Skip to content

Latest commit

 

History

History
240 lines (185 loc) · 8.53 KB

agent_doc.md

File metadata and controls

240 lines (185 loc) · 8.53 KB

Agent.py API Reference

Class: Agent

The Agent class represents an AI agent capable of generating text responses, creating images, and performing text-to-speech operations across multiple AI providers.

Related Modules

This module is a core component of the DeerTick project. For an overview of how it fits into the larger system, see the DeerTick Main Documentation.

Dependencies

  • requests: For making HTTP requests to AI providers
  • json: For parsing JSON responses
  • os: For environment variable access and file operations
  • mistralai: For interacting with the Mistral AI API
  • openai: For interacting with the OpenAI API

Constructor

Agent(model, system_prompt='', provider='', settings=None)
  • model (str): The name or identifier of the AI model to use.
  • system_prompt (str, optional): Initial system prompt for the agent.
  • provider (str, optional): The AI provider (e.g., 'openai', 'openrouter', 'mistral', 'replicate', 'anthropic', 'google', 'huggingface').
  • settings (dict, optional): Additional settings for the agent.

Properties

  • model (str): The current model being used.
  • provider (str): The current AI provider.
  • system_prompt (str): The system prompt for the agent.
  • tools (list): List of available tools for the agent.
  • conversation (dict): Stores the conversation history.
  • content (str): The latest generated content.
  • last_response (list or str): The last response from the AI.
  • nickname (str): A nickname for the agent, defaults to the model name.
  • color (str or None): The color associated with the agent, if any.
  • font (str or None): The font associated with the agent, if any.
  • request (object): Stores the last API request made.
  • response (object): Stores the full response from the last API call.
  • history (bool): Flag to indicate if conversation history should be used.
  • tts_path (str): Path for text-to-speech output files.
  • img_path (str): Path for generated image output files.
  • output_format (str): Format for output files (e.g., 'webp' for images).
  • num_outputs (int): Number of outputs to generate (for image generation).
  • lora_scale (float): Scale factor for LoRA models.
  • aspect_ratio (str): Aspect ratio for generated images.
  • guidance_scale (float): Guidance scale for image generation.
  • num_inference_steps (int): Number of inference steps for image generation.
  • disable_safety_checker (bool): Flag to disable safety checks in image generation.
  • seed (int or None): Seed for random number generation.
  • audio_path (str): Path for audio output files.
  • max_tokens (int): Maximum number of tokens for text generation.
  • min_tokens (int): Minimum number of tokens for text generation.
  • temperature (float): Temperature for text generation.
  • presence_penalty (float): Presence penalty for text generation.
  • frequency_penalty (float): Frequency penalty for text generation.
  • top_k (int): Top-k sampling parameter.
  • top_p (float): Top-p sampling parameter.
  • repetition_penalty (float): Penalty for repetition in text generation.
  • min_p (float): Minimum probability for token selection.
  • top_a (float): Top-a sampling parameter.
  • logit_bias (dict): Bias for logits in text generation.
  • logprobs (bool): Whether to return log probabilities of output tokens.
  • top_logprobs (int or None): Number of top log probabilities to return.
  • response_format (dict or None): Specified format for the response.
  • stop (list): List of stop sequences for text generation.
  • tool_choice (str or None): Specified tool choice for function calling.
  • prompt_template (str): Template for formatting prompts.
  • settings (dict): Stores generation settings in the correct format for the model.
  • tool_template (dict): Template for defining new tools. Structure: 
    {
    "type": "function",
    "function": {
        "description": "",
        "parameters": {
            "type": "object",
            "properties": {},
            "required": [],
            "additionalProperties": False,
        },
    }
}
    This template is used to create new tools with the create_tool method.

