Skip to content

Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.

Adding new LLM class

João Mouraa19a4a555651
diff --git a/src/crewai/agent.py b/src/crewai/agent.py
index b29c62f3a..a943c994d 100644
--- a/src/crewai/agent.py
+++ b/src/crewai/agent.py
@@ -1,6 +1,6 @@
import os
from inspect import signature
-from typing import Any, List, Optional
+from typing import Any, List, Optional, Union
from pydantic import Field, InstanceOf, PrivateAttr, model_validator
from crewai.agents import CacheHandler
@@ -12,6 +12,7 @@ from crewai.memory.contextual.contextual_memory import ContextualMemory
from crewai.utilities.constants import TRAINED_AGENTS_DATA_FILE, TRAINING_DATA_FILE
from crewai.utilities.training_handler import CrewTrainingHandler
from crewai.utilities.token_counter_callback import TokenCalcHandler
+from crewai.llm import LLM
def mock_agent_ops_provider():
@@ -81,8 +82,8 @@ class Agent(BaseAgent):
default=True,
description="Use system prompt for the agent.",
)
- llm: Any = Field(
- description="Language model that will run the agent.", default="gpt-4o-mini"
+ llm: Union[str, InstanceOf[LLM], Any] = Field(
+ description="Language model that will run the agent.", default=None
)
function_calling_llm: Optional[Any] = Field(
description="Language model that will run the agent.", default=None
@@ -118,17 +119,58 @@ class Agent(BaseAgent):
@model_validator(mode="after")
def post_init_setup(self):
self.agent_ops_agent_name = self.role
- self.llm = (
- getattr(self.llm, "model_name", None)
- or getattr(self.llm, "deployment_name", None)
- or self.llm
- or os.environ.get("OPENAI_MODEL_NAME")
- )
- self.function_calling_llm = (
- getattr(self.function_calling_llm, "model_name", None)
- or getattr(self.function_calling_llm, "deployment_name", None)
- or self.function_calling_llm
- )
+
+ # Handle different cases for self.llm
+ if isinstance(self.llm, str):
+ # If it's a string, create an LLM instance
+ self.llm = LLM(model=self.llm)
+ elif isinstance(self.llm, LLM):
+ # If it's already an LLM instance, keep it as is
+ pass
+ elif self.llm is None:
+ # If it's None, use environment variables or default
+ model_name = os.environ.get("OPENAI_MODEL_NAME", "gpt-4o-mini")
+ llm_params = {"model": model_name}
+
+ api_base = os.environ.get("OPENAI_API_BASE")
+ if api_base:
+ llm_params["base_url"] = api_base
+
+ api_key = os.environ.get("OPENAI_API_KEY")
+ if api_key:
+ llm_params["api_key"] = api_key
+
+ self.llm = LLM(**llm_params)
+ else:
+ # For any other type, attempt to extract relevant attributes
+ llm_params = {
+ "model": getattr(self.llm, "model_name", None)
+ or getattr(self.llm, "deployment_name", None)
+ or str(self.llm),
+ "temperature": getattr(self.llm, "temperature", None),
+ "max_tokens": getattr(self.llm, "max_tokens", None),
+ "logprobs": getattr(self.llm, "logprobs", None),
+ "timeout": getattr(self.llm, "timeout", None),
+ "max_retries": getattr(self.llm, "max_retries", None),
+ "api_key": getattr(self.llm, "api_key", None),
+ "base_url": getattr(self.llm, "base_url", None),
+ "organization": getattr(self.llm, "organization", None),
+ }
+ # Remove None values to avoid passing unnecessary parameters
+ llm_params = {k: v for k, v in llm_params.items() if v is not None}
+ self.llm = LLM(**llm_params)
+
+ # Similar handling for function_calling_llm
+ if self.function_calling_llm:
+ if isinstance(self.function_calling_llm, str):
+ self.function_calling_llm = LLM(model=self.function_calling_llm)
+ elif not isinstance(self.function_calling_llm, LLM):
+ self.function_calling_llm = LLM(
+ model=getattr(self.function_calling_llm, "model_name", None)
+ or getattr(self.function_calling_llm, "deployment_name", None)
+ or str(self.function_calling_llm)
+ )
+
if not self.agent_executor:
self._setup_agent_executor()
diff --git a/src/crewai/agents/crew_agent_executor.py b/src/crewai/agents/crew_agent_executor.py
index f7bf57e8c..e256b56c6 100644
--- a/src/crewai/agents/crew_agent_executor.py
+++ b/src/crewai/agents/crew_agent_executor.py
@@ -13,7 +13,6 @@ from crewai.utilities.exceptions.context_window_exceeding_exception import (
)
from crewai.utilities.logger import Logger
from crewai.utilities.training_handler import CrewTrainingHandler
-from crewai.llm import LLM
from crewai.agents.parser import (
AgentAction,
