Terminal LLM Chatbot

A python terminal-based chatbot application that integrates with language models. Features include multiline text input, streaming responses and many commands to control the conversation.

Features

• Supports Claude 3.7 (with optional extended thinking), ChatGPT-4o and o3-mini
• Terminal-based user interface with rich text formatting and code highlighting
• Support for multiline text input
• Streaming responses from the language model
• Conversations are saved as text files for portability and easy manipulation
• Anonymous chats that don't get saved anywhere
• You can edit the conversation at any point to add/remove context or edit the system prompt
• Change active model/conversation on the fly
• Embed files from the filesystem or webpage contents by url in your prompts
• Quick temporary prompts that don't pollute the current conversation context
• Optional voice input support (install llm-chat-term[voice])
• Configurable via a YAML config file

Installation

You can install llm-chat-term using uv by running:

uv tool install llm-chat-term

or if you don't have uv installed:

pip install llm-chat-term
# or
pipx install llm-chat-term

If you want voice input support, you will need to have portaudio and ffmpeg installed for your system. Then, install the llm_chat_term[voice] instead of the base llm_chat_term package.

Set up your API key(s)

On first run, a config.yaml with default options is created in your Config dir (e.g. ~/.config/llm_chat_term) Edit it to add your API key(s) and customize other options. Then re-run the application.

Alternatively the application can read your API keys from your env (ANTHROPIC_API_KEY, OPENAI_API_KEY)

Usage

Run the chatbot from the command line:

llm_chat_term

Controls

• Type your message and press Alt(Esc)+Enter to send
• Use Enter to add a new line in your message
• Type :exit or press Ctrl+D to exit the application
• Type :help to view help for available commands
• Press Ctrl+C to interrupt and exit

Commands

• :help Displays help about the commands.
• :info Displays info about this chat.
• :edit, :e Opens the conversation history in $EDITOR (vim by default). Edit it and save and it will be reloaded in the message history. You can edit the system prompt for the current conversation this way.
• :chat Display a menu to select a different chat.
• :model Display a menu to select a different chat.
• :redraw Redraw the whole conversation.
• :tmp {prompt} This prompt won't be saved to the conversation history. Ideal for quick one-off questions in the middle of a large conversation
• :think {prompt} Enable thinking mode only for this question (Claude only).
• :read {path} Embed a text file in the prompt (replaces the line with :read).
• :web {url} Embed a webpage contents in the prompt (replaces the line with :web).
• :exit Exits the application. The conversation is saved if not anonymous chat.

Configuration

This is the initial config.yaml file:

llm:
  models:
    - provider: anthropic
      name: claude-3-7-sonnet-20250219
      temperature: null
    - provider: openai
      name: o3-mini
      temperature: null
    - provider: openai
      name: gpt-4o
      temperature: null
  api_keys:
    - provider: anthropic
      api_key: ""
    - provider: openai
      api_key: ""
  system_prompt:
    "You are a helpful assistant responding to a user's questions in
    a PC terminal application.

    The user is an experienced software engineer, your answers should be concise and
    not repetitive.

    Skip conclusions and summarizations of your answers.

    If the user asks for a change in code, don't return the whole code, just the
    changed segment(s).

    Return your answers in markdown format, and wrap code in ``` blocks.
    "
ui:
  prompt_symbol: ">>> "
  user: user
  assistant: assistant
colors:
  user: cyan
  assistant: grey39
  system: yellow

License

MIT