Command Line

Ask Aivana from your terminal with the aivana command. No code required.

The aivana command sends your question to the Aivana API and streams the answer into your terminal. It calls the same API as your application, with the same API key, so it's the quickest way to try Aivana, script it, or pipe files and command output straight in.

Install it with Python or Node.js, whichever you already have. Both install the same aivana command, and it behaves identically either way.

Create a key on the API Keys page, then put it in the AIVANA_API_KEY environment variable.

aivana "Should we migrate billing to DynamoDB?"
aivana "What's the default port for Postgres?" --effort low

The answer streams in as it's written. A progress line shows while Aivana works, but only the answer goes to standard output, so aivana "..." > answer.md saves just the answer.

Text piped in is sent ahead of your question, so you can hand Aivana a diff, a log or a document without copying and pasting.

git diff | aivana "Review this change" --effort high
aivana "Summarize the three biggest risks" < contract.txt
aivana "What's driving the dip in this chart?" --image chart.png

The command's --image option attaches images: up to 5 per question, each a PNG, JPEG, WebP or GIF of up to 8 MB. The Aivana API itself also accepts PDF, Word and CSV files (see Files & Documents); from the command line, extract a document's text first and pipe that in.

• --effort auto | low | medium | high: a ceiling on how much intelligence Aivana may apply. auto, the default, lets Aivana judge from the question.
• --shape: the answer format. One of auto, text, recommendation, summary, tradeoffs, decision or extract.
• --web / --no-web: always, or never, search the web first. With neither, there is no search, the default for API keys. A question that searches is charged every token used to answer it, not just your question and the answer.
• --system "...": your own instructions, such as a persona, tone or format.
• --assistant-name NAME: the name the assistant presents as.
• --max-tokens N: a ceiling on answer length. It can shorten an answer, never lengthen it.
• --temperature T: from 0.0 to 2.0. Leave it off to let Aivana choose per question.
• --image PATH: attach an image. Repeat it for up to 5.
• --trace: show the Intelligence Trace, which explains how Aivana handled your question.
• --json: wait for the whole response and print it as JSON, for scripts.
• -q, --quiet: hide the progress line.

Scripts can tell what happened from the exit code, including whether a retry makes sense:

0    success
1    the request failed
2    bad usage
3    authentication problem: no key, or an invalid or expired one
4    rate limited: wait, then retry
5    out of credits: top up in AI Studio
6    temporary failure (network, timeout, upstream): safe to retry
130  interrupted (Ctrl-C)

• Quote your question. The shell acts on characters such as ?, * and ' before aivana sees them.
• In scripts and CI jobs with nothing to pipe in, add < /dev/null. Otherwise aivana waits up to 3 seconds for piped input before answering.
• Every run is a normal API call and uses credits, charged by tokens.
• --effort low is useful for simpler questions and faster experimentation.
• AIVANA_API_BASE points the command at a different Aivana environment.
• aivana --help lists every option.