Diagram

Agent Class Structure and Workflow

graph TD
    A[Agent] --> B[Constructor]
    A --> C[Properties]
    A --> D[Methods]
    
    B --> B1[Initialize attributes]
    B --> B2[Set up API clients]
    B --> B3[Load configuration]
    
    C --> C1[Model and provider info]
    C --> C2[Conversation history]
    C --> C3[Generation settings]
    C --> C4[Tool templates]
    C --> C5[File paths]
    
    D --> D1[update_provider]
    D --> D2[create_tool]
    D --> D3[generate_response]
    D --> D4[poke]
    D --> D5[tts_poke]
    D --> D6[generate_image]
    D --> D7[tts]
    D --> D8[save_conversation]
    D --> D9[load_conversation]
    D --> D10[help]
    
    D3 --> E1{Provider?}
    E1 -->|OpenAI| F1[OpenAI API call]
    E1 -->|OpenRouter| F2[OpenRouter API call]
    E1 -->|Mistral| F3[Mistral API call]
    E1 -->|Other| F4[Other API calls]
    
    F1 --> G[Process response]
    F2 --> G
    F3 --> G
    F4 --> G
    
    G --> H[Update conversation history]
    H --> I[Return response]
    
    D6 --> J{Provider?}
    J -->|Replicate| K1[Replicate API call]
    J -->|Other| K2[Other image generation]
    
    K1 --> L[Process image URL]
    K2 --> L
    L --> M[Save image file]
    M --> N[Return image URL/path]
    
    D7 --> O[Prepare TTS input]
    O --> P[Replicate TTS API call]
    P --> Q[Process audio URL]
    Q --> R[Save audio file]
    R --> S[Return audio URL/path]
    
    D10 --> T[Display models]
    D10 --> U[Display providers]
Loading

Methods

update_provider(provider)

Updates the provider and model for the agent.

  • provider (str): The name of the provider to use.

create_tool(name, description, parameters)

Creates a new tool for the agent.

  • name (str): The name of the tool.
  • description (str): A brief description of what the tool does.
  • parameters (dict): A dictionary describing the parameters for the tool.

generate_response(system_prompt, prompt)

Generates a response based on the given system prompt and user prompt.

  • system_prompt (str): The system instructions for the AI.
  • prompt (str): The user's input prompt.

Returns: The generated response (str or list).

poke(prompt)

A simplified interface to generate a response.

  • prompt (str): The user's input prompt.

Returns: The generated response (str or list).

tts_poke(prompt, voice)

Generates a text-to-speech response.

  • prompt (str): The text to convert to speech.
  • voice (str, optional): The voice sample to use.

Returns: Path to the generated audio file (str).

generate_image(prompt, file_path=None)

Generates an image based on the given prompt.

  • prompt (str): The description of the image to generate.
  • file_path (str, optional): The path to save the generated image.

Returns: URL or path of the generated image (str).

help()

Provides help information about the agent's capabilities.

Returns: A string containing help information.

Usage Example

agent = Agent(model="gpt-3.5-turbo", provider="openai")
response = agent.poke("Tell me a joke about programming.")
print(response)

image_url = agent.generate_image("A serene landscape with mountains and a lake")
print(f"Generated image: {image_url}")

agent.update_provider("mistral")
mistral_response = agent.poke("Explain quantum computing.")
print(mistral_response)

Notes

  • The Agent class supports multiple AI providers and models, including OpenAI, OpenRouter, Mistral, Replicate, Anthropic, Google, and Hugging Face.
  • The update_provider method allows switching between different AI providers dynamically.
  • Image generation and text-to-speech functionalities depend on the chosen provider and model.
  • The tools feature allows for extending the agent's capabilities with custom functions.
  • The class handles different initialization and API call procedures for each provider.
  • Error handling is implemented for API calls, with detailed error messages printed for debugging.
  • The conversation history is maintained and can be used in generating responses if the history flag is set.
  • The class supports various parameters for fine-tuning text generation, such as temperature, top_k, top_p, and various penalties.
  • API keys are expected to be set as environment variables (e.g., OPENAI_API_KEY, MISTRAL_API_KEY, OPENROUTER_API_KEY).