oops

Type oops after a command fails. It figures out the fix and runs it.

Like thefuck, but it reads the actual error text with an LLM instead of matching predefined rules. Zero dependencies. Works offline with a local model — no API key needed.

demo.mov

---

$ git statsu
git: 'statsu' is not a git command. See 'git --help'.

$ oops
↻ reproducing: git statsu
🤖 thinking...

  git status

Run this? [Y/n] y
On branch main
nothing to commit, working tree clean

You mistyped a command, it failed, you sigh — and instead of squinting at the error you just type oops.

Install

pip install oops-cli
oops-fix --install     # adds a tiny shell hook to ~/.zshrc / ~/.bashrc
source ~/.zshrc        # or ~/.bashrc, or open a new terminal

Then point it at a brain — pick one:

# Anthropic — fast and smart.   https://console.anthropic.com/
export ANTHROPIC_API_KEY=sk-ant-...

# Gemini — generous free tier.  https://aistudio.google.com/apikey
export GEMINI_API_KEY=...

# Groq — very fast, free tier.  https://console.groq.com/keys
export GROQ_API_KEY=...

Add the export line to ~/.zshrc (or ~/.bashrc) so it survives a new shell. Or skip the key entirely and stay fully offline with Ollama: ollama pull llama3.2.

After any failed command, type oops.

Your key, your machine

There is no server. oops reads the key from your shell environment and talks to the provider directly over HTTPS — nothing routes through me. The entire provider layer is one short file; read it if you're not sure.

How it works

1. The shell hook hands oops the command you just ran.
2. oops re-runs it to capture the exact error output.
3. It asks the model for a corrected command.
4. It shows you the fix and runs it only if you say yes. Nothing executes without your confirmation.

Caveats

• It re-runs your failed command to capture the error, so don't point it at anything destructive (deploys, rm, migrations). Re-running has side effects.
• The shell hook can mis-capture multi-line pastes. If you paste several commands at once, oops may pick the wrong one — type commands on their own lines instead. Fix is on the list.

Config

Env var	Default	What
ANTHROPIC_API_KEY	–	Use the Anthropic API
GEMINI_API_KEY	–	Use the Gemini API
GROQ_API_KEY	–	Use the Groq API
OOPS_MODEL	claude-haiku-4-5-20251001	Anthropic model
OOPS_GEMINI_MODEL	gemini-2.0-flash	Gemini model
OOPS_GROQ_MODEL	llama-3.3-70b-versatile	Groq model
OOPS_OLLAMA_MODEL	llama3.2	Local model name
OOPS_OLLAMA_URL	http://localhost:11434	Ollama endpoint

Provider precedence: Anthropic → Gemini → Groq → local Ollama. The first key found wins.

Why

The error message almost always contains the answer. oops just reads it for you so you don't have to context-switch. It's the difference between a 200ms oops and a 30-second detour to your browser.

Contributing

Issues and PRs welcome. Good first ones: more providers (OpenAI, Mistral), a --yes auto-run flag, multi-suggestion mode, and a --explain mode that explains instead of fixing.

License

MIT