llmcli

A command line interface for interacting with various LLM APIs (OpenAI, ChatGLM) with support for assistants and chat history.

Features

• Support for multiple LLM providers (ChatGLM, OpenAI)
• Assistant system with customizable prompts
• Chat history management
• Pipe support for processing file content
• Markdown rendering for responses
• Configurable chat context window

Installation

From Source

Requirements:

• Go 1.19 or higher
• GCC (for SQLite support)

Build Steps:

1. git clone https://github.com/fengwj33/llm_cli.git
2. cd llmcli
3. go mod download
4. go build -o llmcli
5. Optional: sudo mv llmcli /usr/local/bin/

Configuration

Before first run, use llmcli -c to create and edit the configuration file at ~/.config/llm_cli/config.json. This will allow you to add your API keys and configure models and assistants. Example configuration structure:

1. Default Model:

   • Specify which model to use when no model is explicitly selected
   • Example: "default": "gpt4"

2. Models:

   • Configure direct API access to language models
   • Each model entry requires:
     • API: provider name (e.g., "openai", "chatglm")
     • Model: specific model identifier (e.g., "gpt-4", "chatglm-6b")
     • API_KEY: authentication key for the API
   • Used with -m flag for one-off queries without context

3. Assistants:

   • Define specialized chat interfaces with persistent context
   • Each assistant entry includes:
     • model: which model to use (must match a configured model name)
     • prompt: system prompt that defines assistant's behavior
     • chatContextWindow: number of previous exchanges to include
   • Used with -a flag for contextual conversations
   • Maintains chat history for continuous dialogue

Example use case:

• Models: Direct questions like "llmcli -m gpt4 'what is 2+2?'"
• Assistants: Complex tasks like "llmcli -a code_review 'review this function'" where the assistant remembers context and follows a specific prompt

Usage

Basic Commands

• llmcli "hello world" - Use default assistant
• llmcli -m gpt4 "what is golang" - Use specific model
• llmcli -a coding "tell me about channels" - Use specific assistant
• llmcli -c - Edit configuration
• echo "some text" | llmcli - Process text from pipe

Chat History Commands

• llmcli -h assistant_name - Show chat history
• llmcli -h assistant_name 5 - Show last 5 messages
• llmcli --clear assistant_name - Clear chat history

Examples

Using Default Assistant

llmcli "explain what is golang"

Using Specific Model

llmcli -m chatglm "tell me a story"

Using Assistant with Pipe

cat code.go | llmcli -a code_review

Using History

llmcli -h code_review 10

Using Aliases for Quick Access

Add aliases to your shell configuration file (.bashrc, .zshrc, etc.) for faster access:

For example, add these lines to your .zshrc:

alias a1="llmcli -a assistant1"    # Quick access to assistant1
alias a2="llmcli -a code_review"   # Quick access to code review assistant
alias a3="llmcli -a translator"    # Quick access to translator assistant

Then you can simply use:

a1 "hello"              # Chat with assistant1
a2 "review this code"   # Start code review
a3 "translate to zh"    # Use translator

Project Structure

• config/ - Configuration management
• llm/ - Core LLM functionality
  • api/ - API providers implementation
  • assistant.go - Assistant functionality
  • llm.go - Main LLM interface
• utils/ - Utility functions
• tests/ - Unit tests
• main.go - CLI entry point

Development

To add a new LLM provider:

1. Create new provider file in llm/api/
2. Implement the LLMProvider interface
3. Add provider to the Providers map in api.go

License

Licensed under the Apache License, Version 2.0. See LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.