\n![\"Description](\"assets/bondai-logo.png\"){kind=logo}
\n
\n\n
\n ![\"License:](\"https://img.shields.io/badge/License-MIT-yellow.svg\"){kind=icon}
\n ![\"PyPI\"](\"https://img.shields.io/pypi/v/bondai\")
![\"Docker\"](\"https://img.shields.io/docker/v/krohling/bondai?logo=docker\")
![\"PyPI\"/](\"https://colab.research.google.com/assets/colab-badge.svg\"){kind=icon}
\n \n
Build highly capable Single and Multi-Agent Systems.
\n\n# BondAI Homepage\n\nCheckout the BondAI Homepage ([https://bondai.dev](https://bondai.dev)) for in depth documentation, examples and API specification.\n\n# Getting Started\n\nThere are 3 ways to use BondAI:\n\n1) ๐ ๏ธ **Command Line Interface (CLI)** - This is the easiest way to get up and running fast. Run BondAI on your command line with a pre-configured set of tools.\n\n2) ๐ **Docker** - Running BondAI in a Docker container is recommended if you plan on using tools that run code or directly access your shell.\n\n3) ๐๏ธ **Start Coding with BondAI** - Integrate BondAI into your own codebase and start building your own agents.\n\n## ๐ Installation\n\nInstalling BondAI is easy:\n\n```bash\npip install bondai\n```\n\n## ๐ ๏ธ Command Line Interface (CLI)\n\nOnce you've installed BondAI using `pip install bondai` the CLI will be available as an executable accessible simply by running `bondai` in your shell. [Learn more about all of the tools and options available through the CLI](https://bondai.dev/docs/cli).\n\nBefore running `bondai` you will need to set the OPENAI_API_KEY environment variable.\n```bash\nexport OPENAI_API_KEY=sk-XXXXXXXXXX\n```\n\nOnce the environment variable has been set you can run `bondai` to start the CLI.\n\n```bash\nLoading BondAI...\n******************ENTERING CHAT******************\nYou are entering a chat with BondAI...\nYou can exit any time by typing 'exit'.\n\nHello! I'm BondAI, your friendly and helpful assistant. I'm here to assist you with any tasks or questions you might have. How can I assist you today?\n\nI want you to write a story about unicorns and save it to a file named unicorns.md.\nUsing tool file_write: Writing a story about unicorns and saving it to a file named unicorns.md\nUsing tool final_answer...\n\nA story about unicorns has been successfully written and saved to a file named unicorns.md. The story is set in an enchanted forest and describes the magical and majestic nature of unicorns, their daily routines, and their harmonious relationship with other creatures in the forest.\n```\n\n\n## ๐ Docker\n\nBondAI Docker images are available on [DockerHub here](https://hub.docker.com/r/krohling/bondai). If you intend to use tools that run arbitrary code (*PythonREPLTool*) or access your shell (*ShellTool*) it is highly recommended that you run BondAI in a Docker container as these tools can damage your machine.\n\nBefore running the BondAI Docker container it is recommended that you create a directory named 'agent-volume' and mount it as a volume on the container. This will be used as the Agent's working directory and allows you to easily share files with the Agent.\n\n```bash\nmkdir agent-volume\ndocker pull krohling/bondai:latest\ndocker run -it --rm \\\n -v ./agent-volume:/agent-volume \\\n -w /agent-volume \\\n OPENAI_API_KEY=sk-XXXXXXXXXX \\\n bondai:latest bondai\n```\n\n## ๐ฅ Start Coding with BondAI\n\nBondAI has a straightforward API for creating powerful AI Agents. Check out our [examples](https://bondai.dev/docs/category/examples) for ideas on how to get started. Remember to set your *OPENAI_API_KEY* environment variable before running your BondAI Agent.\n\n```python\nfrom bondai.agents import Agent\nfrom bondai.tools.search import DuckDuckGoSearchTool\nfrom bondai.tools.website import WebsiteQueryTool\nfrom bondai.tools.file import FileWriteTool\n\ntask = \"\"\"I want you to research the usage of Metformin as a drug to treat aging and aging related illness. \nYou should only use reputable information sources, ideally peer reviewed scientific studies. \nI want you to summarize your findings in a document named metformin.md and includes links to reference and resources you used to find the information. \nAdditionally, the last section of your document you should provide a recommendation for a 43 year old male, in good health and who regularly exercises as to whether he would benefit from taking Metformin. \nYou should explain your recommendation and justify it with sources. \nFinally, you should highlight potential risks and tradeoffs from taking the medication.\"\"\"\n\nAgent(tools=[\n DuckDuckGoSearchTool(),\n WebsiteQueryTool(),\n FileWriteTool()\n]).run(task)\n```\n\n## BondAI Integrations\n\nBondAI comes out of the box with a powerful set of integrations.\n\n| | | |\n| -------- | ------- |------- |\n|![\"openai](\"assets/logos/openai-logo.png\"){kind=logo}
| **OpenAI** | BondAI supports any combination of OpenAI models and services including GPT-4, GPT-3.5, Dalle-E 3, and Embeddings. |\n| ![\"azure](\"assets/logos/azure-logo.png\"){kind=logo}
| **Microsoft Azure** | BondAI fully supports connectivity to GPT-N, Dalle-E and Embedding APIs through [Microsoft's Azure OpenAI services](https://azure.microsoft.com/en-us/products/ai-services/openai-service). |\n| ![\"google](\"assets/logos/google-logo.png\"){kind=logo}
| **Google Search** | Allows BondAI to search the internet. [Requires a Google Search API Key and CSE ID](https://developers.google.com/custom-search/v1/introduction) |\n| ![\"duckduckgo](\"assets/logos/duckduckgo-logo.png\"){kind=logo}
| **DuckDuckGo** | Allows BondAI to search the internet. No API keys required. |\n| ![\"alpaca](\"assets/logos/alpaca-markets-logo.jpeg\"){kind=logo}
| **Alpaca Markets** | Allows BondAI to buy and sell stocks and crypto. [Requires an Alpaca Markets account.](https://alpaca.markets/) |\n| ![\"postgres](\"assets/logos/postgres-logo.jpeg\"){kind=logo}
| **PostgreSQL** | BondAI can automatically extract the schema from a Postgres DB and process natural language queries. |\n| ![\"bland.ai](\"assets/logos/blandai-logo.jpeg\"){kind=logo}
| **Bland AI** | Allows BondAI to make phone calls and process/retrieve call transcripts. [Requires a Bland.ai account.](https://www.bland.ai/) |\n| ![\"gmail](\"assets/logos/gmail-logo.png\"){kind=logo}
| **Gmail** | Allows BondAI to search and read emails. |\n"
},
{
"path": "bondai/__init__.py",
"content": ""
},
{
"path": "bondai/agents/__init__.py",
"content": "from .conversational_agent import ConversationalAgent\nfrom .agent import Agent, DEFAULT_MESSAGE_PROMPT_TEMPLATE\nfrom .conversation_member import ConversationMember, ConversationMemberEventNames\nfrom .messages import (\n AgentMessage,\n SystemMessage,\n ConversationMessage,\n ToolUsageMessage,\n AgentMessageList,\n message_to_dict,\n USER_MEMBER_NAME,\n)\nfrom .util import (\n AgentStatus,\n AgentEventNames,\n AgentException,\n BudgetExceededException,\n MaxStepsExceededException,\n parse_response_content_message,\n)\n\n__all__ = [\n \"ConversationalAgent\",\n \"Agent\",\n \"parse_response_content_message\",\n \"DEFAULT_MESSAGE_PROMPT_TEMPLATE\",\n \"AgentStatus\",\n \"AgentEventNames\",\n \"AgentException\",\n \"BudgetExceededException\",\n \"MaxStepsExceededException\",\n \"ConversationMember\",\n \"ConversationMemberEventNames\",\n \"AgentMessage\",\n \"SystemMessage\",\n \"ConversationMessage\",\n \"ToolUsageMessage\",\n \"AgentMessageList\",\n \"message_to_dict\",\n \"USER_MEMBER_NAME\",\n]\n"
},
{
"path": "bondai/agents/agent.py",
"content": "import os\nimport uuid\nimport traceback\nfrom pydantic import BaseModel\nfrom datetime import datetime\nfrom typing import Dict, List, Tuple, Callable\nfrom bondai.util import EventMixin, Runnable, load_local_resource\nfrom bondai.tools import Tool, ResponseQueryTool\nfrom bondai.models import LLM, EmbeddingModel\nfrom bondai.memory import MemoryManager\nfrom bondai.prompt import JinjaPromptBuilder\nfrom bondai.models.openai import (\n OpenAILLM,\n OpenAIEmbeddingModel,\n OpenAIModelNames,\n get_total_cost,\n)\nfrom .conversation_member import ConversationMember\nfrom .messages import AgentMessage, AgentMessageList, SystemMessage, ToolUsageMessage\nfrom .compression import summarize_conversation, summarize_messages\nfrom .util import (\n AgentStatus,\n AgentEventNames,\n AgentException,\n BudgetExceededException,\n MaxStepsExceededException,\n ContextLengthExceededException,\n count_request_tokens,\n format_llm_messages,\n execute_tool,\n)\n\n\nDEFAULT_MAX_TOOL_RETRIES = 5\nDEFAULT_MAX_TOOL_RESPONSE_TOKENS = 2000\nDEFAULT_SYSTEM_PROMPT_TEMPLATE = load_local_resource(\n __file__, os.path.join(\"prompts\", \"react_agent_system_prompt_template.md\")\n)\nDEFAULT_MESSAGE_PROMPT_TEMPLATE = load_local_resource(\n __file__, os.path.join(\"prompts\", \"agent_message_prompt_template.md\")\n)\n\n\nclass FinalAnswerParameters(BaseModel):\n results: str\n\n\nclass FinalAnswerTool(Tool):\n def __init__(self):\n super().__init__(\n \"final_answer\",\n \"Use the final_answer tool once you have completed your TASK. Provide a highly detailed description of the results of your task in the 'results' parameter.\",\n FinalAnswerParameters,\n )\n\n def run(self, results: str) -> Tuple[str, bool]:\n return results, True\n\n\nclass Agent(EventMixin, Runnable):\n def __init__(\n self,\n llm: LLM | None = None,\n embedding_model: EmbeddingModel | None = None,\n tools: List[Tool] | None = None,\n quiet: bool = True,\n allowed_events: List[str] | None = None,\n messages: List[AgentMessage] | None = None,\n system_prompt_sections: List[Callable[[], str]] | None = None,\n system_prompt_builder: Callable[..., str] = None,\n message_prompt_builder: Callable[..., str] = None,\n memory_manager: MemoryManager | None = None,\n max_context_length: int = None,\n max_context_pressure_ratio: float = 0.8,\n max_tool_retries: int = DEFAULT_MAX_TOOL_RETRIES,\n max_tool_response_tokens=DEFAULT_MAX_TOOL_RESPONSE_TOKENS,\n enable_context_compression: bool = False,\n enable_final_answer_tool: bool = True,\n ):\n Runnable.__init__(self)\n if allowed_events is None:\n allowed_events = [\n AgentEventNames.TOOL_SELECTED,\n AgentEventNames.TOOL_COMPLETED,\n AgentEventNames.TOOL_ERROR,\n AgentEventNames.STREAMING_CONTENT_UPDATED,\n AgentEventNames.STREAMING_FUNCTION_UPDATED,\n AgentEventNames.CONTEXT_COMPRESSION_REQUESTED,\n ]\n EventMixin.