camel-ai 0.2.69a6__tar.gz → 0.2.69a7__tar.gz

{camel_ai-0.2.69a6 → camel_ai-0.2.69a7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: camel-ai
-Version: 0.2.69a6
+Version: 0.2.69a7
 Summary: Communicative Agents for AI Society Study
 Project-URL: Homepage, https://www.camel-ai.org/
 Project-URL: Repository, https://github.com/camel-ai/camel
@@ -31,6 +31,7 @@ Requires-Dist: arxiv<3,>=2.1.3; extra == 'all'
 Requires-Dist: azure-storage-blob<13,>=12.21.0; extra == 'all'
 Requires-Dist: beautifulsoup4<5,>=4; extra == 'all'
 Requires-Dist: botocore<2,>=1.35.3; extra == 'all'
+Requires-Dist: chromadb<1.0.0,>=0.6.0; extra == 'all'
 Requires-Dist: chunkr-ai>=0.0.50; extra == 'all'
 Requires-Dist: cohere<6,>=5.11.0; extra == 'all'
 Requires-Dist: crawl4ai>=0.3.745; extra == 'all'
@@ -288,6 +289,7 @@ Requires-Dist: wikipedia<2,>=1; extra == 'owl'
 Requires-Dist: xls2xlsx>=0.2.0; extra == 'owl'
 Requires-Dist: yt-dlp<2025,>=2024.11.4; extra == 'owl'
 Provides-Extra: rag
+Requires-Dist: chromadb<1.0.0,>=0.6.0; extra == 'rag'
 Requires-Dist: chunkr-ai>=0.0.50; extra == 'rag'
 Requires-Dist: cohere<6,>=5.11.0; extra == 'rag'
 Requires-Dist: crawl4ai>=0.3.745; extra == 'rag'
@@ -311,6 +313,7 @@ Requires-Dist: scholarly[tor]==1.7.11; extra == 'research-tools'
 Provides-Extra: storage
 Requires-Dist: azure-storage-blob<13,>=12.21.0; extra == 'storage'
 Requires-Dist: botocore<2,>=1.35.3; extra == 'storage'
+Requires-Dist: chromadb<1.0.0,>=0.6.0; extra == 'storage'
 Requires-Dist: faiss-cpu<2,>=1.7.2; extra == 'storage'
 Requires-Dist: google-cloud-storage<3,>=2.18.0; extra == 'storage'
 Requires-Dist: mem0ai>=0.1.73; extra == 'storage'

{camel_ai-0.2.69a6 → camel_ai-0.2.69a7}/camel/__init__.py

@@ -14,7 +14,7 @@
 
 from camel.logger import disable_logging, enable_logging, set_log_level
 
-__version__ = '0.2.69a6'
+__version__ = '0.2.69a7'
 
 __all__ = [
     '__version__',

{camel_ai-0.2.69a6 → camel_ai-0.2.69a7}/camel/agents/chat_agent.py

@@ -406,7 +406,10 @@ class ChatAgent(BaseAgent):
             # List of tuples (platform, type)
             resolved_models_list = []
             for model_spec in model_list:
-                platform, type_ = model_spec[0], model_spec[1]  # type: ignore[index]
+                platform, type_ = (  # type: ignore[index]
+                    model_spec[0],
+                    model_spec[1],
+                )
                 resolved_models_list.append(
                     ModelFactory.create(
                         model_platform=platform, model_type=type_
@@ -846,6 +849,185 @@ class ChatAgent(BaseAgent):
         except ValidationError:
             return False
 
