GitHub - lazyville/qsh: q - Slim LLM CLI

A slim LLM CLI for your terminal.
Ask questions, debug errors with session context, and redact secrets — all from a single shell script.

Deps: curl (always present), jq (auto-installed if missing)

Install

curl -sL https://github.com/lazyville/qsh/releases/latest/download/q.sh > ~/.local/bin/q && chmod +x ~/.local/bin/q

Setup

Add session logging so q debug / q wtf have terminal context to work with:

q install-hook   # prints the hook — paste into ~/.zshrc or ~/.bashrc

The hook logs commands (skipping vim, fzf, claude, etc.) and provides q-stop / q-start aliases to pause/resume logging around interactive tools.

Usage

q <prompt>                     one-shot question
q debug [question]             ask with recent terminal output as context
q wtf                          shortcut: "what went wrong and how do I fix it?"
q log                          preview what context would be sent (dry run)
q clear                        wipe the session log
q-stop                         pause logging (shell alias — preserves TTY for vim/claude/fzf)
q-start                        resume logging (shell alias)
q config                       show active config
q install-hook                 print shell hook to add to your rc file
q completions [bash|zsh|fish]  print shell completion script
q ignore                       list redaction patterns
q ignore add <PATTERN>         redact a word/var from context (e.g. API_KEY)
q ignore remove <PATTERN>      remove a pattern

Note: q-stop and q-start are shell aliases, not subcommands. They must run in your current shell to take effect — using q stop / q start won't work.

Shell Completions

Enable tab completion for subcommands:

# zsh — add to ~/.zshrc
eval "$(q completions zsh)"

# bash — add to ~/.bashrc
eval "$(q completions bash)"

# fish — run once
q completions fish > ~/.config/fish/completions/q.fish

Providers

Set Q_PROVIDER (default: openai):

Provider	Endpoint	Key var
`openai`	api.openai.com/v1	`OPENAI_API_KEY`
`nova`	api.nova.amazon.com/v1	`AWS_API_KEY`
`azure`	your resource URL	`AZURE_API_KEY`
`anthropic`	api.anthropic.com (native fmt)	`ANTHROPIC_API_KEY`
`<custom>`	set `Q_ENDPOINT` directly	`Q_API_KEY`

Config

Config is auto-saved on first run to ~/.config/q/config.sh. Override with env vars:

Q_PROVIDER        openai | nova | azure | anthropic | <custom>
Q_MODEL           model name
Q_ENDPOINT        override the preset base URL
Q_API_KEY         override the preset key var
Q_SYSTEM          system prompt
Q_SESSION_LOG     log path  (default: ~/.q/current.log)
Q_CONTEXT_LINES   lines of context to send  (default: 80)

Azure extras: Q_AZURE_BASE, Q_AZURE_DEPLOY, Q_AZURE_API_VER

Show Config

➜  qsh git:(main) q config
provider   openai
endpoint   https://api.nova.amazon.com/v1
model      nova-2-lite-v1
api key    set (36 chars)
log        ~/.q/current.log  (last 80 lines)
config     ~/.config/q/config.sh
ignore     ~/.config/q/qignore

ignore patterns:
  vim
  claude

Examples

q "explain this regex: ^[a-z]+$"

q wtf   # after a command fails

Q_PROVIDER=anthropic Q_MODEL=claude-opus-4-6 q "refactor this"

Q_PROVIDER=azure Q_AZURE_BASE=https://myco.openai.azure.com Q_AZURE_DEPLOY=gpt-4o q "hello"

Q_PROVIDER=custom Q_ENDPOINT=http://localhost:11434/v1 Q_API_KEY=ollama Q_MODEL=llama3 q "hello"

qignore

Redact sensitive values from context before they're sent to the LLM:

q ignore add MY_SECRET_VAR
q ignore add AWS_SECRET_ACCESS_KEY

Patterns are stored in ~/.config/q/qignore and matched as word boundaries and VAR=value pairs.

推荐订阅源

Hacker News - Newest: "LLM"