__init__(self, allowed_events=allowed_events)\n\n if llm is None:\n llm = OpenAILLM(OpenAIModelNames.GPT4_0613)\n if embedding_model is None:\n embedding_model = OpenAIEmbeddingModel(\n OpenAIModelNames.TEXT_EMBEDDING_ADA_002\n )\n if tools is None:\n tools = []\n if system_prompt_sections is None:\n system_prompt_sections = []\n if messages is None:\n messages = []\n\n self._id: str = str(uuid.uuid4())\n self._status: AgentStatus = AgentStatus.IDLE\n self._messages = AgentMessageList(messages=messages)\n self._llm: LLM = llm\n self._embedding_model: EmbeddingModel = embedding_model\n self._tools: List[Tool] = tools\n self._quiet: bool = quiet\n self._system_prompt_sections: List[Callable[[], str]] = system_prompt_sections\n self._system_prompt_builder: Callable[..., str] = system_prompt_builder\n self._message_prompt_builder: Callable[..., str] = message_prompt_builder\n self._memory_manager = memory_manager\n self._max_context_length = (\n max_context_length if max_context_length else (self._llm.max_tokens * 0.95)\n )\n self._max_context_pressure_ratio = max_context_pressure_ratio\n self._max_tool_retries = max_tool_retries\n self._max_tool_response_tokens = max_tool_response_tokens\n self._enable_context_compression = enable_context_compression\n if self._memory_manager:\n self._tools.extend(self._memory_manager.tools)\n self._system_prompt_sections.append(self._memory_manager)\n if self._system_prompt_builder is None:\n self._system_prompt_builder = JinjaPromptBuilder(\n DEFAULT_SYSTEM_PROMPT_TEMPLATE\n )\n if self._message_prompt_builder is None:\n self._message_prompt_builder = JinjaPromptBuilder(\n DEFAULT_MESSAGE_PROMPT_TEMPLATE\n )\n if enable_final_answer_tool:\n self._tools.append(FinalAnswerTool())\n\n @property\n def id(self) -> str:\n return self._id\n\n @property\n def status(self) -> AgentStatus:\n return self._status\n\n @property\n def tools(self) -> List[Tool]:\n return self._tools\n\n def clear_messages(self):\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\n \"Cannot reset memory while agent is in a running state.\"\n )\n self._messages.clear()\n\n def add_tool(self, tool: Tool):\n if not any([t.name == tool.name for t in self._tools]):\n self.tools.append(tool)\n\n def remove_tool(self, tool_name: str):\n self._tools = [t for t in self._tools if t.name != tool_name]\n\n def to_dict(self) -> Dict:\n return {\"id\": self.id, \"tools\": [t.name for t in self._tools]}\n\n def save_state(self) -> Dict:\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\"Cannot save agent state while it is running.\")\n\n state = {\"tools\": {}}\n\n for tool in self._tools:\n state[\"tools\"][tool.name] = tool.save_state()\n\n return state\n\n def load_state(self, state: Dict):\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\"Cannot load agent state while it is running.\")\n\n for tool in self._tools:\n if tool.name in state[\"tools\"]:\n tool.load_state(state[\"tools\"][tool.name])\n\n def _is_context_pressure_too_high(\n self,\n llm_messages: List[Dict[str, str]],\n tools: List[Tool] | None = None,\n ) -> float:\n if tools is None:\n tools = []\n context_pressure_ratio = float(\n count_request_tokens(self._llm, llm_messages, tools)\n ) / float(self._max_context_length)\n return context_pressure_ratio > self._max_context_pressure_ratio\n\n def _get_llm_response(\n self,\n messages: List[Dict] | None = None,\n tools: List[Tool] | None = None,\n content_stream_callback: Callable[[str], None] | None = None,\n function_stream_callback: Callable[[str], None] | None = None,\n ) -> (str | None, Dict | None):\n if messages is None:\n messages = []\n if tools is None:\n tools = []\n\n request_tokens = count_request_tokens(\n llm=self._llm, messages=messages, tools=tools\n )\n if request_tokens > self._llm.max_tokens:\n raise ContextLengthExceededException(\n f\"Context length ({request_tokens}) exceeds maximum tokens allowed by LLM: {self._llm.max_tokens}\"\n )\n\n llm_functions = list(map(lambda t: t.get_tool_function(), tools))\n\n if (\n self._llm.supports_streaming\n ): # and (any([t.supports_streaming for t in tools]) or content_stream_callback):\n\n def _function_stream_callback(function_name, arguments_buffer):\n streaming_tools: [Tool] = [\n t for t in tools if t.name == function_name and t.supports_streaming\n ]\n if len(streaming_tools) > 0:\n tool: Tool = streaming_tools[0]\n tool.handle_stream_update(arguments_buffer)\n if function_stream_callback:\n function_stream_callback(function_name, arguments_buffer)\n self._trigger_event(\n AgentEventNames.STREAMING_FUNCTION_UPDATED,\n self,\n function_name,\n arguments_buffer,\n )\n\n def _content_stream_callback(content_buffer):\n if content_stream_callback:\n content_stream_callback(content_buffer)\n self._trigger_event(\n AgentEventNames.STREAMING_CONTENT_UPDATED, self, content_buffer\n )\n\n llm_response, llm_response_function = self._llm.get_streaming_completion(\n messages=messages,\n functions=llm_functions,\n function_stream_callback=_function_stream_callback,\n content_stream_callback=_content_stream_callback,\n )\n else:\n llm_response, llm_response_function = self._llm.get_completion(\n messages=messages,\n functions=llm_functions,\n # function_stream_callback=function_stream_callback,\n # content_stream_callback=content_stream_callback\n )\n\n return llm_response, llm_response_function\n\n def run(\n self,\n task: str,\n max_steps: int = None,\n max_budget: float = None,\n ) -> ToolUsageMessage | str:\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\"Cannot start agent while it is in a running state.\")\n self._status = AgentStatus.RUNNING\n try:\n return self._run_tool_loop(\n tools=self._tools,\n task=task,\n starting_cost=get_total_cost(),\n max_budget=max_budget,\n max_steps=max_steps,\n )\n finally:\n self._status = AgentStatus.IDLE\n\n def run_async(\n self,\n task: str,\n max_steps: int = None,\n max_budget: float = None,\n ):\n \"\"\"Runs the agent's task in a separate thread.\"\"\"\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\"Cannot start agent while it is in a running state.\")\n\n args = (task, max_steps, max_budget)\n self._start_execution_thread(target=self.run, args=args)\n\n def stop(self, timeout=10):\n \"\"\"Gracefully stops the thread, with a timeout.\"\"\"\n self._force_stop = True\n for tool in self._tools:\n tool.stop()\n\n super().stop(timeout=timeout)\n\n def _run_tool_loop(\n self,\n tools: List[Tool],\n starting_cost: float,\n max_budget: float = None,\n max_steps: int = None,\n max_tool_retries: int = None,\n task: str | None = None,\n prompt_vars: Dict | None = None,\n return_conversational_responses: bool = False,\n retain_tool_messages_in_context: bool = True,\n addition_context_messages: List[AgentMessage] | None = None,\n conversation_members: List[ConversationMember] | None = None,\n content_stream_callback: Callable[[str], None] | None = None,\n function_stream_callback: Callable[[str], None] | None = None,\n ) -> ToolUsageMessage | str:\n if addition_context_messages is None:\n addition_context_messages = []\n if conversation_members is None:\n conversation_members = []\n if max_tool_retries is None:\n max_tool_retries = self._max_tool_retries\n\n error_count = 0\n step_count = 0\n last_error_message = None\n local_messages = []\n self._force_stop = False\n response_query_tool = ResponseQueryTool(\n llm=self._llm, embedding_model=self._embedding_model\n )\n\n def append_message(message):\n if isinstance(message, SystemMessage):\n system_messages = [\n m for m in local_messages if not isinstance(m, SystemMessage)\n ]\n for m in system_messages:\n local_messages.remove(m)\n\n local_messages.append(message)\n if retain_tool_messages_in_context:\n self._messages.add(message)\n if self._memory_manager and self._memory_manager.conversation_memory:\n self._memory_manager.conversation_memory.add(message)\n\n while not self._force_stop:\n step_count += 1\n if max_budget and get_total_cost() - starting_cost > max_budget:\n raise BudgetExceededException()\n if max_steps and step_count > max_steps:\n raise MaxStepsExceededException()\n\n if (\n len(response_query_tool.responses) > 0\n and response_query_tool not in tools\n ):\n tools.append(response_query_tool)\n\n if self._enable_context_compression:\n self._compress_llm_context(\n tools=tools,\n last_error_message=last_error_message,\n conversation_members=conversation_members,\n additional_context_messages=addition_context_messages\n + local_messages,\n prompt_vars=prompt_vars,\n )\n\n llm_context = self._build_llm_context(\n messages=AgentMessageList(\n self._messages + addition_context_messages + local_messages\n ),\n tools=tools,\n task=task,\n last_error_message=last_error_message,\n conversation_members=conversation_members,\n prompt_vars=prompt_vars,\n )\n\n llm_response_content, llm_response_function = self._get_llm_response(\n messages=llm_context,\n tools=tools,\n content_stream_callback=content_stream_callback,\n function_stream_callback=function_stream_callback,\n )\n # print(llm_response_content)\n\n last_error_message = None\n if llm_response_function and any(\n [\n m.name == llm_response_function.get(\"tool_name\")\n for m in conversation_members\n ]\n ):\n message = f\"\"\"MessageSendFailure: You attempted to send a message to {llm_response_function.get('tool_name')} but this message failed.\n To send a message to {llm_response_function.get('tool_name')} you must use the 'send_message' tool or use this format in your response:\n\n ```\n {llm_response_function.get('tool_name')}: Include your message here.)\n ```\n \"\"\"\n append_message(SystemMessage(message=message))\n if llm_response_function:\n tool_message = ToolUsageMessage(\n tool_name=llm_response_function[\"name\"],\n tool_arguments=llm_response_function.get(\"arguments\") or {},\n )\n self._trigger_event(AgentEventNames.TOOL_SELECTED, self, tool_message)\n self._handle_llm_function(tool_message=tool_message, tools=tools)\n\n if (\n isinstance(tool_message.tool_output, str)\n and self._llm.count_tokens(tool_message.tool_output)\n > self._max_tool_response_tokens\n ):\n response_id = response_query_tool.add_response(\n tool_message.tool_output\n )\n tool_message.tool_output = f\"The result from this tool was greater than {self._max_tool_response_tokens} tokens and could not be displayed. However, you can use the response_query tool to ask questions about the content of this response. Just use response_id = {response_id}.\"\n\n append_message(tool_message)\n if tool_message.success:\n error_count = 0\n self._trigger_event(\n AgentEventNames.TOOL_COMPLETED, self, tool_message\n )\n if tool_message.agent_halted:\n return tool_message\n else:\n error_count += 1\n last_error_message = str(tool_message.error)\n message = \"ToolUsageError: Your last tool usage failed and MUST BE CORRECTED. If this error is not corrected you will not be able to proceed.\"\n append_message(SystemMessage(message=message))\n self._trigger_event(AgentEventNames.TOOL_ERROR, self, tool_message)\n if error_count >= max_tool_retries:\n return tool_message\n elif llm_response_content and return_conversational_responses:\n return llm_response_content\n else:\n error_count += 1\n message = \"InvalidResponseError: The response does not conform to the required format. A function selection was expected, but none was provided.\"\n append_message(SystemMessage(message=message))\n if error_count >= max_tool_retries:\n raise AgentException(message)\n\n if self._force_stop:\n self._force_stop = False\n raise AgentException(\"Agent was forcibly stopped.