+    def _check_tools_strict_compatibility(self) -> bool:
+        r"""Check if all tools are compatible with OpenAI strict mode.
+
+        Returns:
+            bool: True if all tools are strict mode compatible,
+                False otherwise.
+        """
+        tool_schemas = self._get_full_tool_schemas()
+        for schema in tool_schemas:
+            if not schema.get("function", {}).get("strict", True):
+                return False
+        return True
+
+    def _convert_response_format_to_prompt(
+        self, response_format: Type[BaseModel]
+    ) -> str:
+        r"""Convert a Pydantic response format to a prompt instruction.
+
+        Args:
+            response_format (Type[BaseModel]): The Pydantic model class.
+
+        Returns:
+            str: A prompt instruction requesting the specific format.
+        """
+        try:
+            # Get the JSON schema from the Pydantic model
+            schema = response_format.model_json_schema()
+
+            # Create a prompt based on the schema
+            format_instruction = (
+                "\n\nPlease respond in the following JSON format:\n" "{\n"
+            )
+
+            properties = schema.get("properties", {})
+            for field_name, field_info in properties.items():
+                field_type = field_info.get("type", "string")
+                description = field_info.get("description", "")
+
+                if field_type == "array":
+                    format_instruction += (
+                        f'    "{field_name}": ["array of values"]'
+                    )
+                elif field_type == "object":
+                    format_instruction += f'    "{field_name}": {{"object"}}'
+                elif field_type == "boolean":
+                    format_instruction += f'    "{field_name}": true'
+                elif field_type == "number":
+                    format_instruction += f'    "{field_name}": 0'
+                else:
+                    format_instruction += f'    "{field_name}": "string value"'
+
+                if description:
+                    format_instruction += f'  // {description}'
+
+                # Add comma if not the last item
+                if field_name != list(properties.keys())[-1]:
+                    format_instruction += ","
+                format_instruction += "\n"
+
+            format_instruction += "}"
+            return format_instruction
+
+        except Exception as e:
+            logger.warning(
+                f"Failed to convert response_format to prompt: {e}. "
+                f"Using generic format instruction."
+            )
+            return (
+                "\n\nPlease respond in a structured JSON format "
+                "that matches the requested schema."
+            )
+
+    def _handle_response_format_with_non_strict_tools(
+        self,
+        input_message: Union[BaseMessage, str],
+        response_format: Optional[Type[BaseModel]] = None,
+    ) -> Tuple[Union[BaseMessage, str], Optional[Type[BaseModel]], bool]:
+        r"""Handle response format when tools are not strict mode compatible.
+
+        Args:
+            input_message: The original input message.
+            response_format: The requested response format.
+
+        Returns:
+            Tuple: (modified_message, modified_response_format,
+                   used_prompt_formatting)
+        """
+        if response_format is None:
+            return input_message, response_format, False
+
+        # Check if tools are strict mode compatible
+        if self._check_tools_strict_compatibility():
+            return input_message, response_format, False
+
+        # Tools are not strict compatible, convert to prompt
+        logger.info(
+            "Non-strict tools detected. Converting response_format to "
+            "prompt-based formatting."
+        )
+
+        format_prompt = self._convert_response_format_to_prompt(
+            response_format
+        )
+
+        # Modify the message to include format instruction
+        modified_message: Union[BaseMessage, str]
+        if isinstance(input_message, str):
+            modified_message = input_message + format_prompt
+        else:
+            modified_message = input_message.create_new_instance(
+                input_message.content + format_prompt
+            )
+
+        # Return None for response_format to avoid strict mode conflicts
+        # and True to indicate we used prompt formatting
+        return modified_message, None, True
+
+    def _apply_prompt_based_parsing(
+        self,
+        response: ModelResponse,
+        original_response_format: Type[BaseModel],
+    ) -> None:
+        r"""Apply manual parsing when using prompt-based formatting.
+
+        Args:
+            response: The model response to parse.
+            original_response_format: The original response format class.
+        """
+        for message in response.output_messages:
+            if message.content:
+                try:
+                    # Try to extract JSON from the response content
+                    import json
+                    import re
+
+                    from pydantic import ValidationError
+
+                    # Try to find JSON in the content
+                    content = message.content.strip()
+
+                    # Try direct parsing first
+                    try:
+                        parsed_json = json.loads(content)
+                        message.parsed = (
+                            original_response_format.model_validate(
+                                parsed_json
+                            )
+                        )
+                        continue
+                    except (json.JSONDecodeError, ValidationError):
+                        pass
+
+                    # Try to extract JSON from text
+                    json_pattern = r'\{[^{}]*(?:\{[^{}]*\}[^{}]*)*\}'
+                    json_matches = re.findall(json_pattern, content, re.DOTALL)
+
+                    for json_str in json_matches:
+                        try:
+                            parsed_json = json.loads(json_str)
+                            message.parsed = (
+                                original_response_format.model_validate(
+                                    parsed_json
+                                )
+                            )
+                            # Update content to just the JSON for consistency
+                            message.content = json.dumps(parsed_json)
+                            break
+                        except (json.JSONDecodeError, ValidationError):
+                            continue
+
+                    if not message.parsed:
+                        logger.warning(
+                            f"Failed to parse JSON from response: "
+                            f"{content[:100]}..."
+                        )
+
+                except Exception as e:
+                    logger.warning(f"Error during prompt-based parsing: {e}")
+
     def _format_response_if_needed(
         self,
         response: ModelResponse,
@@ -932,6 +1114,14 @@ class ChatAgent(BaseAgent):
         except ImportError:
             pass  # Langfuse not available
 
