fix: change inner_thoughts to thinking in kwargs (#3165)

letta/agent.py

@@ -36,6 +36,7 @@ from letta.interface import AgentInterface
 from letta.llm_api.helpers import calculate_summarizer_cutoff, get_token_counts_for_messages, is_context_overflow_error
 from letta.llm_api.llm_api_tools import create
 from letta.llm_api.llm_client import LLMClient
+from letta.local_llm.constants import INNER_THOUGHTS_KWARG
 from letta.local_llm.utils import num_tokens_from_functions, num_tokens_from_messages
 from letta.log import get_logger
 from letta.memory import summarize_messages
@@ -548,8 +549,8 @@ class Agent(BaseAgent):
                 return messages, False, True  # force a heartbeat to allow agent to handle error
 
             # Check if inner thoughts is in the function call arguments (possible apparently if you are using Azure)
-            if "inner_thoughts" in function_args:
-                response_message.content = function_args.pop("inner_thoughts")
+            if INNER_THOUGHTS_KWARG in function_args:
+                response_message.content = function_args.pop(INNER_THOUGHTS_KWARG)
             # The content if then internal monologue, not chat
             if response_message.content and not nonnull_content:
                 self.interface.internal_monologue(response_message.content, msg_obj=messages[-1], chunk_index=chunk_index)

letta/local_llm/constants.py

@@ -25,7 +25,7 @@ DEFAULT_OLLAMA_MODEL = "dolphin2.2-mistral:7b-q6_K"
 DEFAULT_WRAPPER = ChatMLInnerMonologueWrapper
 DEFAULT_WRAPPER_NAME = "chatml"
 
-INNER_THOUGHTS_KWARG = "inner_thoughts"
+INNER_THOUGHTS_KWARG = "thinking"  # used to be "inner_thoughts"
 INNER_THOUGHTS_KWARG_VERTEX = "thinking"
 INNER_THOUGHTS_KWARG_DESCRIPTION = "Deep inner monologue private to you only."
 INNER_THOUGHTS_KWARG_DESCRIPTION_GO_FIRST = f"Deep inner monologue private to you only. Think before you act, so always generate arg '{INNER_THOUGHTS_KWARG}' first before any other arg."

letta/local_llm/json_parser.py

@@ -1,5 +1,6 @@
 import json
 import re
+import warnings
 
 from letta.errors import LLMJSONParsingError
 from letta.helpers.json_helpers import json_loads
@@ -77,10 +78,19 @@ def add_missing_heartbeat(llm_json):
 
 
 def clean_and_interpret_send_message_json(json_string):
+    from letta.local_llm.constants import INNER_THOUGHTS_KWARG
+
     # If normal parsing fails, attempt to clean and extract manually
     cleaned_json_string = re.sub(r"[^\x00-\x7F]+", "", json_string)  # Remove non-ASCII characters
     function_match = re.search(r'"function":\s*"send_message"', cleaned_json_string)
-    inner_thoughts_match = re.search(r'"inner_thoughts":\s*"([^"]+)"', cleaned_json_string)
+    if INNER_THOUGHTS_KWARG == "inner_thoughts":
+        inner_thoughts_match = re.search(r'"inner_thoughts":\s*"([^"]+)"', cleaned_json_string)
+    elif INNER_THOUGHTS_KWARG == "thinking":
+        inner_thoughts_match = re.search(r'"thinking":\s*"([^"]+)"', cleaned_json_string)
+    else:
+        warnings.warn(f"INNER_THOUGHTS_KWARG is not valid: {INNER_THOUGHTS_KWARG}")
+        inner_thoughts_match = re.search(r'"inner_thoughts":\s*"([^"]+)"', cleaned_json_string)
+
     message_match = re.search(r'"message":\s*"([^"]+)"', cleaned_json_string)
 
     if function_match and inner_thoughts_match and message_match:

letta/streaming_utils.py

@@ -1,6 +1,7 @@
 from typing import Optional, Tuple
 
 from letta.constants import DEFAULT_MESSAGE_TOOL_KWARG
+from letta.local_llm.constants import INNER_THOUGHTS_KWARG
 
 
 class JSONInnerThoughtsExtractor:
@@ -34,7 +35,7 @@ class JSONInnerThoughtsExtractor:
 
     """
 
-    def __init__(self, inner_thoughts_key="inner_thoughts", wait_for_first_key=False):
+    def __init__(self, inner_thoughts_key=INNER_THOUGHTS_KWARG, wait_for_first_key=False):
         self.inner_thoughts_key = inner_thoughts_key
         self.wait_for_first_key = wait_for_first_key
         self.main_buffer = ""