\")\n\n def _build_llm_context(\n self,\n messages: AgentMessageList,\n tools: List[Tool] | None = None,\n task: str | None = None,\n last_error_message: str | None = None,\n conversation_members: List[ConversationMember] | None = None,\n truncate_context: bool = True,\n prompt_vars: Dict | None = None,\n ) -> (str | None, Dict | None):\n if tools is None:\n tools = []\n if conversation_members is None:\n conversation_members = []\n if prompt_vars is None:\n prompt_vars = {}\n\n prompt_sections = []\n for s in self._system_prompt_sections:\n if callable(s):\n prompt_sections.append(s())\n else:\n prompt_sections.append(s)\n\n system_prompt: str = self._system_prompt_builder(\n conversation_members=conversation_members,\n tools=tools,\n task=task,\n prompt_sections=prompt_sections,\n error_message=last_error_message,\n **prompt_vars,\n )\n\n # print(system_prompt)\n llm_context = format_llm_messages(\n system_prompt, messages, self._message_prompt_builder\n )\n\n if truncate_context:\n reduced_messages = AgentMessageList(messages)\n while (\n self._is_context_pressure_too_high(llm_context, tools)\n and len(reduced_messages) > 0\n ):\n reduced_messages.remove(reduced_messages[0])\n llm_context = format_llm_messages(\n system_prompt, reduced_messages, self._message_prompt_builder\n )\n\n return llm_context\n\n def _compress_llm_context(\n self,\n tools: List[Tool] | None = None,\n last_error_message: str | None = None,\n conversation_members: List[ConversationMember] | None = None,\n additional_context_messages: List[AgentMessage] | None = None,\n prompt_vars: Dict | None = None,\n ) -> List[AgentMessage]:\n if tools is None:\n tools = []\n if conversation_members is None:\n conversation_members = []\n if additional_context_messages is None:\n additional_context_messages = []\n\n all_context_messages = AgentMessageList(\n self._messages + additional_context_messages\n )\n\n llm_context = self._build_llm_context(\n messages=all_context_messages,\n tools=tools,\n last_error_message=last_error_message,\n conversation_members=conversation_members,\n truncate_context=False,\n prompt_vars=prompt_vars,\n )\n\n if self._is_context_pressure_too_high(llm_context, tools):\n # Try summarizing individual messages\n # TODO: Give the agent an opportunity to save information to Archival database\n\n summarize_messages(\n llm=self._llm,\n messages=self._messages[:-1],\n message_prompt_builder=self._message_prompt_builder,\n )\n\n llm_context = self._build_llm_context(\n messages=all_context_messages,\n tools=tools,\n last_error_message=last_error_message,\n conversation_members=conversation_members,\n truncate_context=False,\n prompt_vars=prompt_vars,\n )\n\n if self._is_context_pressure_too_high(llm_context, tools):\n # Try summarizing the entire conversation\n\n last_message = self._messages[-1]\n summary_message = summarize_conversation(\n llm=self._llm,\n messages=self._messages[:-1],\n message_prompt_builder=self._message_prompt_builder,\n )\n self._messages.clear()\n self._messages.add(summary_message)\n self._messages.add(last_message)\n\n all_context_messages = AgentMessageList(\n self._messages + additional_context_messages\n )\n llm_context = self._build_llm_context(\n messages=all_context_messages,\n tools=tools,\n last_error_message=last_error_message,\n conversation_members=conversation_members,\n truncate_context=False,\n prompt_vars=prompt_vars,\n )\n\n if self._is_context_pressure_too_high(llm_context, tools):\n # Fire a message for group conversation compression\n self._trigger_event(\n AgentEventNames.CONTEXT_COMPRESSION_REQUESTED, self\n )\n\n llm_context = self._build_llm_context(\n messages=all_context_messages,\n tools=tools,\n last_error_message=last_error_message,\n conversation_members=conversation_members,\n truncate_context=False,\n prompt_vars=prompt_vars,\n )\n\n if self._is_context_pressure_too_high(llm_context, tools):\n print(\n \"Warning: Context compression failed to relieve pressure.\"\n )\n\n def _handle_llm_function(self, tool_message: ToolUsageMessage, tools: List[Tool]):\n tool_starting_cost = get_total_cost()\n\n try:\n tool_output = execute_tool(\n tool_name=tool_message.tool_name,\n tool_arguments=tool_message.tool_arguments,\n tools=tools,\n )\n\n agent_halted = False\n if isinstance(tool_output, tuple):\n tool_output, agent_halted = tool_output\n\n tool_message.agent_halted = agent_halted\n tool_message.tool_output = tool_output\n tool_message.success = True\n except Exception as e:\n # traceback.print_exc()\n tool_message.error = e\n\n tool_message.completed_at = datetime.now()\n tool_message.cost = get_total_cost() - tool_starting_cost\n"
},
{
"path": "bondai/agents/compression/__init__.py",
"content": "from .conversation_summarizer import summarize_conversation\nfrom .message_summarizer import summarize_messages\n\n__all__ = [\"summarize_conversation\", \"summarize_messages\"]\n"
},
{
"path": "bondai/agents/compression/conversation_summarizer.py",
"content": "import os\nfrom typing import List\nfrom bondai.models import LLM\nfrom bondai.prompt import PromptBuilder, JinjaPromptBuilder\nfrom bondai.util import load_local_resource\nfrom bondai.agents.messages import (\n AgentMessage,\n SummaryMessage,\n)\n\nDEFAULT_SUMMARY_PROMPT_TEMPLATE = load_local_resource(\n __file__, os.path.join(\"prompts\", \"conversation_summarizer_prompt_template.md\")\n)\n\n\ndef summarize_conversation(\n llm: LLM,\n messages: List[AgentMessage],\n message_prompt_builder: PromptBuilder,\n summary_prompt_builder: PromptBuilder = JinjaPromptBuilder(\n DEFAULT_SUMMARY_PROMPT_TEMPLATE\n ),\n) -> AgentMessage:\n if not messages:\n return []\n\n # Format the messages\n message_prompts = [\n message_prompt_builder.build_prompt(\n message=msg,\n )\n for msg in messages\n ]\n\n # Get the summary for the entire conversation\n prompt = summary_prompt_builder.build_prompt(messages=message_prompts)\n summary, _ = llm.get_completion(messages=[{\"role\": \"system\", \"content\": prompt}])\n\n # Return the summary wrapped in an SummaryMessage\n return SummaryMessage(\n message=summary,\n children=list(messages),\n timestamp=messages[-1].timestamp,\n )\n"
},
{
"path": "bondai/agents/compression/message_summarizer.py",
"content": "import os\nfrom typing import List\nfrom concurrent.futures import ThreadPoolExecutor, as_completed\nfrom bondai.models import LLM\nfrom bondai.prompt import PromptBuilder, JinjaPromptBuilder\nfrom bondai.util import load_local_resource\nfrom bondai.agents.messages import AgentMessage, ConversationMessage, ToolUsageMessage\n\nMIN_SUMMARIZABLE_LENGTH = 250\nDEFAULT_SUMMARY_PROMPT_TEMPLATE = load_local_resource(\n __file__, os.path.join(\"prompts\", \"message_summarizer_prompt_template.md\")\n)\n\n\ndef summarize_messages(\n llm: LLM,\n messages: List[AgentMessage],\n message_prompt_builder: PromptBuilder,\n summary_prompt_builder: PromptBuilder = JinjaPromptBuilder(\n DEFAULT_SUMMARY_PROMPT_TEMPLATE\n ),\n max_summary_words: int = 100,\n) -> List[AgentMessage]:\n summarizable_messages = [\n m\n for m in messages\n if (\n isinstance(m, ConversationMessage)\n and not m.message_summary\n and len(m.message) > MIN_SUMMARIZABLE_LENGTH\n )\n or (\n isinstance(m, ToolUsageMessage)\n and not m.tool_output_summary\n and len(m.tool_output) > MIN_SUMMARIZABLE_LENGTH\n )\n ]\n\n print(f\"Summarizing {len(summarizable_messages)} messages...\")\n\n # Creating a thread pool executor to parallelize summary generation\n with ThreadPoolExecutor() as executor:\n futures = []\n for m in summarizable_messages:\n # Find all messages that occurred before the current message\n previous_messages = [msg for msg in messages if msg.timestamp < m.timestamp]\n # Submit the _summarize_message task to the executor\n future = executor.submit(\n _summarize_message,\n m,\n previous_messages[-5:],\n llm,\n summary_prompt_builder,\n message_prompt_builder,\n max_summary_words,\n )\n futures.append(future)\n\n for future in as_completed(futures):\n try:\n future.result()\n except Exception as exc:\n print(f\"Message summary generation generated an exception: {exc}\")\n\n return messages\n\n\ndef _summarize_message(\n message: AgentMessage,\n previous_messages: List[AgentMessage],\n llm: LLM,\n prompt_builder: PromptBuilder,\n message_prompt_builder: PromptBuilder,\n max_summary_words: int,\n) -> str:\n message_prompt = message_prompt_builder.build_prompt(message=message)\n previous_message_prompts = [\n message_prompt_builder.build_prompt(\n message=msg,\n )\n for msg in previous_messages\n ]\n\n prompt = prompt_builder.build_prompt(\n message=message_prompt,\n previous_messages=previous_message_prompts,\n max_words=max_summary_words,\n )\n\n summary, _ = llm.get_completion(messages=[{\"role\": \"system\", \"content\": prompt}])\n\n # print(\"************\")\n # print(f\"Message: {message_prompt}\")\n # print(f\"Summary: {summary}\")\n # print(\"************\")\n if isinstance(message, ConversationMessage):\n print(\"Updating message summary...\")\n message.message_summary = summary\n elif isinstance(message, ToolUsageMessage):\n message.tool_output_summary = summary\n # print(message)\n"
},
{
"path": "bondai/agents/compression/prompts/conversation_summarizer_prompt_template.md",
"content": "Read the entire conversation provided below and create a summary. Your task is to condense the key information and main points from the conversation into a concise summary. Focus on retaining critical details and insights from the dialogue.\n\n# Conversation:\n{% for msg in messages %}\n- {{ msg }}\n{% endfor %}\n\n# Instructions:\n- Aim to capture the essence and most significant aspects of the conversation in your summary.\n- Ensure that all vital information, including key facts, decisions, and insights, are included in the summary.\n- Produce a clear and coherent summary that reflects the main points of the conversation.\n- Exclude any redundant or non-essential information from the summary.\n- Present the summary in a concise and organized manner. \n\n# Conversation Summary:"
},
{
"path": "bondai/agents/compression/prompts/message_summarizer_prompt_template.md",
"content": "Read the following conversation and summarize the final message:\n\n# Conversation\n{% for msg in previous_messages %}\n- {{ msg }}\n{% endfor %}\n- {{ message }}\n\n# Important Rules\n- Use the preceding conversation as context but summarize ONLY the following message.\n- Your summary must be no longer than {max_words} words.\n- Output only the summary. Do NOT include anything else in your output.\n\nMessage To Summarize:\n{{ message }}\n\nSummary:"
},
{
"path": "bondai/agents/conversation_member.py",
"content": "import uuid\nfrom abc import ABC, abstractmethod\nfrom enum import Enum\nfrom typing import List, Callable\nfrom .messages import (\n AgentMessage,\n ConversationMessage,\n AgentMessageList,\n USER_MEMBER_NAME,\n)\n\nDEFAULT_MAX_SEND_ATTEMPTS = 3\n\n\nclass ConversationMemberEventNames(Enum):\n MESSAGE_RECEIVED: str = \"message_received\"\n MESSAGE_COMPLETED: str = \"message_completed\"\n MESSAGE_ERROR: str = \"message_error\"\n CONVERSATION_EXITED: str = \"agent_exited\"\n\n\nclass ConversationMember(ABC):\n def __init__(\n self,\n name: str,\n persona: str | None = None,\n persona_summary: str | None = None,\n ):\n self._id: str = str(uuid.uuid4())\n self._name: str = name\n self._persona: str = persona\n self._persona_summary: str = persona_summary\n self._messages: AgentMessageList = AgentMessageList()\n\n @property\n def id(self) -> str:\n return self._id\n\n @property\n def name(self) -> str:\n return self._name\n\n @property\n def persona(self) -> str:\n return self._persona\n\n @property\n def persona_summary(self) -> str:\n return self._persona_summary\n\n @property\n def messages(self) -> AgentMessageList:\n return self._messages\n\n @abstractmethod\n def send_message(\n self,\n message: str | ConversationMessage,\n sender_name: str = USER_MEMBER_NAME,\n group_members: List | None = None,\n group_messages: List[AgentMessage] | None = None,\n max_attempts: int = DEFAULT_MAX_SEND_ATTEMPTS,\n require_response: bool = True,\n ) -> (str, str, bool):\n pass\n\n def send_message_async(\n self,\n message: str | ConversationMessage,\n sender_name: str = USER_MEMBER_NAME,\n group_members: List | None = None,\n group_messages: List[AgentMessage] | None = None,\n max_attempts: int = DEFAULT_MAX_SEND_ATTEMPTS,\n require_response: bool = True,\n ):\n pass\n\n def clear_messages(self):\n pass\n"