AgentFinish,
@@ -104,23 +103,10 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
try:
while not isinstance(formatted_answer, AgentFinish):
if not self.request_within_rpm_limit or self.request_within_rpm_limit():
- if isinstance(self.llm, str):
- llm = LLM(
- model=self.llm,
- stop=self.stop if self.use_stop_words else None,
- callbacks=self.callbacks,
- )
- elif isinstance(self.llm, LLM):
- llm = self.llm
- else:
- llm = LLM(
- model=self.llm.model,
- provider=getattr(self.llm, "provider", "litellm"),
- stop=self.stop if self.use_stop_words else None,
- callbacks=self.callbacks,
- **getattr(self.llm, "llm_kwargs", {}),
- )
- answer = llm.call(self.messages)
+ answer = self.llm.call(
+ self.messages,
+ callbacks=self.callbacks,
+ )
if not self.use_stop_words:
try:
@@ -139,6 +125,7 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
action_result = self._use_tool(formatted_answer)
formatted_answer.text += f"\nObservation: {action_result}"
formatted_answer.result = action_result
+ print("formatted_answer", formatted_answer)
self._show_logs(formatted_answer)
if self.step_callback:
@@ -194,7 +181,7 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
if isinstance(formatted_answer, AgentAction):
thought = re.sub(r"\n+", "\n", formatted_answer.thought)
formatted_json = json.dumps(
- json.loads(formatted_answer.tool_input),
+ formatted_answer.tool_input,
indent=2,
ensure_ascii=False,
)
@@ -253,16 +240,6 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
return tool_result
def _summarize_messages(self) -> None:
- if isinstance(self.llm, str):
- llm = LLM(model=self.llm)
- elif isinstance(self.llm, LLM):
- llm = self.llm
- else:
- llm = LLM(
- model=self.llm.model,
- provider=getattr(self.llm, "provider", "litellm"),
- **getattr(self.llm, "llm_kwargs", {}),
- )
messages_groups = []
for message in self.messages:
@@ -272,7 +249,7 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
summarized_contents = []
for group in messages_groups:
- summary = llm.call(
+ summary = self.llm.call(
[
self._format_msg(
self._i18n.slices("summarizer_system_message"), role="system"
@@ -280,7 +257,8 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
self._format_msg(
self._i18n.errors("sumamrize_instruction").format(group=group),
),
- ]
+ ],
+ callbacks=self.callbacks,
)
summarized_contents.append(summary)
diff --git a/src/crewai/crew.py b/src/crewai/crew.py
index 91382c8ac..00d2c4b2a 100644
--- a/src/crewai/crew.py
+++ b/src/crewai/crew.py
@@ -22,6 +22,7 @@ from crewai.agent import Agent
from crewai.agents.agent_builder.base_agent import BaseAgent
from crewai.agents.cache import CacheHandler
from crewai.crews.crew_output import CrewOutput
+from crewai.llm import LLM
from crewai.memory.entity.entity_memory import EntityMemory
from crewai.memory.long_term.long_term_memory import LongTermMemory
from crewai.memory.short_term.short_term_memory import ShortTermMemory
@@ -211,11 +212,15 @@ class Crew(BaseModel):
if self.output_log_file:
self._file_handler = FileHandler(self.output_log_file)
self._rpm_controller = RPMController(max_rpm=self.max_rpm, logger=self._logger)
- self.function_calling_llm = (
- getattr(self.function_calling_llm, "model_name", None)
- or getattr(self.function_calling_llm, "deployment_name", None)
- or self.function_calling_llm
- )
+ if self.function_calling_llm:
+ if isinstance(self.function_calling_llm, str):
+ self.function_calling_llm = LLM(model=self.function_calling_llm)
+ elif not isinstance(self.function_calling_llm, LLM):
+ self.function_calling_llm = LLM(
+ model=getattr(self.function_calling_llm, "model_name", None)
+ or getattr(self.function_calling_llm, "deployment_name", None)
+ or str(self.function_calling_llm)
+ )
self._telemetry = Telemetry()
self._telemetry.set_tracer()
return self
diff --git a/src/crewai/llm.py b/src/crewai/llm.py
index 006733cd8..a1df0eeef 100644
--- a/src/crewai/llm.py
+++ b/src/crewai/llm.py
@@ -85,6 +85,3 @@ class LLM:
except Exception as e:
logging.error(f"LiteLLM call failed: {str(e)}")
raise # Re-raise the exception after logging
-
- def __getattr__(self, name):
- return self.kwargs.get(name)
diff --git a/src/crewai/telemetry/telemetry.py b/src/crewai/telemetry/telemetry.py
index b49391407..14a3ec865 100644
--- a/src/crewai/telemetry/telemetry.py
