Pipes and scripts

kunji ask answers one question and exits, which makes it easy to use with other commands.

Pipe output into a model

npm test 2>&1 | kunji ask "Why are these tests failing?"
git log --oneline -20 | kunji ask "Write release notes from these commits"
cat config.yaml | kunji ask "Is anything wrong with this config?"

Piped text is added under your question. Very large input is cut to the last 200,000 characters.

Plain output for scripts

When the output goes to a file or another command, kunji prints plain text with no colours or boxes. Force it with --raw:

kunji ask --raw "Write a .gitignore for a Node project" > .gitignore
git commit -m "$(kunji @commit --raw)"

Keys without prompts

In CI or on servers, use environment variables instead of the keychain:

export OPENAI_API_KEY=sk-...
kunji ask -m openai:gpt-4o-mini "Summarize this" < report.txt

Or store a key from a script without typing it:

echo "$GROQ_API_KEY" | kunji keys add groq --key-stdin --model llama-3.3-70b-versatile

All variables: Environment variables.

Exit codes

Code	Meaning
0	Success
1	Something failed (bad key, provider error, missing file); the message says why
2	The command was used wrongly (missing argument, interactive command without a terminal)
130	Cancelled with Ctrl+C

Interactive vs. non-interactive

kunji chat and the menu need a real terminal. In scripts, use kunji ask, kunji @name and commands with flags (for example kunji keys remove groq --yes).