + name value"""
|
|
try:
|
|
user_message_json = dict(json.loads(user_message_text, strict=JSON_LOADS_STRICT))
|
|
# Special handling for AutoGen messages with 'name' field
|
|
# Treat 'name' as a special field
|
|
# If it exists in the input message, elevate it to the 'message' level
|
|
name = user_message_json.pop("name", None)
|
|
clean_message = json.dumps(user_message_json, ensure_ascii=JSON_ENSURE_ASCII)
|
|
|
|
except Exception as e:
|
|
print(f"{CLI_WARNING_PREFIX}handling of 'name' field failed with: {e}")
|
|
|
|
return clean_message, name
|
|
|
|
def validate_json(user_message_text: str, raise_on_error: bool) -> str:
|
|
try:
|
|
user_message_json = dict(json.loads(user_message_text, strict=JSON_LOADS_STRICT))
|
|
user_message_json_val = json.dumps(user_message_json, ensure_ascii=JSON_ENSURE_ASCII)
|
|
return user_message_json_val
|
|
except Exception as e:
|
|
print(f"{CLI_WARNING_PREFIX}couldn't parse user input message as JSON: {e}")
|
|
if raise_on_error:
|
|
raise e
|
|
|
|
try:
|
|
# Step 0: add user message
|
|
if user_message is not None:
|
|
if isinstance(user_message, Message):
|
|
# Validate JSON via save/load
|
|
user_message_text = validate_json(user_message.text, False)
|
|
cleaned_user_message_text, name = strip_name_field_from_user_message(user_message_text)
|
|
|
|
if name is not None:
|
|
# Update Message object
|
|
user_message.text = cleaned_user_message_text
|
|
user_message.name = name
|
|
|
|
# Recreate timestamp
|
|
if recreate_message_timestamp:
|
|
user_message.created_at = get_utc_time()
|
|
|
|
elif isinstance(user_message, str):
|
|
# Validate JSON via save/load
|
|
user_message = validate_json(user_message, False)
|
|
cleaned_user_message_text, name = strip_name_field_from_user_message(user_message)
|
|
|
|
# If user_message['name'] is not None, it will be handled properly by dict_to_message
|
|
# So no need to run strip_name_field_from_user_message
|
|
|
|
# Create the associated Message object (in the database)
|
|
user_message = Message.dict_to_message(
|
|
agent_id=self.agent_state.id,
|
|
user_id=self.agent_state.user_id,
|
|
model=self.model,
|
|
openai_message_dict={"role": "user", "content": cleaned_user_message_text, "name": name},
|
|
created_at=timestamp,
|
|
)
|
|
|
|
else:
|
|
raise ValueError(f"Bad type for user_message: {type(user_message)}")
|
|
|
|
self.interface.user_message(user_message.text, msg_obj=user_message)
|
|
|
|
input_message_sequence = self._messages + [user_message]
|
|
# Alternatively, the requestor can send an empty user message
|
|
else:
|
|
input_message_sequence = self._messages
|
|
|
|
if len(input_message_sequence) > 1 and input_message_sequence[-1].role != "user":
|
|
printd(f"{CLI_WARNING_PREFIX}Attempting to run ChatCompletion without user as the last message in the queue")
|
|
|
|
# Step 1: send the conversation and available functions to GPT
|
|
if not skip_verify and (first_message or self.messages_total == self.messages_total_init):
|
|
printd(f"This is the first message. Running extra verifier on AI response.")
|
|
counter = 0
|
|
while True:
|
|
response = self._get_ai_reply(
|
|
message_sequence=input_message_sequence,
|
|
first_message=True, # passed through to the prompt formatter
|
|
stream=stream,
|
|
)
|
|
if verify_first_message_correctness(response, require_monologue=self.first_message_verify_mono):
|
|
break
|
|
|
|
counter += 1
|
|
if counter > first_message_retry_limit:
|
|
raise Exception(f"Hit first message retry limit ({first_message_retry_limit})")
|
|
|
|
else:
|
|
response = self._get_ai_reply(
|
|
message_sequence=input_message_sequence,
|
|
stream=stream,
|
|
)
|
|
|
|
# Step 2: check if LLM wanted to call a function
|
|
# (if yes) Step 3: call the function
|
|
# (if yes) Step 4: send the info on the function call and function response to LLM
|
|
response_message = response.choices[0].message
|
|
response_message.model_copy() # TODO why are we copying here?