+++ b/src/crewai/telemetry/telemetry.py
@@ -117,8 +117,10 @@ class Telemetry:
"max_iter": agent.max_iter,
"max_rpm": agent.max_rpm,
"i18n": agent.i18n.prompt_file,
- "function_calling_llm": agent.function_calling_llm,
- "llm": agent.llm,
+ "function_calling_llm": agent.function_calling_llm.model
+ if agent.function_calling_llm
+ else "",
+ "llm": agent.llm.model,
"delegation_enabled?": agent.allow_delegation,
"allow_code_execution?": agent.allow_code_execution,
"max_retry_limit": agent.max_retry_limit,
@@ -182,8 +184,10 @@ class Telemetry:
"verbose?": agent.verbose,
"max_iter": agent.max_iter,
"max_rpm": agent.max_rpm,
- "function_calling_llm": agent.function_calling_llm,
- "llm": agent.llm,
+ "function_calling_llm": agent.function_calling_llm.model
+ if agent.function_calling_llm
+ else "",
+ "llm": agent.llm.model,
"delegation_enabled?": agent.allow_delegation,
"allow_code_execution?": agent.allow_code_execution,
"max_retry_limit": agent.max_retry_limit,
@@ -488,7 +492,7 @@ class Telemetry:
"max_iter": agent.max_iter,
"max_rpm": agent.max_rpm,
"i18n": agent.i18n.prompt_file,
- "llm": agent.llm,
+ "llm": agent.llm.model,
"delegation_enabled?": agent.allow_delegation,
"tools_names": [
tool.name.casefold() for tool in agent.tools or []
diff --git a/src/crewai/tools/tool_usage.py b/src/crewai/tools/tool_usage.py
index ff8a57351..cbf070758 100644
--- a/src/crewai/tools/tool_usage.py
+++ b/src/crewai/tools/tool_usage.py
@@ -72,7 +72,8 @@ class ToolUsage:
# Set the maximum parsing attempts for bigger models
if (
- self._is_gpt(self.function_calling_llm)
+ self.function_calling_llm
+ and self._is_gpt(self.function_calling_llm)
and self.function_calling_llm in OPENAI_BIGGER_MODELS
):
self._max_parsing_attempts = 2
@@ -85,6 +86,7 @@ class ToolUsage:
def use(
self, calling: Union[ToolCalling, InstructorToolCalling], tool_string: str
) -> str:
+ print("calling", calling)
if isinstance(calling, ToolUsageErrorException):
error = calling.message
if self.agent.verbose:
@@ -299,9 +301,9 @@ class ToolUsage:
def _is_gpt(self, llm) -> bool:
return (
- "gpt" in str(llm).lower()
- or "o1-preview" in str(llm).lower()
- or "o1-mini" in str(llm).lower()
+ "gpt" in str(llm.model).lower()
+ or "o1-preview" in str(llm.model).lower()
+ or "o1-mini" in str(llm.model).lower()
)
def _tool_calling(
@@ -309,11 +311,16 @@ class ToolUsage:
) -> Union[ToolCalling, InstructorToolCalling]:
try:
if self.function_calling_llm:
+ print("self.function_calling_llm")
model = (
InstructorToolCalling
if self._is_gpt(self.function_calling_llm)
else ToolCalling
)
+ print("model", model)
+ print(
+ "self.function_calling_llm.model", self.function_calling_llm.model
+ )
converter = Converter(
text=f"Only tools available:\n###\n{self._render()}\n\nReturn a valid schema for the tool, the tool name must be exactly equal one of the options, use this text to inform the valid output schema:\n\n### TEXT \n{tool_string}",
llm=self.function_calling_llm,
@@ -329,7 +336,15 @@ class ToolUsage:
),
max_attempts=1,
)
- calling = converter.to_pydantic()
+ print("converter", converter)
+ tool_object = converter.to_pydantic()
+ print("tool_object", tool_object)
+ calling = ToolCalling(
+ tool_name=tool_object["tool_name"],
+ arguments=tool_object["arguments"],
+ log=tool_string, # type: ignore
+ )
+ print("calling", calling)
if isinstance(calling, ConverterError):
raise calling
diff --git a/src/crewai/utilities/converter.py b/src/crewai/utilities/converter.py
index 7a1664784..368f7607f 100644
--- a/src/crewai/utilities/converter.py
+++ b/src/crewai/utilities/converter.py
@@ -27,8 +27,7 @@ class Converter(OutputConverter):
if self.is_gpt:
return self._create_instructor().to_pydantic()
else:
- llm = self._create_llm()
- return llm.call(
+ return self.llm.call(
[
{"role": "system", "content": self.instructions},
{"role": "user", "content": self.text},
@@ -47,9 +46,8 @@ class Converter(OutputConverter):
if self.is_gpt:
return self._create_instructor().to_json()
else:
- llm = self._create_llm()
return json.dumps(
- llm.call(
+ self.llm.call(
[
{"role": "system", "content": self.instructions},
{"role": "user", "content": self.text},
@@ -61,19 +59,6 @@ class Converter(OutputConverter):
return self.to_json(current_attempt + 1)
return ConverterError(f"Failed to convert text into JSON, error: {e}.")
