cgpt is a command-line tool for interacting with Large Language Models (LLMs) using various backends.

• Supports multiple backends: Anthropic, OpenAI, Ollama, and Google AI
• Interactive mode for continuous conversations
• Streaming output
• History management with automatic metadata tracking
• Configurable via YAML file and environment variables
• Vim plugin for easy integration

cgpt provides sophisticated session history management:

• Auto-generated sessions: Use -H auto to create timestamped session files in ~/.cgpt/history/sessions/
• Session metadata: Each session includes creation time, auto-generated description, and fork tracking
• Fork tracking: When using -I input.yaml -O output.yaml, cgpt tracks the parent session
• Quick continuation: Use -C to continue your most recent session
• Descriptive metadata: Sessions automatically generate descriptions from the first user message

• Go 1.23 or higher (required for standard library packages like cmp, log/slog, and slices)
• One of the following API keys:
  • Anthropic API key
  • OpenAI API key
  • Google AI API key

brew install tmc/tap/cgpt

cgpt is written in Go. To build from source:

1. Install Go 1.23 or higher:

   • Visit the Go installation instructions
   • Verify your Go installation and version:

     go version

2. Set up your Go environment:

   # Add Go binary directory to PATH if not already done
   echo 'export PATH=$PATH:$HOME/go/bin' >> ~/.bashrc  # or ~/.zshrc for zsh
   source ~/.bashrc  # or source ~/.zshrc for zsh

3. Install cgpt:

   go install github.com/tmc/cgpt/cmd/cgpt@latest

4. Verify installation:

   cgpt --version

Download the latest release from the GitHub Releases page.

cgpt [flags]

• -b, --backend string: The backend to use (default "anthropic")
• -m, --model string: The model to use (default "claude-3-7-sonnet-20250219")
• -i, --input string: Direct string input (overrides -f)
• -f, --file string: Input file path. Use '-' for stdin (default "-")
• -c, --continuous: Run in continuous mode (interactive)
• -s, --system-prompt string: System prompt to use
• -p, --prefill string: Prefill the assistant's response
• -I, --history-in string: File to read completion history from
• -O, --history-out string: File to store completion history in (or - for stdout)
• -H, --history string: Read and write same history file (or 'auto' for auto-generated)
• -C, --continue: Continue most recent session
• --config string: Path to the configuration file (default "config.yaml")
• -v, --verbose: Verbose output
• --debug: Debug output
• -n, --completions int: Number of completions (when running non-interactively with history)
• -t, --max-tokens int: Maximum tokens to generate (default 8000)
• --completion-timeout duration: Maximum time to wait for a response (default 2m0s)

Before using cgpt, you need to set up your API keys. Choose one or more of the following:

# Anthropic (recommended default)
export ANTHROPIC_API_KEY='your-key-here'

# OpenAI
export OPENAI_API_KEY='your-key-here'

# Google AI
export GOOGLE_API_KEY='your-key-here'

For persistent configuration, add these to your shell's configuration file (~/.bashrc, ~/.zshrc, etc.).

cgpt can be configured using a YAML file. By default, it looks for config.yaml in the current directory. You can specify a different configuration file using the --config flag.

Example config.yaml:

backend: "anthropic"
model: "claude-3-7-sonnet-20250219"
stream: true
maxTokens: 2048
systemPrompt: "You are a helpful assistant."

cgpt includes a Vim plugin for easy integration. To use it, copy the vim/plugin/cgpt.vim file to your Vim plugin directory.

1. Visually select the text you want to process with cgpt.
2. Press cg or use the :CgptRun command to run cgpt on the selected text.
3. The output will be appended after the visual selection.

• g:cgpt_backend: Set the backend for cgpt (default: 'anthropic')
• g:cgpt_model: Set the model for cgpt (default: 'claude-3-7-sonnet-20250219')
• g:cgpt_system_prompt: Set the system prompt for cgpt
• g:cgpt_config_file: Set the path to the cgpt configuration file
• g:cgpt_include_filetype: Include the current filetype in the prompt (default: 0)

1. Old Go Version

   • Error: Package not found errors mentioning cmp, log/slog, or slices
   • Solution: Upgrade to Go 1.23 or higher

2. Missing API Keys

   • Error: Authentication errors or "API key not found"
   • Solution: Set the appropriate environment variable for your chosen backend

3. Configuration File Issues

   • Error: "Config file not found" or YAML parsing errors
   • Solution: Ensure your config.yaml is properly formatted and in the correct location

For additional help, please check the GitHub Issues page.

# Simple query
echo "Explain quantum computing" | cgpt

# Interactive mode
cgpt -c

# Use a specific backend and model
cgpt -b openai -m gpt-4 -i "Translate 'Hello, world!' to French"

# Save conversation history
cgpt -H auto -i "Explain quantum computing"

# Continue a specific session
cgpt -H session.yaml -i "Continue our discussion"

# Continue the most recent session
cgpt -C -i "What did we talk about?"

# Read from one file, write to another (fork)
cgpt -I old_session.yaml -O new_session.yaml -i "Let's try a different approach"

This project is licensed under the ISC License. See the LICENSE file for details.

This README provides an overview of the cgpt tool, including its features, installation instructions, usage examples, configuration options, and information about the Vim plugin. It also includes details about the supported backends and environment variables for API keys.

Happy hacking! 🚀