|
|
all_response_messages, heartbeat_request, function_failed = self._handle_ai_response(response_message)
|
|
|
|
# Add the extra metadata to the assistant response
|
|
# (e.g. enough metadata to enable recreating the API call)
|
|
# assert "api_response" not in all_response_messages[0]
|
|
# all_response_messages[0]["api_response"] = response_message_copy
|
|
# assert "api_args" not in all_response_messages[0]
|
|
# all_response_messages[0]["api_args"] = {
|
|
# "model": self.model,
|
|
# "messages": input_message_sequence,
|
|
# "functions": self.functions,
|
|
# }
|
|
|
|
# Step 4: extend the message history
|
|
if user_message is not None:
|
|
if isinstance(user_message, Message):
|
|
all_new_messages = [user_message] + all_response_messages
|
|
else:
|
|
raise ValueError(type(user_message))
|
|
else:
|
|
all_new_messages = all_response_messages
|
|
|
|
# Check the memory pressure and potentially issue a memory pressure warning
|
|
current_total_tokens = response.usage.total_tokens
|
|
active_memory_warning = False
|
|
# We can't do summarize logic properly if context_window is undefined
|
|
if self.agent_state.llm_config.context_window is None:
|
|
# Fallback if for some reason context_window is missing, just set to the default
|
|
print(f"{CLI_WARNING_PREFIX}could not find context_window in config, setting to default {LLM_MAX_TOKENS['DEFAULT']}")
|
|
print(f"{self.agent_state}")
|
|
self.agent_state.llm_config.context_window = (
|
|
LLM_MAX_TOKENS[self.model] if (self.model is not None and self.model in LLM_MAX_TOKENS) else LLM_MAX_TOKENS["DEFAULT"]
|
|
)
|
|
if current_total_tokens > MESSAGE_SUMMARY_WARNING_FRAC * int(self.agent_state.llm_config.context_window):
|
|
printd(
|
|
f"{CLI_WARNING_PREFIX}last response total_tokens ({current_total_tokens}) > {MESSAGE_SUMMARY_WARNING_FRAC * int(self.agent_state.llm_config.context_window)}"
|
|
)
|
|
# Only deliver the alert if we haven't already (this period)
|
|
if not self.agent_alerted_about_memory_pressure:
|
|
active_memory_warning = True
|
|
self.agent_alerted_about_memory_pressure = True # it's up to the outer loop to handle this
|
|
else:
|
|
printd(
|
|
f"last response total_tokens ({current_total_tokens}) < {MESSAGE_SUMMARY_WARNING_FRAC * int(self.agent_state.llm_config.context_window)}"
|
|
)
|
|
|
|
self._append_to_messages(all_new_messages)
|
|
messages_to_return = [msg.to_openai_dict() for msg in all_new_messages] if return_dicts else all_new_messages
|
|
|
|
# update state after each step
|
|
self.update_state()
|
|
|
|
return messages_to_return, heartbeat_request, function_failed, active_memory_warning, response.usage
|
|
|
|
except Exception as e:
|
|
printd(f"step() failed\nuser_message = {user_message}\nerror = {e}")
|
|
|
|
# If we got a context alert, try trimming the messages length, then try again
|
|
if is_context_overflow_error(e):
|
|
# A separate API call to run a summarizer
|
|
self.summarize_messages_inplace()
|
|
|
|
# Try step again
|
|
return self.step(user_message, first_message=first_message, return_dicts=return_dicts)
|
|
else:
|
|
printd(f"step() failed with an unrecognized exception: '{str(e)}'")
|
|
raise e
|
|
|
|
def summarize_messages_inplace(self, cutoff=None, preserve_last_N_messages=True, disallow_tool_as_first=True):
|
|
assert self.messages[0]["role"] == "system", f"self.messages[0] should be system (instead got {self.messages[0]})"
|
|
|
|
# Start at index 1 (past the system message),
|
|
# and collect messages for summarization until we reach the desired truncation token fraction (eg 50%)
|
|
# Do not allow truncation of the last N messages, since these are needed for in-context examples of function calling
|
|
token_counts = [count_tokens(str(msg)) for msg in self.messages]
|
|
message_buffer_token_count = sum(token_counts[1:]) # no system message
|
|
desired_token_count_to_summarize = int(message_buffer_token_count * MESSAGE_SUMMARY_TRUNC_TOKEN_FRAC)
|
|