- def _create_llm(self):
- """Create an LLM instance."""
- if isinstance(self.llm, str):
- return LLM(model=self.llm)
- elif isinstance(self.llm, LLM):
- return self.llm
- else:
- return LLM(
- model=self.llm.model,
- provider=getattr(self.llm, "provider", "litellm"),
- **getattr(self.llm, "llm_kwargs", {}),
- )
-
def _create_instructor(self):
"""Create an instructor."""
from crewai.utilities import InternalInstructor
@@ -93,7 +78,7 @@ class Converter(OutputConverter):
)
parser = CrewPydanticOutputParser(pydantic_object=self.model)
- result = LLM(model=self.llm).call(
+ result = self.llm.call(
[
{"role": "system", "content": self.instructions},
{"role": "user", "content": self.text},
@@ -105,9 +90,9 @@ class Converter(OutputConverter):
def is_gpt(self) -> bool:
"""Return if llm provided is of gpt from openai."""
return (
- "gpt" in str(self.llm).lower()
- or "o1-preview" in str(self.llm).lower()
- or "o1-mini" in str(self.llm).lower()
+ "gpt" in str(self.llm.model).lower()
+ or "o1-preview" in str(self.llm.model).lower()
+ or "o1-mini" in str(self.llm.model).lower()
)
@@ -157,6 +142,7 @@ def handle_partial_json(
converter_cls: Optional[Type[Converter]] = None,
) -> Union[dict, BaseModel, str]:
match = re.search(r"({.*})", result, re.DOTALL)
+ print("handle_partial_json")
if match:
try:
exported_result = model.model_validate_json(match.group(0))
@@ -185,8 +171,11 @@ def convert_with_instructions(
agent: Any,
converter_cls: Optional[Type[Converter]] = None,
) -> Union[dict, BaseModel, str]:
+ print("convert_with_instructions")
llm = agent.function_calling_llm or agent.llm
+ print("llm", llm)
instructions = get_conversion_instructions(model, llm)
+ print("instructions", instructions)
converter = create_converter(
agent=agent,
converter_cls=converter_cls,
@@ -195,10 +184,11 @@ def convert_with_instructions(
model=model,
instructions=instructions,
)
-
+ print("converter", converter)
exported_result = (
converter.to_pydantic() if not is_json_output else converter.to_json()
)
+ print("exported_result", exported_result)
if isinstance(exported_result, ConverterError):
Printer().print(
@@ -218,12 +208,12 @@ def get_conversion_instructions(model: Type[BaseModel], llm: Any) -> str:
return instructions
-def is_gpt(llm: Any) -> bool:
+def is_gpt(llm: LLM) -> bool:
"""Return if llm provided is of gpt from openai."""
return (
- "gpt" in str(llm).lower()
- or "o1-preview" in str(llm).lower()
- or "o1-mini" in str(llm).lower()
+ "gpt" in str(llm.model).lower()
+ or "o1-preview" in str(llm.model).lower()
+ or "o1-mini" in str(llm.model).lower()
)
diff --git a/src/crewai/utilities/evaluators/task_evaluator.py b/src/crewai/utilities/evaluators/task_evaluator.py
index ce19f3630..9206ef4a6 100644
--- a/src/crewai/utilities/evaluators/task_evaluator.py
+++ b/src/crewai/utilities/evaluators/task_evaluator.py
@@ -93,9 +93,9 @@ class TaskEvaluator:
def _is_gpt(self, llm) -> bool:
return (
- "gpt" in str(self.llm).lower()
- or "o1-preview" in str(self.llm).lower()
- or "o1-mini" in str(self.llm).lower()
+ "gpt" in str(self.llm.model).lower()
+ or "o1-preview" in str(self.llm.model).lower()
+ or "o1-mini" in str(self.llm.model).lower()
)
def evaluate_training_data(
diff --git a/src/crewai/utilities/internal_instructor.py b/src/crewai/utilities/internal_instructor.py
index 5a213dcdc..0d9deaa24 100644
--- a/src/crewai/utilities/internal_instructor.py
+++ b/src/crewai/utilities/internal_instructor.py
@@ -42,6 +42,6 @@ class InternalInstructor:
if self.instructions:
messages.append({"role": "system", "content": self.instructions})
model = self._client.chat.completions.create(
- model=self.llm, response_model=self.model, messages=messages
+ model=self.llm.model, response_model=self.model, messages=messages
)
return model