+        # Handle response format compatibility with non-strict tools
+        original_response_format = response_format
+        input_message, response_format, used_prompt_formatting = (
+            self._handle_response_format_with_non_strict_tools(
+                input_message, response_format
+            )
+        )
+
         # Convert input message to BaseMessage if necessary
         if isinstance(input_message, str):
             input_message = BaseMessage.make_user_message(
@@ -1014,6 +1204,13 @@ class ChatAgent(BaseAgent):
             break
 
         self._format_response_if_needed(response, response_format)
+
+        # Apply manual parsing if we used prompt-based formatting
+        if used_prompt_formatting and original_response_format:
+            self._apply_prompt_based_parsing(
+                response, original_response_format
+            )
+
         self._record_final_output(response.output_messages)
 
         return self._convert_to_chatagent_response(
@@ -1065,6 +1262,14 @@ class ChatAgent(BaseAgent):
         except ImportError:
             pass  # Langfuse not available
 
+        # Handle response format compatibility with non-strict tools
+        original_response_format = response_format
+        input_message, response_format, used_prompt_formatting = (
+            self._handle_response_format_with_non_strict_tools(
+                input_message, response_format
+            )
+        )
+
         if isinstance(input_message, str):
             input_message = BaseMessage.make_user_message(
                 role_name="User", content=input_message
@@ -1098,6 +1303,11 @@ class ChatAgent(BaseAgent):
             )
             iteration_count += 1
 
+            # Accumulate API token usage
+            self._update_token_usage_tracker(
+                step_token_usage, response.usage_dict
+            )
+
             # Terminate Agent if stop_event is set
             if self.stop_event and self.stop_event.is_set():
                 # Use the _step_terminate to terminate the agent with reason
@@ -1139,13 +1349,14 @@ class ChatAgent(BaseAgent):
             break
 
         await self._aformat_response_if_needed(response, response_format)
-        self._record_final_output(response.output_messages)
 
-        # Create token usage tracker for this step
-        step_token_usage = self._create_token_usage_tracker()
+        # Apply manual parsing if we used prompt-based formatting
+        if used_prompt_formatting and original_response_format:
+            self._apply_prompt_based_parsing(
+                response, original_response_format
+            )
 
-        # Update with response usage
-        self._update_token_usage_tracker(step_token_usage, response.usage_dict)
+        self._record_final_output(response.output_messages)
 
         return self._convert_to_chatagent_response(
             response,
@@ -1924,7 +2135,9 @@ class ChatAgent(BaseAgent):
                 schema for schema in self._external_tool_schemas.values()
             ],
             response_terminators=self.response_terminators,
-            scheduling_strategy=self.model_backend.scheduling_strategy.__name__,
+            scheduling_strategy=(
+                self.model_backend.scheduling_strategy.__name__
+            ),
             max_iteration=self.max_iteration,
             stop_event=self.stop_event,
         )

{camel_ai-0.2.69a6 → camel_ai-0.2.69a7}/camel/memories/context_creators/score_based.py

@@ -155,16 +155,21 @@ class ScoreBasedContextCreator(BaseContextCreator):
         # ======================
         # 6. Truncation Logic with Tool Call Awareness
         # ======================
-        logger.warning(
-            f"Context truncation required "
-            f"({total_tokens} > {self.token_limit}), "
-            f"pruning low-score messages."
-        )
-
         remaining_units = self._truncate_with_tool_call_awareness(
             regular_units, tool_call_groups, system_tokens
         )
 
