nanoterm

Built by the Nano Collective — a community collective building AI tooling not for profit, but for the community.

nanoterm is an ultra-lightweight, high-performance AI terminal companion that translates natural language requests into bash commands securely and instantly.

It allows you to bypass writing complex regex, find, awk, or sed commands by simply asking your terminal what you want it to do in plain English. nanoterm connects directly to your chosen provider (OpenAI, Anthropic, Google, Atlas Cloud) via the Vercel AI SDK.

What it is / What it is not

nanoterm is a highly secure, privacy-first command generator designed for developers who value safety and speed.

What it does:

• Automatically scrubs sensitive data (emails, IPs, phone numbers) before sending your prompt to a cloud LLM, ensuring your privacy.
• Commands matching common destructive patterns are clearly flagged and require explicit confirmation.
• Interactively prompts you for approval before executing any command on your machine.
• Shares configuration seamlessly with the Nanocoder ecosystem.

What it does not do:

• It does not blindly execute commands without your explicit consent.
• It is not an autonomous agent; it only acts when invoked and waits for approval.
• It does not track you, send telemetry data, or require an account.

Important

Always review the generated bash command carefully before pressing y (Yes) to execute. You always have the option to press edit to safely modify it before it runs!

Security model

• Commands that do not need shell features are executed directly, without a shell. Pipelines, redirects, expansions, and shell built-ins still require the configured shell.
• Commands are evaluated for safety: Destructive commands (e.g. rm -rf) require typing the full word yes. Cautionary commands (e.g. dynamically built commands like $(...)) trigger a warning but only require a simple y. This is an additional review boundary, not a sandbox.
• Cloud prompts are scrubbed before transmission. Scrubbing is skipped only for explicitly known local providers or URLs whose parsed hostname is a loopback address.
• Recent command output is bounded, expires after ten minutes, and is sent to the model as untrusted user-context data rather than as a system instruction.
• Provider credentials and their configuration directory are restricted to the current user on platforms that support POSIX permissions.

Quick Start

Install globally to use the CLI:

npm install -g @nanocollective/nanoterm

Configuration Wizard

Before generating commands, set up your API keys and default models:

nanoterm config

This will launch a beautiful interactive setup wizard. You can choose from OpenAI, Anthropic, Google Gemini, Atlas Cloud, or any Custom OpenAI-compatible provider.

Usage Examples

CLI: Generate and execute a command

nanoterm find all png files in the current directory

Proposed command:
> find . -type f -name "*.png"

Execute? [y/N/edit/?]: y

When prompted to execute, you have several options:

• Press y to execute the command.
• Press Enter (or N) to safely abort (this is the default).
• Type edit to manually modify the command before it runs.
• Type ? to get a concise explanation of what the command does before you decide.

Session Chaining

nanoterm automatically feeds the output (stdout/stderr) of your last executed command into the AI's context for your next request. This allows for seamless session-chaining!

For example:

1. nanoterm run the tests
2. nanoterm fix the failing test in the output (The AI already knows what failed!)

You can even omit the quotes entirely! nanoterm list all open ports works seamlessly out of the box.

Warning

Unquoted requests are subject to shell expansion before nanoterm sees them. For example, nanoterm find all *.png files will expand *.png in your shell if matches exist. If your request includes glob patterns, variables, or shell metacharacters, it is safer to wrap your request in quotes.

Community

• Discord: Join the Nano Collective Discord
• Contributing: Read our Contributing Guide to get started.
• Issues: Check the GitHub Issues for planned work or to report bugs.