CLI (One-Shot)

Run a prompt, stream the response, exit. This is the default mode when you pass a prompt on the command line.

ra "What is the capital of France?"

ra streams the model's response to stdout, executes any tool calls the model makes, and exits when the agent loop completes.

Common flags

ra --provider openai --model gpt-4.1-mini "Explain this"
ra --file report.pdf "Summarize in three bullets"
ra --skill code-review --file diff.patch "Review this diff"
ra --system-prompt "You are a JSON extractor. Output only JSON." "Extract fields from: ..."
ra --thinking high "Design a distributed cache"

Piping

When input is piped, ra reads stdin and auto-switches to CLI mode. The piped content becomes part of the prompt.

cat error.log | ra "What is causing this error?"
git diff | ra --skill code-review "Review this diff"
git diff | ra "Summarize these changes"
echo "hello world" | ra                             # stdin becomes the prompt

Chaining

Chain ra commands together for multi-step workflows:

ra "List all TODO comments" | ra "Group by priority and format as a table"

File attachments

Attach images, PDFs, and text files with --file:

ra --file screenshot.png "What's wrong with this UI?"
ra --file report.pdf --file data.csv "Summarize both files"

See File Attachments for supported formats and provider compatibility.

Resuming sessions

When the agent calls AskUserQuestion, the session ID is printed to stderr so you can resume later:

ra --resume <session-id> "Continue from where we left off"

See Sessions for more on session persistence.

Exit codes

Code	Meaning
0	Success
Non-zero	Error (provider failure, config error, etc.)

See also

• REPL — for interactive sessions
• Configuration — all CLI flags and their config equivalents
• Recipes — piping and chaining patterns