+        # Log only after truncation is actually performed so that both
+        # the original and the final token counts are visible.
+        tokens_after = system_tokens + sum(
+            u.num_tokens for u in remaining_units
+        )
+        logger.warning(
+            "Context truncation performed: "
+            f"before={total_tokens}, after={tokens_after}, "
+            f"limit={self.token_limit}"
+        )
+
         # ======================
         # 7. Output Assembly
         # ======================

{camel_ai-0.2.69a6 → camel_ai-0.2.69a7}/camel/messages/base.py

@@ -69,7 +69,7 @@ class BaseMessage:
         image_detail (Literal["auto", "low", "high"]): Detail level of the
             images associated with the message. (default: :obj:`auto`)
         video_detail (Literal["auto", "low", "high"]): Detail level of the
-            videos associated with the message. (default: :obj:`low`)
+            videos associated with the message. (default: :obj:`auto`)
         parsed: Optional[Union[Type[BaseModel], dict]]: Optional object which
             is parsed from the content. (default: :obj:`None`)
     """
@@ -82,7 +82,7 @@ class BaseMessage:
     video_bytes: Optional[bytes] = None
     image_list: Optional[List[Image.Image]] = None
     image_detail: Literal["auto", "low", "high"] = "auto"
-    video_detail: Literal["auto", "low", "high"] = "low"
+    video_detail: Literal["auto", "low", "high"] = "auto"
     parsed: Optional[Union[BaseModel, dict]] = None
 
     @classmethod

{camel_ai-0.2.69a6 → camel_ai-0.2.69a7}/camel/societies/workforce/workforce.py

@@ -19,7 +19,7 @@ import time
 import uuid
 from collections import deque
 from enum import Enum
-from typing import Any, Coroutine, Deque, Dict, List, Optional
+from typing import Any, Coroutine, Deque, Dict, List, Optional, Set, Tuple
 
 from colorama import Fore
 
@@ -37,6 +37,7 @@ from camel.societies.workforce.role_playing_worker import RolePlayingWorker
 from camel.societies.workforce.single_agent_worker import SingleAgentWorker
 from camel.societies.workforce.task_channel import TaskChannel
 from camel.societies.workforce.utils import (
+    TaskAssignment,
     TaskAssignResult,
     WorkerConf,
     check_if_running,
@@ -1164,22 +1165,30 @@ class Workforce(BaseNode):
             )
         return info
 
-    def _find_assignee(
-        self,
-        tasks: List[Task],
+    def _get_valid_worker_ids(self) -> set:
+        r"""Get all valid worker IDs from child nodes.
+
+        Returns:
+            set: Set of valid worker IDs that can be assigned tasks.
+        """
+        valid_worker_ids = {child.node_id for child in self._children}
+        return valid_worker_ids
+
+    def _call_coordinator_for_assignment(
+        self, tasks: List[Task], invalid_ids: Optional[List[str]] = None
     ) -> TaskAssignResult:
-        r"""Assigns multiple tasks to worker nodes with the best capabilities.
+        r"""Call coordinator agent to assign tasks with optional validation
+        feedback in the case of invalid worker IDs.
 
-        Parameters:
-            tasks (List[Task]): The tasks to be assigned.
+        Args:
+            tasks (List[Task]): Tasks to assign.
+            invalid_ids (List[str], optional): Invalid worker IDs from previous
+                attempt (if any).
 
         Returns:
-            TaskAssignResult: Assignment result containing task assignments
-                with their dependencies.
+            TaskAssignResult: Assignment result from coordinator.
         """
-        self.coordinator_agent.reset()
-
-        # Format tasks information for the prompt
+        # format tasks information for the prompt
         tasks_info = ""
         for task in tasks:
             tasks_info += f"Task ID: {task.id}\n"
@@ -1188,29 +1197,212 @@ class Workforce(BaseNode):
                 tasks_info += f"Additional Info: {task.additional_info}\n"
             tasks_info += "---\n"
 
-        prompt = ASSIGN_TASK_PROMPT.format(
-            tasks_info=tasks_info,
-            child_nodes_info=self._get_child_nodes_info(),
+        prompt = str(
+            ASSIGN_TASK_PROMPT.format(
+                tasks_info=tasks_info,
+                child_nodes_info=self._get_child_nodes_info(),
+            )
         )
 