candidate_messages_to_summarize = self.messages[1:]
|
|
token_counts = token_counts[1:]
|
|
|
|
if preserve_last_N_messages:
|
|
candidate_messages_to_summarize = candidate_messages_to_summarize[:-MESSAGE_SUMMARY_TRUNC_KEEP_N_LAST]
|
|
token_counts = token_counts[:-MESSAGE_SUMMARY_TRUNC_KEEP_N_LAST]
|
|
|
|
# if disallow_tool_as_first:
|
|
# # We have to make sure that a "tool" call is not sitting at the front (after system message),
|
|
# # otherwise we'll get an error from OpenAI (if using the OpenAI API)
|
|
# while len(candidate_messages_to_summarize) > 0:
|
|
# if candidate_messages_to_summarize[0]["role"] in ["tool", "function"]:
|
|
# candidate_messages_to_summarize.pop(0)
|
|
# else:
|
|
# break
|
|
|
|
printd(f"MESSAGE_SUMMARY_TRUNC_TOKEN_FRAC={MESSAGE_SUMMARY_TRUNC_TOKEN_FRAC}")
|
|
printd(f"MESSAGE_SUMMARY_TRUNC_KEEP_N_LAST={MESSAGE_SUMMARY_TRUNC_KEEP_N_LAST}")
|
|
printd(f"token_counts={token_counts}")
|
|
printd(f"message_buffer_token_count={message_buffer_token_count}")
|
|
printd(f"desired_token_count_to_summarize={desired_token_count_to_summarize}")
|
|
printd(f"len(candidate_messages_to_summarize)={len(candidate_messages_to_summarize)}")
|
|
|
|
# If at this point there's nothing to summarize, throw an error
|
|
if len(candidate_messages_to_summarize) == 0:
|
|
raise LLMError(
|
|
f"Summarize error: tried to run summarize, but couldn't find enough messages to compress [len={len(self.messages)}, preserve_N={MESSAGE_SUMMARY_TRUNC_KEEP_N_LAST}]"
|
|
)
|
|
|
|
# Walk down the message buffer (front-to-back) until we hit the target token count
|
|
tokens_so_far = 0
|
|
cutoff = 0
|
|
for i, msg in enumerate(candidate_messages_to_summarize):
|
|
cutoff = i
|
|
tokens_so_far += token_counts[i]
|
|
if tokens_so_far > desired_token_count_to_summarize:
|
|
break
|
|
# Account for system message
|
|
cutoff += 1
|
|
|
|
# Try to make an assistant message come after the cutoff
|
|
try:
|
|
printd(f"Selected cutoff {cutoff} was a 'user', shifting one...")
|
|
if self.messages[cutoff]["role"] == "user":
|
|
new_cutoff = cutoff + 1
|
|
if self.messages[new_cutoff]["role"] == "user":
|
|
printd(f"Shifted cutoff {new_cutoff} is still a 'user', ignoring...")
|
|
cutoff = new_cutoff
|
|
except IndexError:
|
|
pass
|
|
|
|
# Make sure the cutoff isn't on a 'tool' or 'function'
|
|
if disallow_tool_as_first:
|
|
while self.messages[cutoff]["role"] in ["tool", "function"] and cutoff < len(self.messages):
|
|
printd(f"Selected cutoff {cutoff} was a 'tool', shifting one...")
|
|
cutoff += 1
|
|
|
|
message_sequence_to_summarize = self._messages[1:cutoff] # do NOT get rid of the system message
|
|
if len(message_sequence_to_summarize) <= 1:
|
|
# This prevents a potential infinite loop of summarizing the same message over and over
|
|
raise LLMError(
|
|
f"Summarize error: tried to run summarize, but couldn't find enough messages to compress [len={len(message_sequence_to_summarize)} <= 1]"
|
|
)
|
|
else:
|
|
printd(f"Attempting to summarize {len(message_sequence_to_summarize)} messages [1:{cutoff}] of {len(self._messages)}")
|
|
|
|
# We can't do summarize logic properly if context_window is undefined
|
|
if self.agent_state.llm_config.context_window is None:
|
|
# Fallback if for some reason context_window is missing, just set to the default
|
|
print(f"{CLI_WARNING_PREFIX}could not find context_window in config, setting to default {LLM_MAX_TOKENS['DEFAULT']}")
|
|
print(f"{self.agent_state}")
|
|
self.agent_state.llm_config.context_window = (
|
|
LLM_MAX_TOKENS[self.model] if (self.model is not None and self.model in LLM_MAX_TOKENS) else LLM_MAX_TOKENS["DEFAULT"]
|
|
)
|
|
summary = summarize_messages(agent_state=self.agent_state, message_sequence_to_summarize=message_sequence_to_summarize)
|
|
printd(f"Got summary: {summary}")
|
|
|
|
# Metadata that's useful for the agent to see
|
|
all_time_message_count = self.messages_total
|
|
remaining_message_count = len(self.messages[cutoff:])
