ai-fix: terminal tool that uses AI to diagnose and fix failed commands automatically

ai-fix captures a failed command's error output, sends it to Claude Haiku (or local Ollama/GPT-4o-mini), and returns a structured fix with a confidence score (high/medium/low) that users can apply with one keystroke or review first with --dry-run.

The tool reads command history from zsh, bash, or fish; sends only the failed command, error (truncated to 3000 chars), OS version, language versions, and project file names—never file contents, environment variables, or API keys—keeping data exposure minimal.

Each fix costs roughly $0.0003 using Claude Haiku (less than 1/10th of a cent); Ollama (local, free) is the default if running; latency is 0.5–1.5s for cloud backends and ~50–200ms for Ollama, by design to avoid context-switching to a browser.

Requires Python 3.11+; install with export ANTHROPIC_API_KEY or OPENAI_API_KEY, then run ai-fix (for last failed command), ai-fix -y [command], or ai-fix --dry-run to preview fixes without applying.