llm-gemini

You can prevent requests from hanging indefinitely using the timeout option.

CLI usage:

llm -m gemini-flash-latest 'prompt' -o timeout 1.5

Python usage: When using the llm library, pass the timeout parameter to model.prompt(). This raises an httpx.TimeoutException if exceeded.

import httpx, llm

model = llm.get_model("gemini/gemini-flash-latest")

try:
    response = model.prompt("epic saga about mice", timeout=1.5)
    print(response.text())
except httpx.TimeoutException:
    print("Timeout exceeded")

You must provide a Google Gemini API key to use the plugin. You can configure it using the llm keys set command or by setting the LLM_GEMINI_KEY environment variable.

To set it via the CLI:

1. Run llm keys set gemini.
2. Paste your API key when prompted.

Gemini models support multi-modal prompting. Use the -a flag to attach files or URLs.

Images:

llm -m gemini-flash-latest 'extract text' -a image.jpg
llm -m gemini-flash-latest 'describe image' -a https://example.com/image.jpg

Audio:

llm -m gemini-flash-latest 'transcribe audio' -a audio.mp3

Video:

llm -m gemini-flash-latest 'describe what happens' -a video.mp4

YouTube Videos:

llm -m gemini-flash-latest -a 'https://www.youtube.com/watch?v=VIDEO_ID' 'Summarize this video'

Note on YouTube resolution: YouTube videos are processed with low resolution by default. Use -o media_resolution X where X is medium, high, or unspecified to change this.

llm -m gemini-flash-latest -a 'https://www.youtube.com/watch?v=9o1_DL9uNlM' 'Produce a summary...'

Run prompts against Gemini models using the -m flag to specify the model ID. You can also set a default model to avoid specifying it every time.

Basic usage:

llm -m gemini-flash-latest "Your prompt here"

Set a default model:

llm models default gemini-flash-latest
llm "Your prompt here"

llm -m gemini-flash-latest "A short joke about a pelican and a walrus"

The plugin supports Gemini embedding models. You can generate embeddings for single strings or bulk-embed files into a SQLite database.

Single string embedding:

llm embed -m gemini-embedding-2 -c 'hello world'

Bulk embedding files: Use embed-multi to embed files into a collection in a SQLite database.

llm embed-multi readmes -d embed.db -m gemini-embedding-2-768 --files . '*/README.md'

Similarity search:

llm similar readmes -c 'upload csvs to stuff' -d embed.db

Model variants: You can request specific vector dimensions by appending a suffix to the model ID:

• gemini-embedding-2-768 (768 dimensions)
• gemini-embedding-2-1536 (1536 dimensions)
• gemini-embedding-2 (3072 dimensions - default)

You can enable specific server-side capabilities using the -T flag.

Code Execution: Allows the model to write and execute Python code in a sandbox.

llm -m gemini-3.6-flash -T CodeExecution 'use python to calculate (factorial of 13) * 3'

Google Search: Enables Grounding with Google Search.

llm -m gemini-3.6-flash -T GoogleSearch 'What happened in Ireland today?'

Note: To inspect grounding metadata, run llm logs -c --json after the prompt.

URL Context: Allows the model to fetch content from URLs.

llm -m gemini-2.5-flash -T URLContext 'Latest headline on simonwillison.net'

Note: Use --usage to see token counts for tool use.

To start an interactive chat session with a Gemini model, use the chat command.

llm chat -m gemini-flash-latest

Install the llm-gemini plugin into your existing llm environment using pip via the llm install command.

llm install llm-gemini

To ensure the model returns a valid JSON object, use the -o json_object 1 option.

llm -m gemini-flash-latest -o json_object 1 '3 largest cities in California, list of {"name": "..."}'

The plugin provides access to various Gemini and Gemma models. You can use the full ID (e.g., gemini/gemini-3.5-flash) or the shorter alias (e.g., gemini-3.5-flash).

To see all available models via the plugin, run:

llm gemini models

Commonly used models include:

• gemini-flash-latest (Latest Gemini Flash)
• gemini-3.5-flash (Gemini 3.5 Flash)
• gemini-3.5-flash-lite (Gemini 3.5 Flash Lite)
• gemma-4-31b-it (Gemma 4 31B Instruct)
• gemini-embedding-2 (Latest embedding model)

Use llm gemini models to list models exposed by the Gemini API. You can filter by supported generation methods using the --method flag.

Filter by method:

llm gemini models --method embedContent