Refactoring CLI to use config file, connect to Llama Index data sources, and allow for multiple agents

memgpt/agent.py

@@ -1,19 +1,22 @@
 import asyncio
 import inspect
 import datetime
+import glob
 import pickle
 import math
 import os
 import json
 import threading
 
 import openai
-
+from memgpt.persistence_manager import LocalStateManager
+from memgpt.config import AgentConfig
 from .system import get_heartbeat, get_login_event, package_function_response, package_summarize_message, get_initial_boot_messages
 from .memory import CoreMemory as Memory, summarize_messages, a_summarize_messages
 from .openai_tools import acompletions_with_backoff as acreate, completions_with_backoff as create
 from .utils import get_local_time, parse_json, united_diff, printd, count_tokens
 from .constants import (
+    MEMGPT_DIR,
     FIRST_MESSAGE_ATTEMPTS,
     MAX_PAUSE_HEARTBEATS,
     MESSAGE_CHATGPT_FUNCTION_MODEL,
@@ -167,6 +170,7 @@ async def call_function(function_to_call, **function_args):
 class Agent(object):
     def __init__(
         self,
+        config,
         model,
         system,
         functions,
@@ -178,6 +182,8 @@ def __init__(
         persistence_manager_init=True,
         first_message_verify_mono=True,
     ):
+        # agent config
+        self.config = config
         # gpt-4, gpt-3.5-turbo
         self.model = model
         # Store the system instructions (used to rebuild memory)
@@ -194,7 +200,8 @@ def __init__(
         )
         # 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 = self.messages_total
+        # self.messages_total_init = self.messages_total
+        self.messages_total_init = len(self._messages) - 1
         printd(f"AgentAsync initialized, self.messages_total={self.messages_total}")
 
         # Interface must implement:
@@ -331,6 +338,61 @@ def save_to_json_file(self, filename):
         with open(filename, "w") as file:
             json.dump(self.to_dict(), file)
 
+    def save(self):
+        """Save agent state locally"""
+
+        timestamp = get_local_time().replace(" ", "_").replace(":", "_")
+        agent_name = self.config.name  # TODO: fix
+
+        # save agent state
+        filename = f"{timestamp}.json"
+        os.makedirs(self.config.save_state_dir(), exist_ok=True)
+        self.save_to_json_file(os.path.join(self.config.save_state_dir(), filename))
+
+        # save the persistence manager too
+        filename = f"{timestamp}.persistence.pickle"
+        os.makedirs(self.config.save_persistence_manager_dir(), exist_ok=True)
+        self.persistence_manager.save(os.path.join(self.config.save_persistence_manager_dir(), filename))
+
+    @classmethod
+    def load_agent(cls, interface, agent_config: AgentConfig):
+        """Load saved agent state"""
+        # TODO: support loading from specific file
+        agent_name = agent_config.name
+
+        # load state
+        directory = agent_config.save_state_dir()
+        json_files = glob.glob(f"{directory}/*.json")  # This will list all .json files in the current directory.
+        if not json_files:
+            print(f"/load error: no .json checkpoint files found")
+            raise ValueError(f"Cannot load {agent_name}")
+
+        # Sort files based on modified timestamp, with the latest file being the first.
+        filename = max(json_files, key=os.path.getmtime)
+        state = json.load(open(filename, "r"))
+
+        # load persistence manager
+        filename = os.path.basename(filename).replace(".json", ".persistence.pickle")
+        directory = agent_config.save_persistence_manager_dir()
+        persistence_manager = LocalStateManager.load(os.path.join(directory, filename), agent_config)
+
+        messages = state["messages"]
+        agent = cls(
+            config=agent_config,
+            model=state["model"],
+            system=state["system"],
+            functions=state["functions"],
+            interface=interface,
+            persistence_manager=persistence_manager,
+            persistence_manager_init=False,
+            persona_notes=state["memory"]["persona"],
+            human_notes=state["memory"]["human"],
+            messages_total=state["messages_total"] if "messages_total" in state else len(messages) - 1,
+        )
+        agent._messages = messages
+        agent.memory = initialize_memory(state["memory"]["persona"], state["memory"]["human"])
+        return agent
+
     @classmethod
     def load(cls, state, interface, persistence_manager):
         model = state["model"]
@@ -875,6 +937,9 @@ async def step(self, user_message, first_message=False, first_message_retry_limi
 
             if len(input_message_sequence) > 1 and input_message_sequence[-1]["role"] != "user":
                 printd(f"WARNING: attempting to run ChatCompletion without user as the last message in the queue")
+                from pprint import pprint
+
+                pprint(input_message_sequence[-1])
 
             # 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):
@@ -901,9 +966,9 @@ async def step(self, user_message, first_message=False, first_message_retry_limi
 
             # 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]
+            assert "api_response" not in all_response_messages[0], f"api_response already in {all_response_messages[0]}"
             all_response_messages[0]["api_response"] = response_message_copy
-            assert "api_args" not in all_response_messages[0]
+            assert "api_args" not in all_response_messages[0], f"api_args already in {all_response_messages[0]}"
             all_response_messages[0]["api_args"] = {
                 "model": self.model,
                 "messages": input_message_sequence,
@@ -933,6 +998,7 @@ async def step(self, user_message, first_message=False, first_message_retry_limi
 
         except Exception as e:
             printd(f"step() failed\nuser_message = {user_message}\nerror = {e}")
+            print(f"step() failed\nuser_message = {user_message}\nerror = {e}")
 
