AIChat

A customizable AI chat application powered by Flet.

Features

• Supports Diverse Models: Compatible with major LLM providers like OpenAI, Anthropic (Claude), Google (Gemini), and DeepSeek.
• Local Model Support: Run local models using the transformers library (e.g., Phi-4-mini) or models optimized for Apple Silicon using mlx-lm (e.g., Gemma 3, DeepSeek-R1).
• MCP (Model Context Protocol) Ready: Integrate with MCP servers to utilize external tools and resources.
• Easy Customization: Easily add new models and agents.
• Desktop & Web: Run as both a desktop application and a web application.

Setup

1. Install Dependencies:

   uv sync

   • This command installs all necessary dependencies, including those for local models (transformers, mlx-lm based).

2. Configure API Keys: Set the following environment variables according to the models you want to use. You can also manage these in a .env file.

   # Example .env file
   OPENAI_API_KEY="your_openai_api_key"
   ANTHROPIC_API_KEY="your_anthropic_api_key"
   GEMINI_API_KEY="your_gemini_api_key"
   DEEPSEEK_API_KEY="your_deepseek_api_key"
   MISTRAL_API_KEY="your_mistral_api_key" # Required if USE_MISTRAL_OCR is True

Configuration

You can adjust various application settings directly within the aichat/config.py file:

• IS_DEBUG (bool): Enable or disable debug mode. This affects things like the database file used (DEBUG_DB_NAME vs DB_NAME).
• USE_MISTRAL_OCR (bool): Enable or disable OCR functionality using Mistral. Requires a Mistral API key configured in your environment.
• DB_NAME (str): The filename for the main application database.
• DEBUG_DB_NAME (str): The filename for the database used when IS_DEBUG is True.
• USER_NAME (str), USER_AVATAR_COLOR (ft.Colors): Customize the display name and avatar color for the user.
• AGENT_NAME (str), AGENT_AVATAR_COLOR (ft.Colors): Customize the base display name and avatar color for the AI agent. The specific model name is usually appended to this.
• APP_ROLE_NAME (str), APP_ROLE_AVATAR_COLOR (ft.Colors): Customize the display name and avatar color for application messages (e.g., errors).
• MAX_REQUEST_COUNT (int): The maximum number of chained requests allowed during tool use interactions before stopping.

How to Run

Launch the application using the following commands:

# Run as a desktop app
uv run flet run -d aichat/main.py

# Run as a web app
uv run flet run -w aichat/main.py

Supported Models

The following models are supported by default (Model names are the Enum names defined in the Agent files):

• OpenAI (Requires OPENAI_API_KEY):
  • GPT4OMINI (gpt-4o-mini)
  • GPT4O (gpt-4o)
  • O1 (o1)
  • O3MINI (o3-mini)
  • GPT45PREVIEW (gpt-4.5-preview)
• Anthropic (Requires ANTHROPIC_API_KEY):
  • CALUDE35HAIKU (claude-3-5-haiku-latest)
  • CLAUDE37SONNET (claude-3-7-sonnet-latest)
• Google (Requires GEMINI_API_KEY):
  • GEMINI2FLASHLITE (gemini-2.0-flash-lite)
  • GEMINI2FLASH (gemini-2.0-flash)
  • GEMINI25PRO (gemini-2.5-pro-exp-03-25)
• DeepSeek (Requires DEEPSEEK_API_KEY):
  • DEEPSEEKCHAT (deepseek-chat)
  • DEEPSEEKREASONER (deepseek-reasoner)
• Local (Transformers):
  • PHI4MINI (microsoft/Phi-4-mini-instruct)
• Local (MLX) (Apple Silicon Mac):
  • GEMMA3_27B_4BIT (mlx-community/gemma-3-27b-it-4bit)
  • DEEPSEEK_R1_32B_4BIT (mlx-community/DeepSeek-R1-Distill-Qwen-32B-4bit)
  • QWEN3_30B_4BIT (mlx-community/Qwen3-30B-A3B-4bit)
• Dummy: A dummy model for testing purposes.

How to Add New Models

To add a new model, you primarily need to modify two files:

1. Create an Agent: Create an Agent class for the model you want to use. Refer to existing Agent files (e.g., aichat/agents/openai_agent.py) for examples. Use McpHandler if you need to utilize MCP tools.

   # Example: aichat/agents/your_agent.py
   from enum import StrEnum
   from typing import Any, AsyncGenerator
   from .agent import Agent
   from .mcp_handler import McpHandler # If using MCP
   import config
   from models.role import Role
   from models.message import Message
   
   class YourModel(StrEnum):
       MODEL_A = "model-a"
       MODEL_B = "model-b"
   
   class YourAgent(Agent): # Inherit from Agent
       def __init__(self, model: YourModel, mcp_handler: McpHandler | None = None): # Accept MCP handler (optional)
           super().__init__(model, mcp_handler) # Initialize parent class
           self.model = model
           self.role = Role(f"{config.AGENT_NAME} ({self.model})", config.AGENT_AVATAR_COLOR)
           self.streamable = True # Streaming support flag
           # self.client = ... # Initialize API client, etc.
   
       def _construct_request(self, message: Message) -> dict[str, Any]:
           # Convert to model-specific request format
           # ...
           pass
   
       async def request(self, messages: list[Message]) -> list[str]:
           # Handle standard request
           # ...
           pass
   
       async def request_streaming(self, messages: list[Message]) -> AsyncGenerator[str, None]:
           # Handle streaming request (if self.streamable is True)
           # ...
           pass

2. Register the Agent: Register the created Model and Agent in aichat/agents/__init__.py.

   # aichat/agents/__init__.py
   
   # ... (other imports)
   from .your_agent import YourAgent, YourModel # Import your created Agent and Model
   
   all_models = list(
       itertools.chain.from_iterable(
           [
               OpenAIModel,
               GeminiModel,
               ClaudeModel,
               DeepSeekModel,
               LocalModel,
               MLXModel,
               YourModel, # Add your model
               DummyModel,
           ]
       )
   )
   
   # ... (definition of _mcp_handler_instance)
   
   def get_agent_by_model(model: StrEnum) -> Agent:
       """Gets an agent instance based on the model enum."""
       if model in OpenAIModel:
           return OpenAIAgent(model, mcp_handler=_mcp_handler_instance)
       # ... (branches for other models)
       elif model in YourModel: # Add your model
           return YourAgent(model, mcp_handler=_mcp_handler_instance) # Pass handler if using MCP
       elif model in DummyModel:
           return DummyAgent(model)
       else:
           raise ValueError(f"Invalid model: {model}")

MCP (Model Context Protocol)

Utilize MCP servers to integrate with external functionalities like file operations or web browsing.

• Configuration: Create aichat/agents/servers.json based on aichat/agents/servers.json.sample and describe the MCP servers you want to use.
• Usage: Call self.mcp_handler.use_tool(...) within an Agent to use MCP server functionalities. Refer to aichat/agents/mcp_handler.py for details.

Contributing

Please report bugs or suggest features via Issues, and submit code improvements via Pull Requests.

License