@deckyard/gemini

A client for the Google Gemini API as Google AI Studio serves it, for Jennifer programs: generate content from text and media with a system instruction, function calling and a tool loop that keeps thought signatures, structured output, thinking and safety settings; count tokens; embeddings; model listing; the Files API for large media; context caching. Requests and replies are Google's own JSON, every request has a deadline and a reply cap, and rate limits are retried with backoff.

Documentation

Install

jvc add @deckyard/gemini

This vendors the deck to vendor/@deckyard/gemini/. Without jvc, unpack the src/ subtree of a release archive (deckyard-gemini-<version>.tar.gz) there.

Usage

import "@deckyard/gemini/" as gemini;
use io;
use json;
use os;

def c as gemini.Client init gemini.client(os.getEnv("GEMINI_API_KEY"));
def req as json.Value init gemini.newRequest("gemini-3.8-flash");
$req = gemini.withSystem($req, "Answer in one sentence.");
$req = gemini.addUserText($req, "Why is the sky blue?");
io.printf("%s\n", gemini.text(gemini.chat($c, $req)));

A key comes from Google AI Studio; the free tier needs no billing account. The examples in examples/ (a conversation, a tool loop, a file upload) read it from GEMINI_API_KEY.

Requirements

Jennifer 0.26.0 or newer, the default binary: the deck opens network connections (capability net), so it does not load on jennifer-tiny. No streaming yet (it waits for streaming http); no Vertex AI yet.

Development

tools/check-all.sh                   # the full gate, as CI runs it
jennifer test src/gemini_test.j      # the test overlay, against a loopback fake
tools/check-docs.sh                  # docs against the code
tools/pack.sh dist                   # build the release archive

Licence

LGPL-3.0-only, see LICENSE. Copyright (C) 2026 mplx <mplx@jennifer-lang.dev>.