aitermite 0.5.0

AI terminal copilot: fixes failed commands and answers plain thoughts (claude/codex), with an arrow-key run menu

AITERMITE

AITERMITE is an AI terminal copilot for Windows, macOS, and Linux. It does two things:

1. Fixes failed commands — manually, before Enter where the shell supports it, and automatically after a command fails through shell hooks.
2. Answers plain thoughts — type a question at your prompt with no command and no prefix; AITERMITE recognises it as natural language, asks Claude or Codex, and if the answer contains runnable shell commands, an arrow-key menu lets you confirm and run one.

Ask mode (new in v0.5.0)

Just type what you're thinking and press Enter:

$ how do I list the 5 largest files here
AITERMITE · claude
The classic way with find:

  find . -maxdepth 1 -type f -printf '%s %p\n' | sort -rn | head -5

  ↑/↓ move · Enter run · e edit · q cancel
 › [medium] find . -maxdepth 1 -type f -printf '%s %p\n' | sort -rn | head -5
   [medium] du -ah . | sort -rh | head -5

• No prefix needed. A failed line that looks like natural language is routed to the AI instead of the typo fixer (a real-command typo like gti status still goes to the fixer).
• Or be explicit: ai how do I tar a folder / aitermite ask "how do I tar a folder".
• Backends, auto-detected: the installed claude CLI, then codex, then a local ollama. No API keys required when a CLI agent is signed in. Force one with --ask-provider claude|codex|ollama or AITERMITE_ASK_PROVIDER.
• Arrow-key run menu with per-command risk badges. Enter runs the highlighted command through your shell, e edits it first, q cancels. Commands matching dangerous patterns (e.g. rm -rf /) are shown but blocked from running.
• --json prints the answer and extracted commands instead of opening the menu (handy for scripts and CI).

Current packaged release

Latest PyPI release: v0.5.0 Ask Mode.

Main features:

• Ask plain thoughts at the prompt — no command, no prefix.
• Answers from the claude / codex CLIs (auto-detected), or local ollama.
• Arrow-key menu to confirm and run commands found in the answer.
• Manual command fixing with aitermite.
• Inline fixing with aitermite <command>.
• Automatic post-failure suggestion hooks.
• Pre-enter typo detection where supported.
• Ollama local provider support.
• OpenAI fallback support.
• Heuristic offline fallback.
• Windows PowerShell / Windows Terminal support.
• macOS/Linux zsh, bash, and fish support.
• cmd.exe helper macros.
• Clink bootstrap integration for true cmd.exe auto-hooks.
• Secret redaction before provider calls.
• Dangerous command blocking and safety checks.

Install from PyPI

python -m pip install aitermite
aitermite --doctor

Upgrade to the latest published version:

python -m pip install --upgrade aitermite
aitermite --version

Install a specific version:

python -m pip install aitermite==0.5.0

Quick start

aitermite --doctor
aitermite git push                       # fix a command
aitermite ask "how do I tar a folder"    # ask a thought, get an arrow-key run menu
ai how do I free up disk space           # 'ai' alias, after shell integration
aitermite --precheck gti status
aitermite --install-shell auto

After shell integration, restart the terminal. AITERMITE then prints suggestions after failed commands, and — where the shell hook is supported — routes a plain natural-language line straight to the AI ask flow with no command or prefix.

Install from wheel

python -m pip install aitermite-0.5.0-py3-none-any.whl
aitermite --doctor

Install from source

git clone https://github.com/sathkruthdamera/AITERMITE.git
cd AITERMITE
python -m pip install -e .
aitermite --doctor

Shell integration

aitermite --install-shell auto

Supported shell targets:

aitermite --install-shell zsh
aitermite --install-shell bash
aitermite --install-shell fish
aitermite --install-shell powershell
aitermite --install-shell cmd
aitermite --install-shell clink
aitermite --install-shell universal

Automatic post-failure behavior

Target behavior:

Wrong command entered -> command fails -> AITERMITE automatically prints suggestion

Example:

gti status

Output:

command not found: gti

AITERMITE suggestion after failed command
Typed: gti status
Fix: git status
Why: Known command typo detected.
Confidence: high (0.94)
Run manually: git status

Provider settings

export AITERMITE_PROVIDER=auto
export AITERMITE_POSTFAIL_PROVIDER=auto
export AITERMITE_POSTFAIL_TIMEOUT_MS=900

For Ollama:

ollama pull gemma3:latest
export AITERMITE_PROVIDER=ollama
export AITERMITE_OLLAMA_MODEL=gemma3:latest

For OpenAI:

export OPENAI_API_KEY="your_api_key"
export AITERMITE_PROVIDER=openai

Common commands

aitermite --doctor
aitermite git push
aitermite --json git push
aitermite --precheck gti status
aitermite --postfail 127 gti status
aitermite --install-shell auto

Safety

AITERMITE blocks or avoids risky commands such as destructive delete patterns, curl | sh, fork bombs, device writes, shutdown/reboot patterns, and unsupported shell control operators in auto-apply flows.

Validation from local build

CLI builtin typo tests: 2/2 passed
Precheck tests: 2/2 passed
Redaction tests: 1/1 passed
Safety tests: 3/3 passed
Ask provider tests: 5/5 passed
Command extraction tests: 7/7 passed
NL detection tests: 6/6 passed
Arrow-key menu tests: 6/6 passed
Total targeted tests: 32/32 passed
Pentest checks: passed
Editable install: passed
Wheel build: passed (aitermite-0.5.0-py3-none-any.whl)
CLI smoke test: passed
Version: 0.5.0

Repository import status

This repository has been initialized from ChatGPT. The full v0.4.2 source ZIP and wheel were generated in the workspace. Source import should include:

src/aitermite/
tests/
security/
examples/
assets/
README.md
pyproject.toml
LICENSE
FULL_DOCUMENTATION.md
CROSS_PLATFORM_SHELL_INTEGRATION.md
POSTFAIL_AUTOMATION.md
SHELL_AND_ANIMATION.md
VALIDATION_REPORT.md