|
|
hidden_message_count = all_time_message_count - remaining_message_count
|
|
summary_message_count = len(message_sequence_to_summarize)
|
|
summary_message = package_summarize_message(summary, summary_message_count, hidden_message_count, all_time_message_count)
|
|
printd(f"Packaged into message: {summary_message}")
|
|
|
|
prior_len = len(self.messages)
|
|
self._trim_messages(cutoff)
|
|
packed_summary_message = {"role": "user", "content": summary_message}
|
|
self._prepend_to_messages(
|
|
[
|
|
Message.dict_to_message(
|
|
agent_id=self.agent_state.id,
|
|
user_id=self.agent_state.user_id,
|
|
model=self.model,
|
|
openai_message_dict=packed_summary_message,
|
|
)
|
|
]
|
|
)
|
|
|
|
# reset alert
|
|
self.agent_alerted_about_memory_pressure = False
|
|
|
|
printd(f"Ran summarizer, messages length {prior_len} -> {len(self.messages)}")
|
|
|
|
def heartbeat_is_paused(self):
|
|
"""Check if there's a requested pause on timed heartbeats"""
|
|
|
|
# Check if the pause has been initiated
|
|
if self.pause_heartbeats_start is None:
|
|
return False
|
|
|
|
# Check if it's been more than pause_heartbeats_minutes since pause_heartbeats_start
|
|
elapsed_time = get_utc_time() - self.pause_heartbeats_start
|
|
return elapsed_time.total_seconds() < self.pause_heartbeats_minutes * 60
|
|
|
|
def rebuild_memory(self):
|
|
"""Rebuilds the system message with the latest memory object"""
|
|
curr_system_message = self.messages[0] # this is the system + memory bank, not just the system prompt
|
|
|
|
# NOTE: This is a hacky way to check if the memory has changed
|
|
memory_repr = str(self.memory)
|
|
if memory_repr == curr_system_message["content"][-(len(memory_repr)) :]:
|
|
printd(f"Memory has not changed, not rebuilding system")
|
|
return
|
|
|
|
# update memory (TODO: potentially update recall/archival stats seperately)
|
|
new_system_message = initialize_message_sequence(
|
|
self.model,
|
|
self.system,
|
|
self.memory,
|
|
archival_memory=self.persistence_manager.archival_memory,
|
|
recall_memory=self.persistence_manager.recall_memory,
|
|
)[0]
|
|
|
|
diff = united_diff(curr_system_message["content"], new_system_message["content"])
|
|
if len(diff) > 0: # there was a diff
|
|
printd(f"Rebuilding system with new memory...\nDiff:\n{diff}")
|
|
|
|
# Swap the system message out (only if there is a diff)
|
|
self._swap_system_message(
|
|
Message.dict_to_message(
|
|
agent_id=self.agent_state.id, user_id=self.agent_state.user_id, model=self.model, openai_message_dict=new_system_message
|
|
)
|
|
)
|
|
assert self.messages[0]["content"] == new_system_message["content"], (
|
|
self.messages[0]["content"],
|
|
new_system_message["content"],
|
|
)
|
|
|
|
def add_function(self, function_name: str) -> str:
|
|
# TODO: refactor
|
|
raise NotImplementedError
|
|
# if function_name in self.functions_python.keys():
|
|
# msg = f"Function {function_name} already loaded"
|
|
# printd(msg)
|
|
# return msg
|
|
|
|
# available_functions = load_all_function_sets()
|
|
# if function_name not in available_functions.keys():
|
|
# raise ValueError(f"Function {function_name} not found in function library")
|
|
|
|
# self.functions.append(available_functions[function_name]["json_schema"])
|
|
# self.functions_python[function_name] = available_functions[function_name]["python_function"]
|
|
|
|
# msg = f"Added function {function_name}"
|
|
## self.save()
|
|
# self.update_state()
|
|
# printd(msg)
|
|
# return msg
|
|
|
|
def remove_function(self, function_name: str) -> str:
|
|
# TODO: refactor
|
|
raise NotImplementedError
|
|
# if function_name not in self.functions_python.keys():
|
|
# msg = f"Function {function_name} not loaded, ignoring"
|
|
# printd(msg)
|
|
# return msg
|
|
|
|
## only allow removal of user defined functions
|
|
# user_func_path = Path(USER_FUNCTIONS_DIR)
|
|
# func_path = Path(inspect.getfile(self.functions_python[function_name]))
|