-        logger.debug(
-            f"Sending batch assignment request to coordinator "
-            f"for {len(tasks)} tasks."
-        )
+        # add feedback if this is a retry
+        if invalid_ids:
+            valid_worker_ids = list(self._get_valid_worker_ids())
+            feedback = (
+                f"VALIDATION ERROR: The following worker IDs are invalid: "
+                f"{invalid_ids}. "
+                f"VALID WORKER IDS: {valid_worker_ids}. "
+                f"Please reassign ONLY the above tasks using these valid IDs."
+            )
+            prompt = prompt + f"\n\n{feedback}"
 
         response = self.coordinator_agent.step(
             prompt, response_format=TaskAssignResult
         )
+
         if response.msg is None or response.msg.content is None:
             logger.error(
                 "Coordinator agent returned empty response for task assignment"
             )
-            # Return empty result as fallback
             return TaskAssignResult(assignments=[])
 
         result_dict = json.loads(response.msg.content, parse_int=str)
-        task_assign_result = TaskAssignResult(**result_dict)
-        return task_assign_result
+        return TaskAssignResult(**result_dict)
+
+    def _validate_assignments(
+        self, assignments: List[TaskAssignment], valid_ids: Set[str]
+    ) -> Tuple[List[TaskAssignment], List[TaskAssignment]]:
+        r"""Validate task assignments against valid worker IDs.
+
+        Args:
+            assignments (List[TaskAssignment]): Assignments to validate.
+            valid_ids (Set[str]): Set of valid worker IDs.
+
+        Returns:
+            Tuple[List[TaskAssignment], List[TaskAssignment]]:
+                (valid_assignments, invalid_assignments)
+        """
+        valid_assignments: List[TaskAssignment] = []
+        invalid_assignments: List[TaskAssignment] = []
+
+        for assignment in assignments:
+            if assignment.assignee_id in valid_ids:
+                valid_assignments.append(assignment)
+            else:
+                invalid_assignments.append(assignment)
+
+        return valid_assignments, invalid_assignments
+
+    def _handle_task_assignment_fallbacks(self, tasks: List[Task]) -> List:
+        r"""Create new workers for unassigned tasks as fallback.
+
+        Args:
+            tasks (List[Task]): Tasks that need new workers.
+
+        Returns:
+            List[TaskAssignment]: Assignments for newly created workers.
+        """
+        fallback_assignments = []
+
+        for task in tasks:
+            logger.info(f"Creating new worker for unassigned task {task.id}")
+            new_worker = self._create_worker_node_for_task(task)
+
+            assignment = TaskAssignment(
+                task_id=task.id,
+                assignee_id=new_worker.node_id,
+                dependencies=[],
+            )
+            fallback_assignments.append(assignment)
+
+        return fallback_assignments
+
+    def _handle_assignment_retry_and_fallback(
+        self,
+        invalid_assignments: List[TaskAssignment],
+        tasks: List[Task],
+        valid_worker_ids: Set[str],
+    ) -> List[TaskAssignment]:
+        r"""Called if Coordinator agent fails to assign tasks to valid worker
+        IDs. Handles retry assignment and fallback worker creation for invalid
+        assignments.
+
+        Args:
+            invalid_assignments (List[TaskAssignment]): Invalid assignments to
+                retry.
+            tasks (List[Task]): Original tasks list for task lookup.
+            valid_worker_ids (set): Set of valid worker IDs.
+
+        Returns:
+            List[TaskAssignment]: Final assignments for the invalid tasks.
+        """
+        invalid_ids = [a.assignee_id for a in invalid_assignments]
+        invalid_tasks = [
+            task
+            for task in tasks
+            if any(a.task_id == task.id for a in invalid_assignments)
+        ]
+
+        # handle cases where coordinator returned no assignments at all
+        if not invalid_assignments:
+            invalid_tasks = tasks  # all tasks need assignment
+            logger.warning(
+                f"Coordinator returned no assignments. "
+                f"Retrying assignment for all {len(invalid_tasks)} tasks."
+            )
+        else:
+            logger.warning(
+                f"Invalid worker IDs detected: {invalid_ids}. "
+                f"Retrying assignment for {len(invalid_tasks)} tasks."
+            )
+
+        # retry assignment with feedback
+        retry_result = self._call_coordinator_for_assignment(
+            invalid_tasks, invalid_ids
+        )
+        final_assignments = []
+
+        if retry_result.assignments:
+            retry_valid, retry_invalid = self._validate_assignments(
+                retry_result.assignments, valid_worker_ids
+            )
+            final_assignments.extend(retry_valid)
+
+            # collect tasks that are still unassigned for fallback
+            if retry_invalid:
+                unassigned_tasks = [
+                    task
+                    for task in invalid_tasks
+                    if any(a.task_id == task.id for a in retry_invalid)
+                ]
+            else:
+                unassigned_tasks = []
+        else:
+            # retry failed completely, all invalid tasks need fallback
+            logger.warning("Retry assignment failed")
+            unassigned_tasks = invalid_tasks
+
+        # handle fallback for any remaining unassigned tasks
+        if unassigned_tasks:
+            logger.warning(
+                f"Creating fallback workers for {len(unassigned_tasks)} "
+                f"unassigned tasks"
+            )
+            fallback_assignments = self._handle_task_assignment_fallbacks(
+                unassigned_tasks
+            )
+            final_assignments.extend(fallback_assignments)
+
+        return final_assignments
+
+    def _find_assignee(
+        self,
+        tasks: List[Task],
+    ) -> TaskAssignResult:
+        r"""Assigns multiple tasks to worker nodes with the best capabilities.
+
+        Parameters:
+            tasks (List[Task]): The tasks to be assigned.
+
+        Returns:
+            TaskAssignResult: Assignment result containing task assignments
+                with their dependencies.
+        """
+        self.coordinator_agent.reset()
+        valid_worker_ids = self._get_valid_worker_ids()
+
+        logger.debug(
+            f"Sending batch assignment request to coordinator "
+            f"for {len(tasks)} tasks."
+        )
+
+        assignment_result = self._call_coordinator_for_assignment(tasks)
+
+        # validate assignments
+        valid_assignments, invalid_assignments = self._validate_assignments(
+            assignment_result.assignments, valid_worker_ids
+        )
+
+        # check if we have assignments for all tasks
+        assigned_task_ids = {
+            a.task_id for a in valid_assignments + invalid_assignments
+        }
+        unassigned_tasks = [t for t in tasks if t.id not in assigned_task_ids]
+
+        # if all assignments are valid and all tasks are assigned, return early
+        if not invalid_assignments and not unassigned_tasks:
+            return TaskAssignResult(assignments=valid_assignments)
+
+        # handle retry and fallback for
+        # invalid assignments and unassigned tasks
+        all_problem_assignments = invalid_assignments
+        retry_and_fallback_assignments = (
+            self._handle_assignment_retry_and_fallback(
+                all_problem_assignments, tasks, valid_worker_ids
+            )
+        )
+        valid_assignments.extend(retry_and_fallback_assignments)
+
+        return TaskAssignResult(assignments=valid_assignments)
 