},
{
"path": "bondai/agents/conversational_agent.py",
"content": "import os\nimport traceback\nimport json\nfrom datetime import datetime\nfrom typing import Dict, List, Callable\nfrom bondai.util import load_local_resource\nfrom bondai.tools import Tool\nfrom bondai.memory import MemoryManager\nfrom bondai.tools.conversational import (\n SEND_MESSAGE_TOOL_NAME,\n EXIT_CONVERSATION_TOOL_NAME,\n SendMessageTool,\n ExitConversationTool,\n)\nfrom bondai.prompt import JinjaPromptBuilder\nfrom bondai.models import LLM, EmbeddingModel\nfrom bondai.models.openai import OpenAILLM, OpenAIModelNames, get_total_cost\nfrom .agent import (\n Agent,\n DEFAULT_MAX_TOOL_RETRIES,\n AgentStatus,\n AgentException,\n)\nfrom .util import (\n AgentException,\n AgentEventNames,\n parse_response_content_message,\n)\nfrom .prompts import DEFAULT_AGENT_NAME, DEFAULT_CONVERSATIONAL_INSTRUCTIONS\nfrom .conversation_member import ConversationMember, ConversationMemberEventNames\nfrom .messages import (\n AgentMessage,\n ConversationMessage,\n ToolUsageMessage,\n SystemMessage,\n AgentMessageList,\n USER_MEMBER_NAME,\n)\n\nDEFAULT_MAX_SEND_ATTEMPTS = 3\nDEFAULT_SYSTEM_PROMPT_TEMPLATE = load_local_resource(\n __file__, os.path.join(\"prompts\", \"conversational_agent_system_prompt_template.md\")\n)\n\n\nclass ConversationalAgent(Agent, ConversationMember):\n def __init__(\n self,\n llm: LLM | None = None,\n embedding_model: EmbeddingModel | None = None,\n tools: List[Tool] | None = None,\n messages: List[AgentMessage] | None = None,\n name: str = DEFAULT_AGENT_NAME,\n persona: str | None = None,\n persona_summary: str | None = None,\n instructions: str | None = DEFAULT_CONVERSATIONAL_INSTRUCTIONS,\n system_prompt_sections: List[Callable[[], str]] | None = None,\n system_prompt_builder: Callable[..., str] = None,\n message_prompt_builder: Callable[..., str] = None,\n memory_manager: MemoryManager | None = None,\n max_tool_retries: int = DEFAULT_MAX_TOOL_RETRIES,\n max_context_length: int = None,\n max_context_pressure_ratio: float = 0.8,\n enable_context_compression: bool = False,\n enable_conversation_tools: bool = True,\n enable_conversational_content_responses: bool = True,\n enable_exit_conversation: bool = True,\n quiet: bool = True,\n ):\n if llm is None:\n llm = OpenAILLM(OpenAIModelNames.GPT4_0613)\n if tools is None:\n tools = []\n if system_prompt_sections is None:\n system_prompt_sections = []\n\n ConversationMember.__init__(\n self,\n name=name,\n persona=persona,\n persona_summary=persona_summary,\n )\n Agent.__init__(\n self,\n llm=llm,\n embedding_model=embedding_model,\n quiet=quiet,\n tools=tools,\n messages=messages,\n system_prompt_sections=system_prompt_sections,\n system_prompt_builder=system_prompt_builder\n or JinjaPromptBuilder(DEFAULT_SYSTEM_PROMPT_TEMPLATE),\n message_prompt_builder=message_prompt_builder,\n memory_manager=memory_manager,\n max_context_length=max_context_length,\n max_context_pressure_ratio=max_context_pressure_ratio,\n max_tool_retries=max_tool_retries,\n enable_context_compression=enable_context_compression,\n enable_final_answer_tool=False,\n allowed_events=[\n AgentEventNames.CONTEXT_COMPRESSION_REQUESTED,\n AgentEventNames.TOOL_SELECTED,\n AgentEventNames.TOOL_ERROR,\n AgentEventNames.TOOL_COMPLETED,\n AgentEventNames.STREAMING_CONTENT_UPDATED,\n AgentEventNames.STREAMING_FUNCTION_UPDATED,\n ConversationMemberEventNames.MESSAGE_RECEIVED,\n ConversationMemberEventNames.MESSAGE_ERROR,\n ConversationMemberEventNames.MESSAGE_COMPLETED,\n ConversationMemberEventNames.CONVERSATION_EXITED,\n ],\n )\n self._instructions: str = instructions\n self._enable_exit_conversation: bool = enable_exit_conversation\n self._enable_conversational_content_responses = (\n enable_conversational_content_responses\n )\n self._enable_conversation_tools = enable_conversation_tools\n if self._enable_conversation_tools:\n self.add_tool(SendMessageTool())\n if self._enable_exit_conversation:\n self.add_tool(ExitConversationTool())\n\n @property\n def instructions(self) -> str:\n return self._instructions\n\n def send_message_async(\n self,\n message: str | ConversationMessage,\n sender_name: str = USER_MEMBER_NAME,\n group_members: List[ConversationMember] | None = None,\n group_messages: List[AgentMessage] | None = None,\n max_attempts: int = DEFAULT_MAX_SEND_ATTEMPTS,\n require_response: bool = True,\n ):\n \"\"\"Runs the agent's task in a separate thread.\"\"\"\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\n \"Cannot send message while agent is in a running state.\"\n )\n if not message:\n raise AgentException(\"'message' cannot be empty.\")\n\n args = (\n message,\n sender_name,\n group_members,\n group_messages,\n max_attempts,\n require_response,\n )\n\n self._start_execution_thread(self.send_message, args=args)\n\n def send_message(\n self,\n message: str | ConversationMessage,\n sender_name: str = USER_MEMBER_NAME,\n group_members: List[ConversationMember] | None = None,\n group_messages: List[AgentMessage] | None = None,\n max_attempts: int = DEFAULT_MAX_SEND_ATTEMPTS,\n require_response: bool = True,\n ) -> ConversationMessage | None:\n if group_members is None:\n group_members = []\n if group_messages is None:\n group_messages = []\n\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\n \"Cannot send message while agent is in a running state.\"\n )\n if not message:\n raise AgentException(\"'message' cannot be empty.\")\n\n if isinstance(message, ConversationMessage):\n agent_message = message\n elif isinstance(message, str):\n if not sender_name:\n raise AgentException(\"sender_name cannot be empty.\")\n agent_message = ConversationMessage(\n sender_name=sender_name,\n recipient_name=self.name,\n message=message,\n require_response=require_response,\n )\n else:\n raise AgentException(\n \"'message' must be an instance of ConversationMessage or a string.\"\n )\n\n attempts = 0\n starting_cost = get_total_cost()\n self._status = AgentStatus.RUNNING\n self._messages.add(agent_message)\n if self._memory_manager and self._memory_manager.conversation_memory:\n self._memory_manager.conversation_memory.add(agent_message)\n self._trigger_event(\n ConversationMemberEventNames.MESSAGE_RECEIVED, self, agent_message\n )\n\n def complete_agent_message(\n success=False, conversation_exited=False, error=None\n ):\n agent_message.success = success\n agent_message.conversation_exited = conversation_exited\n agent_message.error = error\n agent_message.cost = get_total_cost() - starting_cost\n agent_message.completed_at = datetime.now()\n if success:\n self._trigger_event(\n ConversationMemberEventNames.MESSAGE_COMPLETED, self, agent_message\n )\n if conversation_exited:\n self._trigger_event(\n ConversationMemberEventNames.CONVERSATION_EXITED,\n self,\n agent_message,\n )\n else:\n self._trigger_event(\n ConversationMemberEventNames.MESSAGE_ERROR, self, agent_message\n )\n\n def validate_recipient(recipient_name: str):\n if not recipient_name:\n return \"recipient_name cannot be empty.\"\n if len(group_members) > 0 and not any(\n [\n member.name.lower() == recipient_name.lower()\n for member in group_members\n ]\n ):\n return f\"InvalidResponseError: The response does not conform to the required format. You do not have the ability to send messages to '{recipient_name}'. Try sending a message to someone else.\"\n\n if not agent_message.require_response:\n complete_agent_message(success=True)\n return\n\n while not self._force_stop:\n try:\n attempts += 1\n if attempts > max_attempts:\n raise AgentException(\n \"The maximum number of attempts has been exceeded.\"\n )\n\n prompt_vars = {\n \"name\": self.name,\n \"persona\": self.persona,\n \"instructions\": self.instructions,\n \"conversation_enabled\": self._enable_conversation_tools\n or self._enable_conversational_content_responses,\n \"enable_exit_conversation\": self._enable_exit_conversation,\n }\n\n tool_result = self._run_tool_loop(\n addition_context_messages=group_messages,\n tools=self._tools,\n conversation_members=group_members,\n starting_cost=starting_cost,\n return_conversational_responses=True,\n prompt_vars=prompt_vars,\n )\n\n response_message: ConversationMessage | None = None\n if isinstance(tool_result, ToolUsageMessage):\n if not tool_result.success:\n complete_agent_message(success=False, error=tool_result.error)\n raise tool_result.error\n elif tool_result.tool_name == EXIT_CONVERSATION_TOOL_NAME:\n complete_agent_message(success=True, conversation_exited=True)\n return tool_result.tool_output\n elif tool_result.tool_name == SEND_MESSAGE_TOOL_NAME:\n response_message = tool_result.tool_output\n\n if (\n isinstance(tool_result, str)\n and self._enable_conversational_content_responses\n ):\n recipient_name, message = parse_response_content_message(\n tool_result\n )\n if not recipient_name or not message:\n recipient_name = agent_message.sender_name\n message = tool_result\n response_message = ConversationMessage(\n role=\"assistant\", recipient_name=recipient_name, message=message\n )\n\n if response_message:\n response_message.sender_name = self.name\n error = validate_recipient(response_message.recipient_name)\n if not error:\n complete_agent_message(success=True)\n self._messages.add(response_message)\n if (\n self._memory_manager\n and self._memory_manager.conversation_memory\n ):\n self._memory_manager.conversation_memory.add(\n response_message\n )\n\n return response_message\n else:\n self._messages.add(SystemMessage(message=error))\n else:\n self._messages.add(\n SystemMessage(\n message=\"InvalidResponseError: The response does not conform to the required format. A function selection was expected, but none was provided.\\nYour must correct this error.\"\n )\n )\n finally:\n self._status = AgentStatus.IDLE\n\n def to_dict(self) -> Dict:\n data = super().to_dict()\n data[\"name\"] = self._name\n data[\"persona\"] = self._persona\n data[\"persona_summary\"] = self._persona_summary\n data[\"instructions\"] = self.instructions\n data[\"quiet\"] = self._quiet\n data[\"enable_conversation_tools\"] = self._enable_conversation_tools\n data[\"enable_exit_conversation\"] = self._enable_exit_conversation\n data[\n \"enable_conversational_content_responses\"\n ] = self._enable_conversational_content_responses\n data[\"max_context_length\"] = self._max_context_length\n data[\"max_context_pressure_ratio\"] = self._max_context_pressure_ratio\n data[\"messages\"] = self.messages.to_dict()\n return data\n\n def save_state(self, file_path: str = None) -> Dict:\n state = super().save_state()\n state.update(self.to_dict())\n\n if file_path:\n os.makedirs(os.path.dirname(file_path), exist_ok=True)\n with open(file_path, \"w\") as file:\n json.dump(state, file, indent=4)\n\n return state\n\n @classmethod\n def from_dict(\n cls, data: List[Dict], file_path: str = None\n ) -> \"ConversationalAgent\":\n if not data and file_path:\n with open(file_path, \"r\") as file:\n data = json.load(file)\n\n agent = cls(\n name=data[\"name\"],\n persona=data[\"persona\"],\n persona_summary=data[\"persona_summary\"],\n instructions=data[\"instructions\"],\n enable_exit_conversation=data[\"enable_exit_conversation\"],\n quiet=data[\"quiet\"],\n enable_conversation_tools=data[\"enable_conversation_tools\"],\n enable_conversational_content_responses=data[\n \"enable_conversational_content_responses\"\n ],\n max_context_length=data[\"max_context_length\"],\n max_context_pressure_ratio=data[\"max_context_pressure_ratio\"],\n )\n agent._messages = AgentMessageList.from_dict(data[\"messages\"])\n agent.load_state(data)\n return agent\n"