|
# is_subpath = func_path.resolve().parts[: len(user_func_path.resolve().parts)] == user_func_path.resolve().parts
|
|
|
|
# if not is_subpath:
|
|
# raise ValueError(f"Function {function_name} is not user defined and cannot be removed")
|
|
|
|
# self.functions = [f_schema for f_schema in self.functions if f_schema["name"] != function_name]
|
|
# self.functions_python.pop(function_name)
|
|
|
|
# msg = f"Removed function {function_name}"
|
|
## self.save()
|
|
# self.update_state()
|
|
# printd(msg)
|
|
# return msg
|
|
|
|
def update_state(self) -> AgentState:
|
|
memory = {
|
|
"system": self.system,
|
|
"memory": self.memory.to_dict(),
|
|
"messages": [str(msg.id) for msg in self._messages], # TODO: move out into AgentState.message_ids
|
|
}
|
|
self.agent_state = AgentState(
|
|
name=self.agent_state.name,
|
|
user_id=self.agent_state.user_id,
|
|
tools=self.agent_state.tools,
|
|
system=self.system,
|
|
## "model_state"
|
|
llm_config=self.agent_state.llm_config,
|
|
embedding_config=self.agent_state.embedding_config,
|
|
id=self.agent_state.id,
|
|
created_at=self.agent_state.created_at,
|
|
## "agent_state"
|
|
state=memory,
|
|
_metadata=self.agent_state._metadata,
|
|
)
|
|
return self.agent_state
|
|
|
|
def migrate_embedding(self, embedding_config: EmbeddingConfig):
|
|
"""Migrate the agent to a new embedding"""
|
|
# TODO: archival memory
|
|
|
|
# TODO: recall memory
|
|
raise NotImplementedError()
|
|
|
|
def attach_source(self, source_name, source_connector: StorageConnector, ms: MetadataStore):
|
|
"""Attach data with name `source_name` to the agent from source_connector."""
|
|
# TODO: eventually, adding a data source should just give access to the retriever the source table, rather than modifying archival memory
|
|
|
|
filters = {"user_id": self.agent_state.user_id, "data_source": source_name}
|
|
size = source_connector.size(filters)
|
|
# typer.secho(f"Ingesting {size} passages into {agent.name}", fg=typer.colors.GREEN)
|
|
page_size = 100
|
|
generator = source_connector.get_all_paginated(filters=filters, page_size=page_size) # yields List[Passage]
|
|
all_passages = []
|
|
for i in tqdm(range(0, size, page_size)):
|
|
passages = next(generator)
|
|
|
|
# need to associated passage with agent (for filtering)
|
|
for passage in passages:
|
|
assert isinstance(passage, Passage), f"Generate yielded bad non-Passage type: {type(passage)}"
|
|
passage.agent_id = self.agent_state.id
|
|
|
|
# regenerate passage ID (avoid duplicates)
|
|
passage.id = create_uuid_from_string(f"{source_name}_{str(passage.agent_id)}_{passage.text}")
|
|
|
|
# insert into agent archival memory
|
|
self.persistence_manager.archival_memory.storage.insert_many(passages)
|
|
all_passages += passages
|
|
|
|
assert size == len(all_passages), f"Expected {size} passages, but only got {len(all_passages)}"
|
|
|
|
# save destination storage
|
|
self.persistence_manager.archival_memory.storage.save()
|
|
|
|
# attach to agent
|
|
source = ms.get_source(source_name=source_name, user_id=self.agent_state.user_id)
|
|
assert source is not None, f"source does not exist for source_name={source_name}, user_id={self.agent_state.user_id}"
|
|
source_id = source.id
|
|
ms.attach_source(agent_id=self.agent_state.id, source_id=source_id, user_id=self.agent_state.user_id)
|
|
|
|
total_agent_passages = self.persistence_manager.archival_memory.storage.size()
|
|
|
|
printd(
|
|
f"Attached data source {source_name} to agent {self.agent_state.name}, consisting of {len(all_passages)}. Agent now has {total_agent_passages} embeddings in archival memory.",
|
|
)
|
|
|
|
|
|
def save_agent(agent: Agent, ms: MetadataStore):
|
|
"""Save agent to metadata store"""
|
|
|
|
agent.update_state()
|
|
agent_state = agent.agent_state
|
|
|
|
if ms.get_agent(agent_name=agent_state.name, user_id=agent_state.user_id):
|
|
ms.update_agent(agent_state)
|
|
else:
|
|
ms.create_agent(agent_state)
|