     async def _post_task(self, task: Task, assignee_id: str) -> None:
         # Record the start time when a task is posted

{camel_ai-0.2.69a6 → camel_ai-0.2.69a7}/camel/storages/__init__.py

@@ -26,6 +26,7 @@ from .vectordb_storages.base import (
     VectorDBQueryResult,
     VectorRecord,
 )
+from .vectordb_storages.chroma import ChromaStorage
 from .vectordb_storages.faiss import FaissStorage
 from .vectordb_storages.milvus import MilvusStorage
 from .vectordb_storages.oceanbase import OceanBaseStorage
@@ -52,4 +53,5 @@ __all__ = [
     'Mem0Storage',
     'OceanBaseStorage',
     'WeaviateStorage',
+    'ChromaStorage',
 ]

{camel_ai-0.2.69a6 → camel_ai-0.2.69a7}/camel/storages/vectordb_storages/__init__.py

@@ -19,6 +19,7 @@ from .base import (
     VectorDBStatus,
     VectorRecord,
 )
+from .chroma import ChromaStorage
 from .faiss import FaissStorage
 from .milvus import MilvusStorage
 from .oceanbase import OceanBaseStorage
@@ -30,6 +31,7 @@ __all__ = [
     'BaseVectorStorage',
     'VectorDBQuery',
     'VectorDBQueryResult',
+    'ChromaStorage',
     'QdrantStorage',
     'MilvusStorage',
     "TiDBStorage",