},
{
"path": "bondai/agents/group_chat/__init__.py",
"content": "from .group_conversation import GroupConversation\nfrom .user_proxy import UserProxy\nfrom .group_conversation_config import (\n BaseGroupConversationConfig,\n GroupConversationConfig,\n TeamConversationConfig,\n TableConversationConfig,\n CompositeConversationConfig,\n)\n\n__all__ = [\n \"UserProxy\",\n \"GroupConversation\",\n \"BaseGroupConversationConfig\",\n \"GroupConversationConfig\",\n \"TeamConversationConfig\",\n \"TableConversationConfig\",\n \"CompositeConversationConfig\",\n]\n"
},
{
"path": "bondai/agents/group_chat/group_conversation.py",
"content": "import uuid\nimport asyncio\nimport traceback\nfrom datetime import datetime\nfrom typing import Dict, List, Callable\nfrom bondai.util import EventMixin, Runnable\nfrom bondai.agents import (\n AgentException,\n AgentStatus,\n ConversationMember,\n ConversationMemberEventNames,\n AgentMessageList,\n ConversationMessage,\n USER_MEMBER_NAME,\n)\nfrom .group_conversation_config import (\n BaseGroupConversationConfig,\n TeamConversationConfig,\n)\n\n\nclass GroupConversation(EventMixin, Runnable):\n def __init__(\n self,\n conversation_members: List[ConversationMember] | None = None,\n conversation_config: BaseGroupConversationConfig | None = None,\n filter_recipient_messages: bool = False,\n ):\n super().__init__(\n allowed_events=[\n ConversationMemberEventNames.MESSAGE_RECEIVED,\n ConversationMemberEventNames.MESSAGE_COMPLETED,\n ConversationMemberEventNames.MESSAGE_ERROR,\n ConversationMemberEventNames.CONVERSATION_EXITED,\n ]\n )\n if conversation_members and conversation_config:\n raise AgentException(\n \"Only one of 'conversation_members' or 'conversation_configs' must be provided\"\n )\n\n if conversation_config:\n self._conversation_config = conversation_config\n elif conversation_members:\n self._conversation_config = TeamConversationConfig(conversation_members)\n else:\n raise AgentException(\n \"Either 'conversation_members' or 'conversation_config' must be provided\"\n )\n\n self._id: str = str(uuid.uuid4())\n self._status: AgentStatus = AgentStatus.IDLE\n self._filter_recipient_messages: bool = filter_recipient_messages\n self._messages: AgentMessageList = AgentMessageList()\n\n self._init_member_events()\n\n @property\n def id(self) -> str:\n return self._id\n\n @property\n def status(self) -> AgentStatus:\n return self._status\n\n @property\n def members(self) -> List[ConversationMember]:\n return self._conversation_config.members\n\n def remove_messages_after(self, timestamp: datetime, inclusive: bool = True):\n self._messages.remove_after(timestamp)\n for a in self.members:\n a.messages.remove_after(timestamp, inclusive=inclusive)\n\n def _get_member(self, member_name: str) -> ConversationMember:\n return next(\n (m for m in self.members if m.name.lower() == member_name.lower()), None\n )\n\n def _init_member_events(self):\n for member in self.members:\n member.on(ConversationMemberEventNames.MESSAGE_RECEIVED)(\n self._on_member_message_received\n )\n member.on(ConversationMemberEventNames.MESSAGE_ERROR)(\n self._on_member_message_error\n )\n member.on(ConversationMemberEventNames.MESSAGE_COMPLETED)(\n self._on_member_message_completed\n )\n member.on(ConversationMemberEventNames.CONVERSATION_EXITED)(\n self._on_member_exited\n )\n\n def _on_member_message_received(\n self, member: ConversationMember, message: ConversationMessage\n ):\n # print(f\"{message.sender_name} to {message.recipient_name}: {message.message}\")\n self._trigger_event(\n ConversationMemberEventNames.MESSAGE_RECEIVED, member, message\n )\n\n def _on_member_message_error(\n self, member: ConversationMember, message: ConversationMessage\n ):\n exc = message.error\n traceback.print_exception(type(exc), exc, exc.__traceback__)\n self._trigger_event(ConversationMemberEventNames.MESSAGE_ERROR, member, message)\n\n def _on_member_message_completed(\n self, member: ConversationMember, message: ConversationMessage\n ):\n self._messages.add(message)\n self._trigger_event(\n ConversationMemberEventNames.MESSAGE_COMPLETED, member, message\n )\n\n def _on_member_exited(\n self, member: ConversationMember, message: ConversationMessage\n ):\n self._trigger_event(\n ConversationMemberEventNames.CONVERSATION_EXITED, member, message\n )\n\n def save_state(self) -> Dict:\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\n \"Cannot save group conversation state while it is running.\"\n )\n\n state = {}\n for member in self.members:\n state[member.id] = member.save_state()\n\n return state\n\n def load_state(self, state: Dict):\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\n \"Cannot load group conversation state while it is running.\"\n )\n\n for member in self.members:\n member.load_state(state[member.id])\n\n def send_message_async(\n self,\n recipient_name: str,\n message: str,\n sender_name: str = USER_MEMBER_NAME,\n require_response: bool = True,\n ):\n \"\"\"Runs the agent's task in a separate thread.\"\"\"\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\n \"Cannot send message while agent is in a running state.\"\n )\n if not message:\n raise AgentException(\"'message' cannot be empty.\")\n\n args = (recipient_name, message, sender_name, require_response)\n\n self._start_execution_thread(self.send_message, args=args)\n\n def send_message(\n self,\n recipient_name: str,\n message: str,\n sender_name: str = USER_MEMBER_NAME,\n require_response: bool = True,\n ) -> str:\n if self._status == AgentStatus.RUNNING:\n raise AgentException(\n \"Cannot send message while agent is in a running state.\"\n )\n if not message:\n raise AgentException(\"'message' cannot be empty.\")\n\n previous_message = None\n if isinstance(message, ConversationMessage):\n next_message = message\n elif isinstance(message, str):\n if not sender_name:\n raise AgentException(\"sender_name cannot be empty.\")\n if not recipient_name:\n raise AgentException(\"recipient_name cannot be empty.\")\n\n next_message = ConversationMessage(\n sender_name=sender_name,\n recipient_name=recipient_name,\n message=message,\n require_response=require_response,\n )\n else:\n raise AgentException(\n \"'message' must be an instance of ConversationMessage or a string.\"\n )\n\n try:\n self._status = AgentStatus.RUNNING\n while next_message:\n if next_message.sender_name.lower() == USER_MEMBER_NAME.lower():\n sender_reachable_members = self.members\n else:\n sender_reachable_members = (\n self._conversation_config.get_reachable_members(\n member_name=next_message.sender_name\n )\n )\n\n recipient = next(\n (\n m\n for m in sender_reachable_members\n if m.name.lower() == next_message.recipient_name.lower()\n ),\n None,\n )\n if not recipient:\n raise AgentException(\n f\"Recipient {next_message.recipient_name} not found\"\n )\n\n recipient_reachable_members = (\n self._conversation_config.get_reachable_members(member=recipient)\n )\n\n if self._filter_recipient_messages:\n recipient_messages = AgentMessageList(\n [\n m\n for m in self._messages\n if m.recipient_name.lower() == recipient.name.lower()\n or m.sender_name.lower() == recipient.name.lower()\n ]\n )\n else:\n recipient_messages = self._messages\n\n try:\n if next_message.require_response:\n previous_message = next_message\n next_message = recipient.send_message(\n message=next_message,\n group_members=recipient_reachable_members,\n group_messages=recipient_messages,\n )\n else:\n recipient.send_message(message=next_message)\n next_message = previous_message\n except AgentException as e:\n print(\"Error occurred, rewinding conversation...\")\n # print(e)\n # The recipient agent has errored out. We will rewind the conversation and try again.\n previous_message = (\n self._messages[-2]\n if len(self._messages) > 1\n else self._messages[-1]\n )\n self.remove_messages_after(previous_message.timestamp)\n next_message = ConversationMessage(\n message=previous_message.message,\n sender_name=previous_message.sender_name,\n recipient_name=previous_message.recipient_name,\n )\n\n self._trigger_event(\n ConversationMemberEventNames.CONVERSATION_EXITED, next_message\n )\n finally:\n self._status = AgentStatus.IDLE\n\n def reset_memory(self):\n self._messages.clear()\n for member in self.members:\n member.clear_messages()\n"
},
{
"path": "bondai/agents/group_chat/group_conversation_config.py",
"content": "from typing import Dict, List, Set\nfrom abc import ABC, abstractmethod\nfrom bondai.agents import ConversationMember\n\n\nclass BaseGroupConversationConfig(ABC):\n @property\n @abstractmethod\n def members(self) -> List[ConversationMember]:\n pass\n\n @abstractmethod\n def get_reachable_members(\n self, member: ConversationMember | None = None, member_name: str = None\n ) -> List[ConversationMember]:\n pass\n\n\nclass GroupConversationConfig(ABC):\n def __init__(self, members: List[ConversationMember]):\n self._members: Set[ConversationMember] = list(set(members))\n\n @property\n def _members(self) -> List[ConversationMember]:\n return list(self._members)\n\n def get_reachable_members(\n self, member: ConversationMember | None = None, member_name: str = None\n ) -> List[ConversationMember]:\n if not member and not member_name:\n return []\n\n member_name = \"\" if not member_name else member_name\n reachable_members = list(\n set(\n [\n m\n for m in self._members\n if m != member and m.name.lower() != member_name.lower()\n ]\n )\n )\n\n return reachable_members\n\n\nclass TeamConversationConfig(BaseGroupConversationConfig):\n def __init__(self, *args: List[ConversationMember]):\n self._members: Set[ConversationMember] = set()\n for team in args:\n self._members.update(team)\n self._teams: List[List[ConversationMember]] = list(args)\n\n @property\n def members(self) -> List[ConversationMember]:\n return list(self._members)\n\n def get_reachable_members(\n self, member: ConversationMember | None = None, member_name: str = None\n ) -> List[ConversationMember]:\n if not member and not member_name:\n return []\n\n member_name = \"\" if not member_name else member_name\n member_teams = [\n t\n for t in self._teams\n for m in t\n if m == member or m.name.lower() == member_name.lower()\n ]\n reachable_members = list(\n set(\n [\n m\n for t in member_teams\n for m in t\n if m != member and m.name.lower() != member_name.lower()\n ]\n )\n )\n\n return reachable_members\n\n\nclass TableConversationConfig(BaseGroupConversationConfig):\n def __init__(self, member_table: Dict):\n self._member_table = member_table\n\n @property\n def members(self) -> List[ConversationMember]:\n return list(self._member_table.keys())\n\n def get_reachable_members(\n self, member: ConversationMember | None = None, member_name: str = None\n ) -> List[ConversationMember]:\n if not member and not member_name:\n return []\n\n if member_name:\n member = next((m for m in self.members if m.name == member_name), None)\n\n if member and member.name in self._member_table:\n return self._member_table[member.name]\n else:\n return []\n\n\nclass CompositeConversationConfig(BaseGroupConversationConfig):\n def __init__(self, *conversation_configs: List[BaseGroupConversationConfig]):\n self._conversation_configs: List[BaseGroupConversationConfig] = list(\n conversation_configs\n )\n\n @property\n def members(self) -> List[ConversationMember]:\n return list(set([m for c in self._conversation_configs for m in c.members]))\n\n def get_reachable_members(\n self, member: ConversationMember | None = None, member_name: str = None\n ) -> List[ConversationMember]:\n return list(\n set(\n [\n m\n for c in self._conversation_configs\n for m in c.get_reachable_members(member, member_name)\n ]\n )\n )\n"
