ask README

The project's README from github.com/benja/ask. As one markdown file: README.md.

ask

A small CLI for asking AI questions from your terminal.

Motivation • Install • Usage • How it works

Ask a question and get back to your shell:

$ ask how do i find TODO or FIXME comments in src
Use ripgrep:
  rg -n 'TODO|FIXME' src

$ rg -n 'TODO|FIXME' src
src/generated/client.ts:18:// TODO: generated placeholder
src/generated/schema.ts:204:// FIXME: compatibility shim
src/api.ts:42:// TODO: retry failed requests

$ ask -c how do i exclude generated files
Add an exclude:
  rg -n --glob '!generated/**' 'TODO|FIXME' src

$ rg ...

Or just run ask to start a session:

$ ask
> why is this container restarting?
Check its exit code:
  docker inspect -f '{{.State.ExitCode}}' my-container

> it says 137. what does that mean?
Exit code 137 usually means SIGKILL, often from running out of memory.

> thank you
You're welcome.
> ^D

$ ...

Every answer is saved. Use ask -c to continue the latest one in this folder.

Motivation

I made ask because I wanted a normal terminal command for asking AI questions, whichever agent or model I used. I didn't want a TUI. I wanted to ask something, read the answer, and get my shell back.

Coding agents do a lot in the background, and I don't always want that. With ask, the agent stays read-only. I run the commands and make the changes myself. That way I actually learn.

Install

curl -fsSL https://benja.dev/ask/install.sh | sh

The installer supports Intel/ARM Macs and x86_64/ARM64 Linux. It verifies the release checksum and installs to ~/.local/bin/ask. Set ASK_INSTALL_DIR to change the destination.

Usage

ask [QUESTION...]          ask once
ask                        start a session
ask -c [QUESTION...]       continue the latest session here
ask --sessions             reopen a saved session
ask --settings             set defaults for new sessions
ask --upgrade              update ask
ask -V                     print the version

Answers go to stdout. Prompts and errors go to stderr, so one-shot answers are safe to pipe. Agents start in the current directory and run read-only.

ask checks for updates at most once a day. If a new version is available, it lets you know when you get your shell back. You won't get another notice for that version. Set ASK_NO_UPDATE_CHECK=1 to turn this off. ask only updates when you run ask --upgrade.

How it works

Under the hood, ask runs Codex, Claude Code, Pi, or OpenCode using the CLI already installed and logged in on your machine.

For a new session, ask uses your default agent, model, and reasoning. Continuing a session restores those saved choices and the underlying agent session ID. Answer instructions come from the current global setting. Ask starts the selected CLI read-only in your current folder, prints the answer to stdout, and saves the turn automatically.

Agents and models

Run ask --settings to choose the default agent, model, reasoning, and answer instructions. Use /settings inside a session to change that session's model or reasoning, or to edit the global answer instructions. Start a new session to switch agents.

Contributing

Requires Rust 1.88 or newer. Before opening a pull request:

cargo fmt --check
cargo test --locked
cargo clippy --locked --all-targets -- -D warnings

Installer changes also need shellcheck install.sh tests/install.sh and sh tests/install.sh.

License

MIT. Have fun :)

ask (benja/ask)

Searching Database…

139 ·GitHub
Ask
What is this?A terminal CLI that lets you ask AI questions and get a direct answer back in your shell, powered by Codex, Claude Code, Pi, or OpenCode.
Is this its own AI engine or a wrapper around other tools?It is a thin wrapper, not a standalone AI engine — it runs whichever coding-agent CLI you already have installed, in read-only mode.
How do I get it running?Run the installer script, which checks your platform and installs the binary to ~/.local/bin/ask; then just run 'ask' to start a session or 'ask -c' to continue the latest one.
Do I need an AI subscription or API key set up first?You need Codex, Claude Code, Pi, or OpenCode's CLI already installed and logged in — ask itself doesn't ship its own model access.
Can ask actually edit my code or run commands for me?No — the agent always stays read-only; it reads the project, answers, and you're the one who runs commands and makes changes.
Does the installer support Windows?No — the installer only covers Intel/ARM Macs and x86_64/ARM64 Linux, there's no Windows build.
How do I continue a previous conversation?Run 'ask -c' to continue the latest saved session in the current folder, restoring the same agent, model, and reasoning that were used before.
Can I pick which agent or model it uses?Yes — run 'ask --settings' to set the default agent, model, and reasoning, or use '/settings' inside a session to change that session's model.
Does it update itself automatically?It checks for a new version once a day and tells you when one is available, but only actually upgrades when you run 'ask --upgrade'; you can disable the check with ASK_NO_UPDATE_CHECK=1.
Why did the author build this instead of just using a TUI agent?He wanted a normal terminal command that answers a question and hands the shell back immediately, without the agent making background changes, so he still runs the commands himself and learns.
What are similar projects?foreman, fable-orchestrator, pstack, golutra, copperhead, pstack-claude, testsprite-cli, fixing-smartass-opus-5
Which topics is it in?pi-coding-agent, opencode, cli-tool, agentic-ai, coding-agent, ai-agent, rust, productivity, codex, command-line-tool, agent, claude-code
Creator
benja
Size
0.1 MB
Stars
139
Updated
CommitAug 23Releasev0.2.1Aug 23
Age
1mo
Commits · 90 Days