llmsmith.agent.function package

Subpackages

Submodules

llmsmith.agent.function.cohere module

class llmsmith.agent.function.cohere.CohereFunctionAgent(name: str, llm: AsyncClient, llm_options: CohereTextGenOptions, tools: List[CohereTool] = [], max_turns: int = 5)

Bases: Task[str, str]

Agent based on function calling capability of Cohere LLMs. Agent will loop until one of the conditions are met:

  • If the LLM does not choose any tools (function) and simply returns text content in the response.

  • If maximum number of turns are reached in the agent loop.

Parameters:

  • name (str) – The name of the task.

  • llm (cohere.AsyncClient) – An instance of the Cohere client.

  • llm_options (llmsmith.task.textgen.options.cohere.CohereTextGenOptions, optional) – A dictionary of options to pass to the Cohere LLM.

  • tools (llmsmith.agent.tool.cohere.CohereTool) – List of tools (functions) which can be used by the Cohere LLMs. Each tool contains both the declaration and the actual callable.

  • max_turns (str) – Maximum number of turns allowed in the agent loop. Defaults to 5.

Raises:

ValueError – If the name is empty or if max_turns is less than 1.

async execute(task_input: TaskInput[str]) → TaskOutput[str]

Executes the agent by running the loop. The agent exits the loop and returns the LLM response only if no more function calls are required.

Parameters:

task_input (llmsmith.task.models.TaskInput[str]) – The input to the agent task.

Raises:

  • ValueError – If the content of the task input is not a string.

  • PromptBlockedError – If the prompt is blocked by the AI.

  • TextGenFailedError – If AI fails to generate text based on the prompt.

  • MaxTurnsReachedException – If maximum number of turns are reached in the agent loop.

Returns:

The output of the task.

Return type:

llmsmith.task.models.TaskOutput[str]

llmsmith.agent.function.gemini module

class llmsmith.agent.function.gemini.GeminiFunctionAgent(name: str, llm: GenerativeModel, llm_options: GeminiTextGenOptions, tools: List[GeminiTool] = [], max_turns: int = 5)

Bases: Task[str, str]

Agent based on function calling capability of Gemini LLMs. Agent will loop until one of the conditions are met:

  • If the LLM does not choose any tools (function) and simply returns text content in the response.

  • If maximum number of turns are reached in the agent loop.

Parameters:

  • name (str) – The name of the task.

  • llm (google.generativeai.GenerativeModel) – An instance of the Gemini client.

  • llm_options (llmsmith.task.textgen.options.gemini.GeminiTextGenOptions, optional) – A dictionary of options to pass to the Gemini LLM.

  • tools (llmsmith.agent.tool.gemini.GeminiTool) – List of tools (functions) which can be used by the Gemini LLMs. Each tool contains both the declaration and the actual callable.

  • max_turns (str) – Maximum number of turns allowed in the agent loop. Defaults to 5.

Raises:

ValueError – If the name is empty or if max_turns is less than 1.

async execute(task_input: TaskInput[str]) → TaskOutput[str]

Executes the agent by running the loop. The agent exits the loop and returns the LLM response only if no more function calls are required.

Parameters:

task_input (llmsmith.task.models.TaskInput[str]) – The input to the agent task.

Raises:

  • ValueError – If the content of the task input is not a string.

  • PromptBlockedError – If the prompt is blocked by the AI.

  • TextGenFailedError – If AI fails to generate text based on the prompt.

  • MaxTurnsReachedException – If maximum number of turns are reached in the agent loop.

Returns:

The output of the task.

Return type:

llmsmith.task.models.TaskOutput[str]

llmsmith.agent.function.groq module

class llmsmith.agent.function.groq.GroqFunctionAgent(name: str, llm: AsyncGroq, llm_options: GroqTextGenOptions, tools: List[GroqTool] = [], max_turns: int = 5)

Bases: Task[str, str]

Agent based on function calling capability of LLMs in Groq. Agent will loop until one of the conditions are met:

  • If the LLM does not choose any tools (function) and simply returns text content in the response.