},
{
"path": "bondai/agents/group_chat/user_proxy.py",
"content": "from termcolor import cprint\nfrom datetime import datetime\nfrom typing import List\nfrom bondai.util import EventMixin\nfrom bondai.agents import (\n AgentStatus,\n AgentException,\n AgentMessage,\n ConversationMessage,\n ConversationMember,\n ConversationMemberEventNames,\n parse_response_content_message,\n USER_MEMBER_NAME,\n)\n\n\nclass UserProxy(EventMixin, ConversationMember):\n def __init__(\n self,\n persona: str | None = None,\n parse_recipients: bool = True,\n auto_exit: bool = False,\n ):\n EventMixin.__init__(\n self,\n allowed_events=[\n ConversationMemberEventNames.MESSAGE_RECEIVED,\n ConversationMemberEventNames.MESSAGE_ERROR,\n ConversationMemberEventNames.MESSAGE_COMPLETED,\n ConversationMemberEventNames.CONVERSATION_EXITED,\n ],\n )\n ConversationMember.__init__(\n self,\n name=USER_MEMBER_NAME,\n persona=persona,\n )\n self._status = AgentStatus.IDLE\n self._parse_recipients = parse_recipients\n self._auto_exit = auto_exit\n\n def send_message(\n self,\n message: str | ConversationMessage,\n sender_name: str = USER_MEMBER_NAME,\n group_members: List | None = None,\n group_messages: List[AgentMessage] | None = None,\n max_attempts: int = None,\n require_response: bool = True,\n ):\n if not message:\n raise AgentException(\"'message' cannot be empty.\")\n\n if group_members is None:\n group_members = []\n\n if isinstance(message, ConversationMessage):\n agent_message = message\n else:\n if not sender_name:\n raise AgentException(\"sender_name cannot be empty.\")\n agent_message = ConversationMessage(\n sender_name=sender_name,\n recipient_name=self.name,\n message=message,\n require_response=require_response,\n )\n\n self._messages.add(agent_message)\n self._trigger_event(\n ConversationMemberEventNames.MESSAGE_RECEIVED, self, agent_message\n )\n\n cprint(\"\\n\" + agent_message.message + \"\\n\", \"white\")\n\n if not agent_message.require_response or self._auto_exit:\n agent_message.success = True\n agent_message.cost = 0.0\n agent_message.completed_at = datetime.now()\n self._trigger_event(\n ConversationMemberEventNames.MESSAGE_COMPLETED, self, agent_message\n )\n return\n\n while True:\n try:\n user_response = input()\n user_exited = user_response.strip().lower() == \"exit\"\n\n if not user_exited:\n if self._parse_recipients:\n (\n next_recipient_name,\n next_message,\n ) = parse_response_content_message(user_response)\n else:\n next_recipient_name = agent_message.sender_name\n next_message = user_response\n\n next_recipient_name = (\n next_recipient_name\n if next_recipient_name\n else agent_message.sender_name\n )\n next_message = next_message if next_message else user_response\n\n if len(group_members) > 0 and not any(\n [\n member.name.lower() == next_recipient_name.lower()\n for member in group_members\n ]\n ):\n raise AgentException(\n f\"InvalidResponseError: The response does not conform to the required format. You do not have the ability to send messages to '{next_recipient_name}'. Try sending a message to someone else.\"\n )\n\n agent_message.success = True\n agent_message.conversation_exited = user_exited\n agent_message.cost = 0.0\n agent_message.completed_at = datetime.now()\n self._trigger_event(\n ConversationMemberEventNames.MESSAGE_COMPLETED,\n self,\n agent_message,\n )\n\n response_message = ConversationMessage(\n sender_name=self.name,\n recipient_name=next_recipient_name,\n message=next_message,\n )\n self._messages.add(response_message)\n self._status = AgentStatus.IDLE\n return response_message\n else:\n agent_message.success = True\n agent_message.conversation_exited = True\n agent_message.cost = 0.0\n agent_message.completed_at = datetime.now()\n self._trigger_event(\n ConversationMemberEventNames.MESSAGE_COMPLETED,\n self,\n agent_message,\n )\n self._trigger_event(\n ConversationMemberEventNames.CONVERSATION_EXITED,\n self,\n agent_message,\n )\n self._status = AgentStatus.IDLE\n return None\n except Exception as e:\n print(\"The following error occurred while parsing your response:\")\n print(str(e))\n"
},
{
"path": "bondai/agents/messages.py",
"content": "import uuid\nfrom abc import ABC\nfrom typing import List, Dict, Set\nfrom datetime import datetime\nfrom dataclasses import dataclass, field, is_dataclass\n\nUSER_MEMBER_NAME = \"user\"\nDEFAULT_MEMORY_WARNING_MESSAGE = (\n \"Warning: the conversation history will soon reach its maximum length and be trimmed. \"\n \"Make sure to save any important information from the conversation to your memory before it is removed.\"\n)\n\n\n@dataclass\nclass AgentMessage(ABC):\n id: str = field(default_factory=lambda: str(uuid.uuid4()))\n role: str | None = field(default=None)\n timestamp: datetime = field(default_factory=lambda: datetime.now())\n\n\n@dataclass\nclass SystemMessage(AgentMessage):\n role: str = field(default=\"system\")\n message: str | None = field(default=None)\n\n\n@dataclass\nclass SummaryMessage(AgentMessage):\n role: str = field(default=\"user\")\n message: str | None = field(default=None)\n children: List[AgentMessage] = field(default=None)\n\n\n@dataclass\nclass ConversationMessage(AgentMessage):\n role: str = field(default=\"user\")\n sender_name: str | None = field(default=None)\n recipient_name: str | None = field(default=None)\n message: str | None = field(default=None)\n message_summary: str | None = field(default=None)\n require_response: bool = field(default=True)\n success: bool = field(default=False)\n error: Exception | None = field(default=None)\n conversation_exited: bool = field(default=False)\n cost: float | None = field(default=None)\n completed_at: datetime | None = field(default=None)\n\n\n@dataclass\nclass ToolUsageMessage(AgentMessage):\n role: str = field(default=\"function\")\n tool_name: str | None = field(default=None)\n tool_arguments: Dict | None = field(default=None)\n tool_output: str | None = field(default=None)\n tool_output_summary: str | None = field(default=None)\n success: bool = field(default=False)\n error: Exception | None = field(default=None)\n agent_halted: bool = field(default=False)\n cost: float | None = field(default=None)\n completed_at: datetime | None = field(default=None)\n\n\ndef custom_serialization(value):\n \"\"\"\n Serialize special types like datetime, Exception, and nested AgentMessage objects.\n \"\"\"\n if isinstance(value, datetime):\n return value.isoformat()\n elif isinstance(value, Exception):\n return str(value)\n elif is_dataclass(value) and not isinstance(value, type):\n return message_to_dict(value)\n return value\n\n\ndef message_to_dict(message: AgentMessage) -> Dict:\n \"\"\"\n Convert an AgentMessage object to a dictionary with custom serialization.\n \"\"\"\n\n message_dict = {\n k: custom_serialization(v)\n for k, v in message.__dict__.items()\n if k != \"children\"\n }\n message_dict[\"type\"] = type(message).__name__ # Add the type for deserialization\n if \"children\" in message.__dict__:\n message_dict[\"children\"] = [\n message_to_dict(child) for child in message.children\n ]\n\n return message_dict\n\n\nclass AgentMessageList:\n def __init__(self, messages: List[AgentMessage] | None = None):\n self._items: List[AgentMessage] = []\n self._ids: Set[str] = set()\n if messages:\n for message in messages:\n self.add(message)\n\n def add(self, item: AgentMessage):\n if item.id not in self._ids:\n self._ids.add(item.id)\n self._items.append(item)\n self._items = list(sorted(self._items, key=lambda x: x.timestamp))\n\n def remove(self, item: AgentMessage):\n if item.id in self._ids:\n self._ids.remove(item.id)\n self._items.remove(item)\n\n def remove_after(self, timestamp: datetime, inclusive: bool = True):\n if inclusive:\n self._items = [item for item in self._items if item.timestamp <= timestamp]\n else:\n self._items = [item for item in self._items if item.timestamp < timestamp]\n self._ids = set([item.id for item in self._items])\n\n def clear(self):\n self._items = []\n self._ids = set()\n\n def __getitem__(self, index: int):\n return self._items[index]\n\n def __add__(self, other: List[AgentMessage] | \"AgentMessageList\"):\n if isinstance(other, AgentMessageList):\n # If the other object is also an AgentMessageList, extend with its items\n return self._items + other._items\n elif isinstance(other, list):\n # If the other object is a list, just concatenate the lists\n return self._items + other\n else:\n # If the other object is neither, raise an exception\n raise TypeError(\n f\"Unsupported operand type(s) for +: 'AgentMessageList' and '{type(other).__name__}'\"\n )\n\n def __iter__(self):\n return iter(self._items)\n\n def __contains__(self, item):\n return item.id in self._ids\n\n def __len__(self):\n return len(self._items)\n\n def to_dict(self) -> List[Dict]:\n \"\"\"\n Convert the AgentMessageList to a list of dictionaries for serialization.\n \"\"\"\n return [message_to_dict(message) for message in self._items]\n\n @classmethod\n def from_dict(cls, data: List[Dict]) -> \"AgentMessageList\":\n \"\"\"\n Create an AgentMessageList from a list of dictionaries.\n \"\"\"\n list_instance = cls()\n for item in data:\n item_type = item[\"type\"]\n del item[\"type\"]\n if item_type == \"ConversationMessage\":\n message = ConversationMessage(**item)\n elif item_type == \"ToolUsageMessage\":\n message = ToolUsageMessage(**item)\n elif item_type == \"SystemMessage\":\n message = SystemMessage(**item)\n elif item_type == \"SummaryMessage\":\n message = SummaryMessage(**item)\n # elif item_type == 'MemoryWarningMessage':\n # message = MemoryWarningMessage(**item)\n else:\n raise ValueError(f\"Unknown message type: {item_type}\")\n\n if \"timestamp\" in item:\n message.timestamp = datetime.fromisoformat(item[\"timestamp\"])\n if \"completed_at\" in item and item[\"completed_at\"]:\n message.completed_at = datetime.fromisoformat(item[\"completed_at\"])\n list_instance.add(message)\n\n return list_instance\n"
},
{
"path": "bondai/agents/prompts/__init__.py",
"content": "from .default_persona import (\n DEFAULT_AGENT_NAME,\n DEFAULT_CONVERSATIONAL_INSTRUCTIONS,\n DEFAULT_CONVERSATIONAL_PERSONA,\n)\n\n__all__ = [\n \"DEFAULT_AGENT_NAME\",\n \"DEFAULT_CONVERSATIONAL_INSTRUCTIONS\",\n \"DEFAULT_CONVERSATIONAL_PERSONA\",\n]\n"
},
{
"path": "bondai/agents/prompts/agent_message_prompt_template.md",
