shgpt 0.7.1

Chat with LLM in your terminal, be it shell generator, story teller, linux-terminal, etc.

ShellGPT

中文

Chat with LLM in your terminal, be it shell generator, story teller, linux-terminal, etc.

Install

pip install -U shgpt

or if you prefer to use uvtool

uv tool install shgpt

This will install two commands: sg and shellgpt, which are identical.

After install, use sg --init to create the required directory and a default configuration file (~/.shellgpt/config.toml).

Usage

ShellGPT has three modes:

• Direct mode: sg [question] or via pipeline: echo question | sg.
• REPL mode: sg --repl for an interactive chat.
• TUI mode: sg --tui for a text user interface, tailored for shell command inference.

Configuration

ShellGPT is configured via ~/.shellgpt/config.toml. You can change the configuration directory by setting the SHELLGPT_CONF_DIR environment variable.

Multiple Profiles

Define your API settings in the [profiles] section and switch between them using the -p / --profile flag.

# ~/.shellgpt/config.toml
default_profile = "ollama"

[profiles.ollama]
base_url = "http://localhost:11434/v1"
model = "llama3"

[profiles.openai]
base_url = "https://api.openai.com/v1"
# You can provide api_key directly
api_key = "sk-xxxx"
# Or provide the name of the environment variable that contains the API key
# api_key_env = "OPENAI_API_KEY"
model = "gpt-4o"
# Headers support environment variable substitution ($VAR or ${VAR})
headers = { "Authorization" = "Bearer $OPENAI_API_KEY", "X-My-Header" = "value" }

# https://docs.github.com/en/github-models/quickstart
[profiles.github]
base_url = "https://models.github.ai/inference"
api_key = "ghp_xxxx"
model = "gpt-4o"

# https://developers.cloudflare.com/workers-ai/configuration/open-ai-compatibility/
[profiles.cloudflare]
base_url = "https://api.cloudflare.com/client/v4/accounts/<account-id>/ai/v1"
api_key = "<token>"
model = "@cf/meta/llama-3-8b-instruct"

# https://developers.cloudflare.com/ai-gateway/usage/chat-completion/
[profiles.cloudflare-gateway]
base_url = "https://gateway.ai.cloudflare.com/v1/<account-id>/<gateway-id>/compat"
api_key = "sk-xxxx"
model = "gpt-4o"

Usage:

# Uses default_profile (ollama)
sg How to center a div?

# Uses OpenAI profile
sg -p openai Write a poem about rust

Global Settings

You can override global defaults at the top level of the config file:

timeout = 60
stream = true
temperature = 0.8

Roles

Roles allow you to define specific system prompts for different tasks. They are managed in the [roles] section of your config.

Built-in examples (created during sg --init):

• shell: Infer shell commands.
• typo: Correct text typos.
• code: Coding assistant.
• summary: Summarize text in Markdown.
• polish: Polish writing.

Usage:

# List all available roles and their contents
sg --list

# Use a specific role
sg -s typo I is a enginer
git diff | sg -s commit

Custom Roles

Add your own roles to config.toml:

[roles]
translator = "You are a professional translator. Translate everything to Japanese."

Then use it with sg -s translator Hello world.

TUI

Key bindings in TUI mode:

• ctrl+j: Infer answer
• ctrl+r: Run command
• ctrl+y: Yank command

License

GPL-3.0