  • If maximum number of turns are reached in the agent loop.

Parameters:

  • name (str) – The name of the task.

  • llm (groq.AsyncGroq) – An instance of the async Groq client.

  • llm_options (llmsmith.task.textgen.options.groq.GroqTextGenOptions, optional) – A dictionary of options to pass to the LLM in Groq.

  • tools (llmsmith.agent.tool.groq.GroqTool) – List of tools (functions) which can be used by the LLMs in Groq. Each tool contains both the declaration and the actual callable.

  • max_turns (str) – Maximum number of turns allowed in the agent loop. Defaults to 5.

Raises:

ValueError – If the name is empty or if max_turns is less than 1.

async execute(task_input: TaskInput[str]) → TaskOutput[str]

Executes the agent by running the loop. The agent exits the loop and returns the LLM response only if no more function calls are required.

Parameters:

task_input (llmsmith.task.models.TaskInput[str]) – The input to the agent task.

Raises:

  • ValueError – If the content of the task input is not a string.

  • PromptBlockedError – If the prompt is blocked by the AI.

  • TextGenFailedError – If AI fails to generate text based on the prompt.

  • MaxTurnsReachedException – If maximum number of turns are reached in the agent loop.

Returns:

The output of the task.

Return type:

llmsmith.task.models.TaskOutput[str]

llmsmith.agent.function.openai module

class llmsmith.agent.function.openai.OpenAIFunctionAgent(name: str, llm: AsyncOpenAI, assistant: Assistant, tools: List[OpenAIAssistantTool] = [], max_turns: int = 5)

Bases: Task[str, str]

Agent based on function calling capability of OpenAI LLMs. Agent will loop until one of the conditions are met:

  • If the LLM does not choose any tools and simply returns text content in the response.

  • If maximum number of turns are reached in the agent loop.

Parameters:

  • name (str) – The name of the task.

  • llm (openai.AsyncOpenAI) – An instance of the async OpenAI client.

  • assistant (openai.types.beta.Assistant) – The OpenAI assistant.

  • tools (List[llmsmith.agent.tool.openai.OpenAIAssistantTool]) – List of tools (file search, code interpreter, functions) which can be used by the OpenAI LLMs. Tools of function type should have the actual callable too.

  • max_turns (str) – Maximum number of turns allowed in the agent loop. Defaults to 5.

Raises:

ValueError – If the name is empty or if max_turns is less than 1.

async classmethod create(name: str, llm: AsyncOpenAI, assistant_options: OpenAIAssistantOptions, tools: List[OpenAIAssistantTool] = [], max_turns: int = 5)

Factory method for creating an instance of OpenAIFunctionAgent.

Parameters:

  • name (str) – The name of the task.

  • llm (openai.AsyncOpenAI) – An instance of the async OpenAI client.

  • assistant_options (llmsmith.agent.function.options.openai.OpenAIAssistantOptions, optional) – A dictionary of options to pass to the OpenAI assistant.

  • tools (List[llmsmith.agent.tool.openai.OpenAIAssistantTool]) – List of tools (file search, code interpreter, functions) which can be used by the OpenAI LLMs. Tools of function type should have the actual callable too.

  • max_turns (str) – Maximum number of turns allowed in the agent loop. Defaults to 5.

Raises:

ValueError – If the name is empty or if max_turns is less than 1.

async execute(task_input: TaskInput[str]) → TaskOutput[str]

Executes the agent by running the loop. The agent exits the loop and returns the LLM response only if no more function calls are required.

Parameters:

task_input (llmsmith.task.models.TaskInput[str]) – The input to the agent task.

Raises:

  • ValueError – If the content of the task input is not a string.

  • TextGenFailedError – If AI fails to generate text based on the prompt.

  • MaxTurnsReachedException – If maximum number of turns are reached in the agent loop.

Returns:

The output of the task.

Return type:

llmsmith.task.models.TaskOutput[str]

Module contents