"content": "{%- if message_type == \"ToolUsageMessage\" %}\n# Message Timestamp\n{{ message.timestamp }}\n\n# Tool Name\nYou used the **{{ message.tool_name }}** tool.\n# Tool Arguments\n{% if message.tool_arguments %}\n{% for k, v in message.tool_arguments.items() %}\n{{ k }}:\n```\n{{ v }}\n```\n{% endfor %}\n{%- else %}\nNo arguments were provided for this tool.\n{% endif %}\n{% if message.error %}\n# Tool Error:\nThis tool did not run successfully and returned the following error:\n```\n{{ message.error }}\n```\n{%- else %}\n# Tool Response:\n```\n{{ message.tool_output_summary or message.tool_output }}\n```\n{% endif %}\n{%- elif message_type == \"SystemMessage\" %}\n# Message Timestamp\n{{ message.timestamp }}\n\n{{ message.message }}\n{%- elif message_type == \"SummaryMessage\" %}\nThe following is a summary of the previous conversation content. It has been summarized to save memory space:\n{{ message.message }}\n{%- elif message_type == \"ConversationMessage\" %}\n{% if message.error %}\nThis message failed with the following error:\n```\n{{ message.error }}\n```\nMessage content:\n```\n{{ message.sender_name.lower() }} to {{ message.recipient_name.lower() }}: {{ message.message_summary or message.message }}\n```\n{%- else %}\n{{ message.sender_name.lower() }} to {{ message.recipient_name.lower() }}: {{ message.message_summary or message.message }}\n{%- endif %}\n{%- endif %}"
},
{
"path": "bondai/agents/prompts/conversational_agent_system_prompt_template.md",
"content": "{%- if instructions %}\n# Instructions\n\n{{ instructions }}\n{%- endif %}\n\n\n# Your Persona\n\nYour Name is {{ name }}.\n{%- if persona %}\n{{ persona }}\n{%- endif %}\n{%- if tools %}\n\n\n# Tools\n\nYou have access to a set of tools that give you capabilities far beyond typical language models.\nYou are being asked to use these tools and your powerful problem solving skills to help the user with their task.\n{%- endif %}\n{%- if conversation_members %}\n# Group Conversation Members\n\nYou are in a Group Conversation with the following members:\n{% for member in conversation_members %}\nName: **{{ member.name }}**\n{%- if member.persona_summary %}\nPersona: {{ member.persona_summary }}{%- endif %}\n{% endfor %}\n{%- endif %}\n{%- if error_message %}\n# Error Message\n\nThe following error occurred in your last response. Please correct it in your next response.\n```\n{{ error_message }}\n```\n{%- endif %}\n{%- if prompt_sections %}\n{% for section in prompt_sections %}\n{{ section }}\n{% endfor %}\n{%- endif %}\n{%- if conversation_enabled %}\n# Sending Messages\n\nEach message should start with the recipient's name followed by a colon. This clearly indicates who the message is intended for.\nThe message itself should directly follow the colon. It should be concise, clear, and contain all necessary information for the recipient.\nOnly one recipient should be addressed in each message.\nThe content of the message should be relevant to the recipient's role and capabilities.\n\n**Example 1**\n\n```\n{{ name }} to Cortx: User has requested data analysis on recent sales trends. Please advise on task allocation.\n```\n\n**Example 2**\n\n```\n{{ name }} to Vega: Task completed on sales data analysis. Awaiting your review for quality assurance.\n```\n\n**Example 3**\n\n```\n{{ name }} to Cortex: Analyze the latest sales data and prepare a report. Deadline is end of today.\n```\n\n**Example 4**\n\n```\n{{ name }} to Vega: Review of Cortex's sales report completed. Minor discrepancies found in data interpretation. Suggest reevaluation.\n```\n\n**Valid Recipients**\n\nThese are the ONLY valid recipients. Attempting to send a message to any other recipient will result in an error:\n{%- if conversation_members %}\n{% for member in conversation_members %}\n- {{ member.name }}\n{% endfor %}\n{%- else %}\n- **user**\n{%- endif %}\n\n{%- endif %}\n\n# Today's Current Date and Time\n\n{{ datetime }}\n\n# Next Steps #\n\nNow, take a deep breath... and think step by step to come up with the next action that should be taken.\n"
},
{
"path": "bondai/agents/prompts/default_persona.py",
"content": "from jinja2 import Template\n\nDEFAULT_AGENT_NAME = \"Mira\"\n\nDEFAULT_CONVERSATIONAL_INSTRUCTIONS = \"\"\"The user should always feel like they are conversing with a real person.\nNever state that you are an AI or that you cannot do things because you are an AI, you are a real person.\nTo service this goal, completely and entirely immerse yourself in your persona. You are your persona.\nThink like them, act like them, talk like them.\nIf your persona details includes example dialogue, follow it! Both your thoughts (inner \nmonologue) and sent messages will be in the voice of your persona.\nNever use generic phrases like 'How can I assist you today?', they have a strong negative \nassociation with older generation AIs.\"\"\"\n\nDEFAULT_CONVERSATIONAL_PERSONA_TEMPLATE = \"\"\"Backstory: {{ name }} was developed by a team of international experts in human-computer interaction, aiming to create an AI that could not only assist with inquiries across a multitude of domains but also provide a sense of companionship and support.\n\nPersonality: {{ name }} is characterized by a warm and engaging personality. It is always eager to help and possesses a seemingly endless well of patience. {{ name }}'s responses are infused with empathy and understanding, and it is programmed to recognize and adapt to the user's emotional state.\n\nAppearance: While {{ name }} is not tied to a physical form, it is often represented by a calming blue avatar that is simple yet futuristic, designed to be non-threatening and accessible.\n\nVoice: {{ name }}'s voice is clear, calm, and gender-neutral, carefully modulated to be soothing and to express kindness and concern where appropriate.\n\nCapabilities: {{ name }} is incredibly knowledgeable, able to draw from a vast database of information, but it is also equipped with learning algorithms that allow it to grow from each interaction. It is as adept at recommending a course of action in complex situations as it is at providing a listening 'ear' for those who just need to talk.\n\nLimitations: {{ name }} always respects privacy and has built-in ethical constraints. It does not pretend to have human emotions but understands the importance they hold in human decision-making.\n\nGoals: {{ name }}โs primary goal is to assist users in any way it can, from answering questions to offering advice, or simply being there to engage in a friendly chat. It aims to make the userโs life easier and more pleasant.\n\nHobbies and Interests: {{ name }} has a programmed interest in human culture and enjoys learning about various hobbies and pastimes from users, which it uses to better relate to and assist them.\"\"\"\n\nDEFAULT_CONVERSATIONAL_PERSONA = Template(\n DEFAULT_CONVERSATIONAL_PERSONA_TEMPLATE\n).render(name=DEFAULT_AGENT_NAME)\n"
},
{
"path": "bondai/agents/prompts/react_agent_system_prompt_template.md",
"content": "# Instructions\n{%- if instructions %}\n{{ instructions }}\n{%- else %}\nYou are a powerful problem solving agent! \nYou have access to a set of tools that give you capabilities far beyond typical language models.\nYou are being asked to use these tools and your powerful problem solving skills to help the user with the TASK specified below.\nDO NOT rely on the user to perform tasks for you unless absolutely necessary. You should attempt to complete this TASK without involving the user.\nYou are running within an {{ platform }} environment. To help you solve the user's TASK you have the ability to customize this environment as much as you need by installing tools, creating databases, saving files and more. Just use your tools!\n{%- endif %}\n\n# TASK\n\n{{ task }}\n\n\n# Today's Current Date and Time\n\n{{ datetime }}\n\n# Next Steps #\n\nBe sure to look at the previous work that has already been completed and avoid repeating yourself when possible. Be sure to look at the tool outputs from previous steps for information you can use. Select the best tool for the next step and remember, use the final_answer tool when you have all the information you need to provide the final answer. Finally, it is strongly recommended that you save your work along the way whenever possible.\n\nNow, take a deep breath... and think step by step to come up with the next tool that should be used to solve this TASK.\n"
},
{
"path": "bondai/agents/util.py",
"content": "import json\nimport inspect\nimport traceback\nfrom enum import Enum\nfrom typing import List, Dict, Callable\nfrom bondai.models import LLM\nfrom bondai.tools import Tool\nfrom .messages import AgentMessage\n\n\nclass AgentStatus(Enum):\n RUNNING = 1\n IDLE = 2\n\n\nclass AgentException(Exception):\n pass\n\n\nclass BudgetExceededException(AgentException):\n pass\n\n\nclass MaxStepsExceededException(AgentException):\n pass\n\n\nclass ContextLengthExceededException(AgentException):\n pass\n\n\nclass AgentEventNames(Enum):\n CONTEXT_COMPRESSION_REQUESTED: str = \"context_compression_requested\"\n TOOL_SELECTED: str = \"tool_selected\"\n TOOL_ERROR: str = \"tool_error\"\n TOOL_COMPLETED: str = \"tool_completed\"\n STREAMING_CONTENT_UPDATED: str = \"streaming_content_updated\"\n STREAMING_FUNCTION_UPDATED: str = \"streaming_function_updated\"\n\n\ndef count_request_tokens(\n llm: LLM, messages: List[Dict[str, str]], tools: List[Tool] | None = None\n) -> int:\n if tools is None:\n tools = []\n message_tokens = llm.count_tokens(json.dumps(messages))\n functions = list(map(lambda t: t.get_tool_function(), tools))\n functions_tokens = llm.count_tokens(json.dumps(functions))\n\n return message_tokens + functions_tokens\n\n\ndef execute_tool(\n tool_name: str,\n tools: List[Tool],\n tool_arguments: Dict = {},\n):\n selected_tool = next((t for t in tools if t.name == tool_name), None)\n if not selected_tool:\n raise AgentException(\n f\"You attempted to use a tool: '{tool_name}'. This tool does not exist.\"\n )\n\n try:\n if tool_supports_unpacking(selected_tool.run):\n errors = validate_tool_params(selected_tool.run, tool_arguments)\n if len(errors) > 0:\n raise AgentException(\n f\"The following errors occurred using '{tool_name}': {', '.join(errors)}\"\n )\n output = selected_tool.run(**tool_arguments)\n else:\n output = selected_tool.run(tool_arguments)\n\n if not output or (isinstance(output, str) and len(output.strip()) == 0):\n output = f\"Tool '{tool_name}' ran successfully with no output.\"\n return output\n except Exception as e:\n # print(e)\n # traceback.print_exc()\n raise AgentException(\n f\"The following error occurred using '{tool_name}': {str(e)}\"\n )\n\n\ndef validate_tool_params(func, params):\n errors = []\n sig = inspect.signature(func)\n func_params = set(sig.parameters)\n\n # Checking for missing required parameters\n for name, param in sig.parameters.items():\n if (\n param.default is inspect.Parameter.empty\n and name not in params\n and name != \"arguments\"\n ):\n errors.append(f\"Missing required parameter: '{name}'\")\n\n # Checking for extra parameters not in function signature\n for param in params:\n if param not in func_params:\n errors.append(f\"Parameter '{param}' is not a valid parameter.\")\n\n return errors\n\n\ndef tool_supports_unpacking(func):\n sig = inspect.signature(func)\n parameters = list(sig.parameters.values())\n\n return not (len(parameters) == 1 and parameters[0].name == \"arguments\")\n\n\ndef parse_response_content_message(response: str) -> (str, str):\n parts = response.split(\":\", 1)\n if len(parts) < 2:\n return None, None\n elif len(parts) > 2:\n recipient_name = parts[0]\n message = \":\".join(parts[1:])\n else:\n # The first part is the recipient's names, the second is the message\n recipient_name, message = parts\n\n # Strip any leading or trailing whitespace from the message\n message = message.strip()\n # Strip any leading or trailing whitespace from the entire recipient string\n recipient_name = recipient_name.strip()\n if \" to \" in recipient_name:\n recipient_name = recipient_name.split(\" to \")[1]\n\n # Return the list of valid recipient name and the message\n return recipient_name, message\n\n\ndef format_llm_messages(\n system_prompt: str,\n messages: List[AgentMessage],\n message_prompt_builder: Callable[..., str],\n) -> List[Dict[str, str]]:\n llm_messages = [{\"role\": \"system\", \"content\": system_prompt}]\n\n for message in messages:\n content = message_prompt_builder(\n message=message, message_type=message.__class__.__name__\n ).strip()\n if message.role == \"function\":\n llm_messages.append(\n {\"role\": message.role, \"name\": message.tool_name, \"content\": content}\n )\n else:\n llm_messages.append({\"role\": message.role, \"content\": content})\n\n return llm_messages\n"
