use request context when create agent

letta/agent.py

@@ -21,7 +21,7 @@
 )
 from letta.errors import LLMError
 from letta.helpers import ToolRulesSolver
-from letta.interface import AgentInterface
+from letta.interface import AgentInterface, RequestContext
 from letta.llm_api.helpers import is_context_overflow_error
 from letta.llm_api.llm_api_tools import create
 from letta.local_llm.utils import num_tokens_from_functions, num_tokens_from_messages
@@ -532,18 +532,19 @@ def _get_ai_reply(
 
         try:
             response = create(
-                # agent_state=self.agent_state,
-                llm_config=self.agent_state.llm_config,
-                messages=message_sequence,
-                user_id=self.agent_state.user_id,
-                functions=allowed_functions,
-                functions_python=self.functions_python,
-                function_call=function_call,
-                # hint
-                first_message=first_message,
-                # streaming
-                stream=stream,
-                stream_interface=self.interface,
+                RequestContext(
+                    llm_config=self.agent_state.llm_config,
+                    messages=message_sequence,
+                    user_id=self.agent_state.user_id,
+                    functions=allowed_functions,
+                    functions_python=self.functions_python,
+                    function_call=function_call,
+                    # hint
+                    first_message=first_message,
+                    # streaming
+                    stream=stream,
+                    stream_interface=self.interface,
+                )
             )
 
             if len(response.choices) == 0 or response.choices[0] is None:

letta/functions/function_sets/extras.py

@@ -8,6 +8,7 @@
     MESSAGE_CHATGPT_FUNCTION_MODEL,
     MESSAGE_CHATGPT_FUNCTION_SYSTEM_MESSAGE,
 )
+from letta.interface import RequestContext
 from letta.llm_api.llm_api_tools import create
 from letta.schemas.message import Message
 from letta.utils import json_dumps, json_loads
@@ -31,8 +32,10 @@ def message_chatgpt(self, message: str):
     ]
     # TODO: this will error without an LLMConfig
     response = create(
-        model=MESSAGE_CHATGPT_FUNCTION_MODEL,
-        messages=message_sequence,
+        RequestContext(
+            model=MESSAGE_CHATGPT_FUNCTION_MODEL,
+            messages=message_sequence,
+        )
     )
 
     reply = response.choices[0].message.content

letta/interface.py

@@ -1,6 +1,6 @@
 import re
 from abc import ABC, abstractmethod
-from typing import List, Optional
+from typing import List, Optional, Union
 
 from colorama import Fore, Style, init
 
@@ -10,10 +10,30 @@
     INNER_THOUGHTS_CLI_SYMBOL,
 )
 from letta.schemas.message import Message
+from letta.streaming_interface import AgentChunkStreamingInterface, AgentRefreshStreamingInterface
 from letta.utils import json_loads, printd
+from dataclasses import dataclass
+from letta.schemas.llm_config import LLMConfig
 
 init(autoreset=True)
 
+@dataclass
+class RequestContext:
+    llm_config: LLMConfig
+    messages: List[Message]
+    user_id: Optional[str] = None
+    functions: Optional[list] = None
+    functions_python: Optional[dict] = None
+    function_call: str = "auto"
+    first_message: bool = False
+    use_tool_naming: bool = True
+    stream: bool = False
+    stream_interface: Optional[Union[AgentRefreshStreamingInterface, AgentChunkStreamingInterface]] = None
+    max_tokens: Optional[int] = None
+    model_settings: Optional[dict] = None
+
+
+
 # DEBUG = True  # puts full message outputs in the terminal
 DEBUG = False  # only dumps important messages in the terminal