             # If we got a context alert, try trimming the messages length, then try again
             if "maximum context length" in str(e):
@@ -943,6 +1009,7 @@ async def step(self, user_message, first_message=False, first_message_retry_limi
                 return await self.step(user_message, first_message=first_message)
             else:
                 printd(f"step() failed with openai.InvalidRequestError, but didn't recognize the error message: '{str(e)}'")
+                print(e)
                 raise e
 
     async def summarize_messages_inplace(self, cutoff=None):

memgpt/cli/cli.py

@@ -0,0 +1,132 @@
+import typer
+import logging
+import asyncio
+import os
+from prettytable import PrettyTable
+import questionary
+import openai
+
+
+import memgpt.interface  # for printing to terminal
+from memgpt.cli.cli_config import configure
+import memgpt.agent as agent
+import memgpt.system as system
+import memgpt.presets as presets
+import memgpt.constants as constants
+import memgpt.personas.personas as personas
+import memgpt.humans.humans as humans
+import memgpt.utils as utils
+from memgpt.utils import printd
+from memgpt.persistence_manager import LocalStateManager
+from memgpt.config import MemGPTConfig, AgentConfig
+from memgpt.constants import MEMGPT_DIR
+from memgpt.agent import AgentAsync
+
+
+def run(
+    persona: str = typer.Option(None, help="Specify persona"),
+    agent: str = typer.Option(None, help="Specify agent save file"),
+    human: str = typer.Option(None, help="Specify human"),
+    model: str = typer.Option(None, help="Specify the LLM model"),
+    data_source: str = typer.Option(None, help="Specify data source to attach to agent"),
+    first: bool = typer.Option(False, "--first", help="Use --first to send the first message in the sequence"),
+    debug: bool = typer.Option(False, "--debug", help="Use --debug to enable debugging output"),
+    no_verify: bool = typer.Option(False, "--no_verify", help="Bypass message verification"),
+    yes: bool = typer.Option(False, "-y", help="Skip confirmation prompt and use defaults"),
+):
+    """Start chatting with an MemGPT agent
+
+    Example usage: `memgpt run --agent myagent --data-source mydata --persona mypersona --human myhuman --model gpt-3.5-turbo`
+
+    :param persona: Specify persona
+    :param agent: Specify agent name (will load existing state if the agent exists, or create a new one with that name)
+    :param human: Specify human
+    :param model: Specify the LLM model
+    :param data_source: Specify data source to attach to agent (if new agent is being created)
+
+    """
+
+    # setup logger
+    utils.DEBUG = debug
+    logging.getLogger().setLevel(logging.CRITICAL)
+    if debug:
+        logging.getLogger().setLevel(logging.DEBUG)
+
+    if not MemGPTConfig.exists():  # if no config, run configure
+        if yes:
+            # use defaults
+            config = MemGPTConfig()
+        else:
+            # use input
+            configure()
+            config = MemGPTConfig.load()
+    else:  # load config
+        config = MemGPTConfig.load()
+
+    # override with command line arguments
+    if debug:
+        config.debug = debug
+    if no_verify:
+        config.no_verify = no_verify
+
+    # determine agent to use, if not provided
+    if not yes and not agent:
+        agent_files = utils.list_agent_config_files()
+        agents = [AgentConfig.load(f).name for f in agent_files]
+
+        if len(agents) > 0:
+            select_agent = questionary.confirm("Would you like to select an existing agent?").ask()
+            if select_agent:
+                agent = questionary.select("Select agent:", choices=agents).ask()
+
+    # create agent config
+    if agent and AgentConfig.exists(agent):  # use existing agent
+        typer.secho(f"Using existing agent {agent}", fg=typer.colors.GREEN)
+        agent_config = AgentConfig.load(agent)
+        printd("State path:", agent_config.save_state_dir())
+        printd("Persistent manager path:", agent_config.save_persistence_manager_dir())
+        printd("Index path:", agent_config.save_agent_index_dir())
+        # persistence_manager = LocalStateManager(agent_config).load() # TODO: implement load
+        # TODO: load prior agent state
+        assert not any(
+            [persona, human, model]
+        ), f"Cannot override existing agent state with command line arguments: {persona}, {human}, {model}"
+
+        # load existing agent
+        memgpt_agent = AgentAsync.load_agent(memgpt.interface, agent_config)
+    else:  # create new agent
+        # create new agent config: override defaults with args if provided
+        typer.secho("Creating new agent...", fg=typer.colors.GREEN)
+        agent_config = AgentConfig(
+            name=agent if agent else None,
+            persona=persona if persona else config.default_persona,
+            human=human if human else config.default_human,
+            model=model if model else config.model,
+        )
+
+        # attach data source to agent
+        agent_config.attach_data_source(data_source)
+
+        # TODO: allow configrable state manager (only local is supported right now)
+        persistence_manager = LocalStateManager(agent_config)  # TODO: insert dataset/pre-fill
+
+        # save new agent config
+        agent_config.save()
+        typer.secho(f"Created new agent {agent_config.name}.", fg=typer.colors.GREEN)
+
+        # create agent
+        memgpt_agent = presets.use_preset(
+            presets.DEFAULT,
+            agent_config,
+            agent_config.model,
+            agent_config.persona,
+            agent_config.human,
+            memgpt.interface,
+            persistence_manager,
+        )
+
+    # start event loop
+    from memgpt.main import run_agent_loop
+
+    loop = asyncio.get_event_loop()
+    loop.run_until_complete(run_agent_loop(memgpt_agent, first, no_verify, config))  # TODO: add back no_verify