Nitro: A tiny and efficient harness for running Bash commands.

Quickstart

1. Install Nitro:

   npm install -g @aerovato/nitro

2. Add a provider:

   nitro provider add

3. Run your first command:

   nitro "find all TS and TSX files and count total lines"
   nitro "summarize how the Bash tool is implemented src/tools/bash.tsx"
   nitro "read the README and setup the project for development"

Usage

nitro "<request>"              # Execute request and exit

nitro interactive [request]    # Start interactive session
nitro i [request]              # (shorthand)

nitro continue <request>       # Continue last conversation
nitro c <request>              # (shorthand)

nitro resume [request]         # Resume last conversation interactively
nitro r [request]

nitro strict [request]         # Run in strict mode (always confirm commands)
nitro s [request]

nitro provider add             # Add a new provider
nitro provider list            # List all providers
nitro provider edit            # Edit a provider
nitro provider remove          # Remove a provider
nitro provider default         # Set default provider

nitro settings                 # Configure Nitro settings
nitro help                     # Print help message

Features

Command in Natural Language

Nitro translates natural language requests into shell commands.

# Instead of remembering complex commands:
find . -name "node_modules" -prune -o -name "*.md" -print0 | xargs -0 wc -l | sort -rn | head -n 11

# Just describe what you want:
nitro "find all markdown files except node_modules and count total lines, show top 10"

More examples:

nitro "squash last 5 commits into 1 with message: Bug fix"
nitro "replace 'foo' with 'bar' in all txt files"
nitro "compress input.mov to a smaller mp4 (h264) optimized for smaller file"
nitro "show last 20 commits that modify src/main.ts with author, date, and message"
nitro "get 10 most recent open gh issues with P1 label, show id and title"

Multi-Provider Support

Nitro works with your preferred AI provider:

• OpenAI
• Anthropic
• Z.ai Coding Plan
• Qwen
• DeepSeek
• Mistral
• Groq
• Any OpenAI/Anthropic compatible provider

Safety Model

Nitro automatically evaluates command risk before execution:

Risk Levels:

• Read Only: Safe commands (ls, cat, find). Auto-approved unless in strict mode.
• Normal: Low-risk modifications. Requires confirmation.
• Dangerous: File deletions, overwrites. Requires confirmation.
• Extremely Dangerous: System-wide changes, multiple deletions. Requires confirmation.

Behavior Tags:

• Safe, Reversible, Write, Delete, Overwrite, Side Effects, Exfiltration

Strict mode (nitro strict) requires confirmation for all commands including read-only operations.

Safety Advisory

Nitro executes commands on your system. Only use Nitro on code and projects you trust.

• Do not use Nitro on untrusted codebases
• Always review commands before approving
• Keep backups of important data

Customization

Run nitro settings to configure:

• alwaysConfirm: Require confirmation for all commands
• showThinking: Display AI reasoning for supported models
• showTokenSummary: Display token usage summary on session exit
• maxOutputTokens: Maximum output tokens per response
• reasoningEffort: Reasoning effort level

To customize the system prompt body, create ~/.nitro/system_prompt.md.

• The default system prompt body can be found in ~/.nitro/system_prompt_template.md

Configuration

Nitro stores configuration in ~/.nitro/:

• settings.json - User settings
• auth.json - Provider configuration & authentication
• system_prompt.md - Custom system prompt (optional)

Uninstalling

npm uninstall -g @aerovato/nitro
rm -rf ~/.nitro