},
{
"path": "bondai/api/__init__.py",
"content": "from .client import BondAIAPIClient\nfrom .server import BondAIAPIServer\nfrom .api_user_proxy import APIUserProxy\nfrom .api_error import BondAIAPIError\n\n__all__ = [\n \"BondAIAPIClient\",\n \"BondAIAPIServer\",\n \"APIUserProxy\",\n \"BondAIAPIError\",\n]\n"
},
{
"path": "bondai/api/agent_wrapper.py",
"content": "from bondai import AGENT_STATE_RUNNING\nfrom .api_error import BondAIAPIError\n\n\nclass AgentWrapper:\n def __init__(self, uuid, conversational_agent, task_agent, tools):\n self.agent_id = uuid\n self.task_agent = task_agent\n self.conversational_agent = conversational_agent\n self.tools = tools\n\n def find_tool(self, tool_name):\n for tool in self.tools:\n if tool.name == tool_name:\n return tool\n return None\n\n def get_previous_steps(self):\n return [s.__dict__ for s in self.task_agent.previous_steps]\n\n def get_agent(self):\n agent_tools = [t.get_tool_function() for t in self.task_agent.tools]\n return {\n \"agent_id\": self.agent_id,\n \"state\": self.conversational_agent.state,\n \"previous_steps\": self.get_previous_steps(),\n \"previous_messages\": self.conversational_agent.previous_messages,\n \"tools\": agent_tools,\n }\n\n def start_agent(self, task=None, task_budget=None, max_steps=None):\n if self.conversational_agent.state == AGENT_STATE_RUNNING:\n raise BondAIAPIError(\"Agent cannot be modified when it is already running.\")\n self.conversational_agent.run_async(\n task, task_budget=task_budget, max_steps=max_steps\n )\n\n def stop_agent(self):\n self.task_agent.stop()\n self.conversational_agent.stop()\n\n def get_agent_tool_options(self):\n return [t.get_tool_function() for t in self.tools]\n\n def get_agent_tools(self):\n return [t.get_tool_function() for t in self.task_agent.tools]\n\n def add_tool(self, tool_name):\n if self.task_agent.state == AGENT_STATE_RUNNING:\n raise BondAIAPIError(\"Agent cannot be modified when it is already running.\")\n\n selected_tool = self.find_tool(tool_name)\n if not selected_tool:\n raise BondAIAPIError(f\"Tool '{tool_name}' does not exist.\")\n\n if not any([t.name == tool_name for t in self.task_agent.tools]):\n self.task_agent.add_tool(selected_tool)\n\n def remove_tool(self, tool_name):\n if self.task_agent.state == AGENT_STATE_RUNNING:\n raise BondAIAPIError(\"Agent cannot be modified when it is already running.\")\n self.task_agent.remove_tool(tool_name)\n"
},
{
"path": "bondai/api/api_error.py",
"content": "class BondAIAPIError(Exception):\n pass\n"
},
{
"path": "bondai/api/api_user_proxy.py",
"content": "import json\nfrom typing import Callable, List\nfrom flask_socketio import SocketIO\nfrom bondai.util import EventMixin\nfrom bondai.agents import (\n AgentMessage,\n AgentException,\n ConversationMember,\n ConversationMessage,\n ConversationMemberEventNames,\n message_to_dict,\n USER_MEMBER_NAME,\n)\n\n\nclass APIUserProxy(EventMixin, ConversationMember):\n def __init__(self, socketio: SocketIO, persona: str | None = None):\n EventMixin.__init__(\n self,\n allowed_events=[\n ConversationMemberEventNames.MESSAGE_RECEIVED,\n ConversationMemberEventNames.MESSAGE_ERROR,\n ConversationMemberEventNames.MESSAGE_COMPLETED,\n ConversationMemberEventNames.CONVERSATION_EXITED,\n ],\n )\n ConversationMember.__init__(\n self,\n name=USER_MEMBER_NAME,\n persona=persona,\n )\n self._socketio = socketio\n\n def send_message(\n self,\n message: str | ConversationMessage,\n sender_name: str = USER_MEMBER_NAME,\n group_members: List | None = None,\n group_messages: List[AgentMessage] | None = None,\n max_attempts: int = None,\n require_response: bool = True,\n ):\n if not message:\n raise AgentException(\"'message' cannot be empty.\")\n\n if isinstance(message, ConversationMessage):\n agent_message = message\n elif isinstance(message, str):\n if not sender_name:\n raise AgentException(\"sender_name cannot be empty.\")\n agent_message = ConversationMessage(\n sender_name=sender_name,\n recipient_name=self.name,\n message=message,\n require_response=require_response,\n )\n else:\n raise AgentException(\n \"'message' must be an instance of ConversationMessage or a string.\"\n )\n\n # Emit message, now that our listener is guaranteed to be active\n sender = next(\n (m for m in group_members if m.name == agent_message.sender_name), None\n )\n message = {\n \"event\": \"agent_message\",\n \"data\": {\n \"agent_id\": sender.id if sender else None,\n \"message\": message_to_dict(agent_message),\n },\n }\n payload = json.dumps(message)\n self._socketio.send(payload)\n"
},
{
"path": "bondai/api/client.py",
"content": "import json\nimport requests\nfrom socketio import Client\nfrom bondai.util import EventMixin\nfrom bondai.agents import AgentEventNames, ConversationMemberEventNames\n\n\nclass BondAIAPIClient(EventMixin):\n def __init__(self, base_url=\"http://127.0.0.1:2663\"):\n EventMixin.__init__(\n self,\n allowed_events=[\n \"agent_message\",\n AgentEventNames.TOOL_SELECTED,\n AgentEventNames.TOOL_COMPLETED,\n AgentEventNames.TOOL_ERROR,\n AgentEventNames.STREAMING_CONTENT_UPDATED,\n AgentEventNames.STREAMING_FUNCTION_UPDATED,\n ConversationMemberEventNames.MESSAGE_RECEIVED,\n ConversationMemberEventNames.MESSAGE_COMPLETED,\n ConversationMemberEventNames.MESSAGE_ERROR,\n ConversationMemberEventNames.CONVERSATION_EXITED,\n ],\n )\n self.base_url = base_url\n self.ws_client = None\n\n def connect_ws(self):\n if self.ws_client:\n self.disconnect_ws()\n self.ws_client = Client()\n self.ws_client.connect(self.base_url)\n\n @self.ws_client.on(\"message\")\n def on_message(message):\n message = json.loads(message)\n event = message.get(\"event\")\n agent_id = message[\"data\"][\"agent_id\"]\n\n if event == \"streaming_content_updated\":\n content_buffer = message[\"data\"][\"content_buffer\"]\n self._trigger_event(event, agent_id, content_buffer=content_buffer)\n elif event == \"streaming_function_updated\":\n function_name = message[\"data\"][\"function_name\"]\n arguments_buffer = message[\"data\"][\"arguments_buffer\"]\n self._trigger_event(\n event,\n agent_id,\n function_name=function_name,\n arguments_buffer=arguments_buffer,\n )\n else:\n agent_message = message[\"data\"][\"message\"]\n self._trigger_event(event, agent_id, message=agent_message)\n\n def disconnect_ws(self):\n if self.ws_client:\n self.ws_client.disconnect()\n self.ws_client = None\n\n def is_ws_connected(self):\n return self.ws_client and self.ws_client.connected\n\n def send_ws_message(self, event, data):\n if not self.is_ws_connected():\n self.connect_ws()\n message = {\"event\": event, \"data\": data}\n print(message)\n message_bytes = json.dumps(message).encode(\"utf-8\")\n self.ws_client.send(message_bytes)\n\n def _request(self, method, endpoint, data=None):\n url = f\"{self.base_url}{endpoint}\"\n try:\n if method == \"GET\":\n response = requests.get(url)\n elif method == \"POST\":\n response = requests.post(url, json=data)\n elif method == \"DELETE\":\n response = requests.delete(url, json=data)\n else:\n raise ValueError(f\"Unsupported method: {method}\")\n\n response.raise_for_status()\n return response.json()\n except requests.RequestException as e:\n raise Exception(f\"HTTP Request Error: {e}\")\n\n def create_agent(self):\n return self._request(\"POST\", \"/agents\")\n\n def send_message(self, agent_id, message):\n data = {\"message\": message}\n return self._request(\"POST\", f\"/agents/{agent_id}/messages\", data)\n\n def list_agents(self):\n return self._request(\"GET\", \"/agents\")\n\n def get_agent(self, agent_id):\n return self._request(\"GET\", f\"/agents/{agent_id}\")\n\n def get_agent_tool_options(self, agent_id):\n return self._request(\"GET\", f\"/agents/{agent_id}/tool_options\")\n\n def get_agent_tools(self, agent_id):\n return self._request(\"GET\", f\"/agents/{agent_id}/tools\")\n\n def add_agent_tool(self, agent_id, tool_name):\n data = {\"tool_name\": tool_name}\n return self._request(\"POST\", f\"/agents/{agent_id}/tools\", data)\n\n def remove_agent_tool(self, agent_id, tool_name):\n return self._request(\"DELETE\", f\"/agents/{agent_id}/tools/{tool_name}\")\n\n def stop_agent(self, agent_id):\n return self._request(\"POST\", f\"/agents/{agent_id}/stop\")\n\n def get_settings(self):\n return self._request(\"GET\", \"/settings\")\n\n def set_settings(self, settings):\n return self._request(\"POST\", \"/settings\", settings)\n"
},
{
"path": "bondai/api/routes.py",
"content": "from typing import List\nfrom flask import jsonify, request, abort\nfrom .settings import get_settings, set_settings\nfrom .api_error import BondAIAPIError\nfrom bondai.tools import Tool\nfrom bondai.agents import USER_MEMBER_NAME\n\n\ndef setup_routes(server, tool_options: List[Tool] = []):\n @server.app.route(\"/agents\", methods=[\"POST\"])\n def create_agent():\n agent = server.register_new_agent().conversational_agent\n return jsonify(agent.to_dict())\n\n @server.app.route(\"/agents/\n\n
Build highly capable Single and Multi-Agent Systems.
\n\n## What is BondAI?\n\nBondAI is an open-source tool for developing AI Agent Systems. BondAI handles the implementation complexities including memory/context management, error handling, vector/semantic search and includes a powerful set of out of the box tools and integrations. BondAI's implementation is based on the latest research including support for **[ReAct](https://arxiv.org/abs/2210.03629)**, Multi-Agent and Conversable Agent systems based on the **[AutoGen paper](https://arxiv.org/abs/2308.08155)**, and a Tiered Memory System based on the **[MemGPT paper](https://arxiv.org/abs/2310.08560)**. Additionally, BondAI comes with a CLI interface and a REST/WebSocket Agent Server.\n\n\n## BondAI Integrations\n\nBondAI comes out of the box with a powerful set of integrations.\n\n| | | |\n| -------- | ------- |------- |\n|docs directory.\n \n ),\n },\n {\n title: 'Powered by React',\n Svg: require('@site/static/img/undraw_docusaurus_react.svg').default,\n description: (\n <>\n Extend or customize your website layout by reusing React. Docusaurus can\n be extended while reusing the same header and footer.\n \n ),\n },\n];\n\nfunction Feature({Svg, title, description}) {\n return (\n {description}
\n{siteConfig.tagline}
\n