Gemini tools cheat sheet: app, Workspace, API, and CLI

Gemini shows up as a chat app, a side panel inside Google Workspace, an API, and a terminal agent. The names overlap and the model ids do not. This sheet lists only features confirmed on Google’s own pages in October 2026, then the calls you can paste.

Gemini app

You have to be signed in at gemini.google.com. The help pages say a work or school account also needs an admin to allow the feature you want.

Gems

Open the sidebar, open Gems, and choose New Gem. If you have never made one, the help page routes you through Settings and help, then Gems, first. Give it a name and instructions. Under Knowledge, Add files can upload from the device, from Drive, or from a NotebookLM notebook. Drive files need Keep Activity on and Workspace connected to Gemini Apps. A Drive file is read from its current version, so a later edit shows up in the Gem. Save. A Gem you create on the web also shows up in the mobile app and in the Workspace side panel.

Gems cannot be used with Gemini Live. The Learning coach Gem does not cover language learning. Starting in November 2026, Gems on personal Google accounts move to skills. Disable knowledge citations turns off citations for the Gem’s files and for files uploaded in that chat.

Deep Research

On the computer, open gemini.google.com, use Add Files, then Deep Research. You can attach files the same way as a normal prompt. Sources is a separate control. Google Search is included unless you clear it. Gmail, Drive, and other Google services appear when the Workspace app is connected. You can add NotebookLM notebooks. Type the question, submit, and Gemini writes a plan. Edit plan if the scope is wrong, then Start research. Open the report when it is ready.

A report usually takes about 5 to 10 minutes, longer when the question is heavier. You can leave the chat. The web app marks the thread when it is done. The mobile app can notify you. A Google AI Ultra plan may add charts, diagrams, and simulators. This sheet does not list Ultra prices.

Canvas

Add Files, then Canvas, then a prompt that names the format (doc, slides, app, or code). You can attach a file or an image first. Edits auto-save. You can turn the result into an Audio Overview or a quiz. Code, Show console, and Show recent changes sit at the top of the panel. Select and ask highlights a region. Add Gemini features, for signed-in users 18 or older, puts text or image generation inside a Canvas app.

Files, images, and code

Add files, then Upload Files. A code folder is under More Uploads, Import Code, Upload folder. A GitHub repository is the same Import Code path, using the repo or branch URL. Drive is Add from Drive, with the same activity and Workspace connection rules as Gems. Notebooks are under More Uploads.

Limits from that help page, as of October 2026:

If analysis fails, upload the file again. Prompt habits that transfer from chat products are in the ChatGPT prompting cheat sheet. A second research tool is Perplexity for technical research.

No Gemini app menu named Extensions was on the help pages read in October 2026. Canvas is the documented place to create an app. Terminal extensions are under Gemini CLI below.

Gemini in Google Workspace

Open Ask Gemini, usually at the top right. In Docs, switch between the side panel and the bottom bar. Pick a suggested prompt or write one. In Gmail, the same control drafts or revises mail. Docs help says you need an eligible Workspace or Google AI plan, and language support varies. A refresh or closing the document clears the Docs side-panel history. Insert text you need to keep.

The Workspace feature list places that side panel in Gmail, Docs, Sheets, Slides, Drive, and Chat, and it also names Help me write and Meet notes. Your edition may not include every row. If Ask Gemini in Chat is on, the learning page says the side panel is removed. Gems from the Gemini app show up in the Workspace side panel.

Gemini API quickstart

The API docs say the Interactions API has been the default since June 2026. generateContent still works and is marked legacy. Use Interactions for a new script. The official SDK is google-genai (General Availability as of May 2025). The older google-generativeai package is not maintained. The libraries page says legacy libraries were deprecated as of November 30, 2025.

Create a key in Google AI Studio. New users get a project and a key. Otherwise use Create API key on the API keys page. The get-started guide tells you to export it:

pip install -U google-genai
export GEMINI_API_KEY="YOUR_API_KEY"

from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-3.8-flash",
    input="List three checks before you restart a production Deployment. Keep each check to one sentence.",
)
print(interaction.output_text)

Client() with no arguments reads GEMINI_API_KEY. The get-started sample uses gemini-3.8-flash. The API homepage sample uses gemini-4-argon for a short explanation. Both ids are on the models page. The JavaScript package is @google/genai.

The get-started guide says a paid tier raises rate limits and needs Cloud Billing. As of that page in October 2026, the prepaid minimum it names is $5, or the same amount in your currency. Usage is under Dashboard, then Usage, in AI Studio. This sheet does not copy the rest of the price list.

Agents, tool calling, and a managed agent are separate API surfaces on the same docs site. How an agent differs from one interactions.create call is in what is agentic AI. Wiring tools over the Model Context Protocol is in MCP servers explained. A terminal coding agent from another vendor is the Claude Code cheat sheet.

Model lineup as of October 2026

The models page was last updated October 22, 2026. Stable ids are the ones to pin. A preview can be billed and can be deprecated with at least two weeks’ notice. A latest alias can change underneath you. An experimental id is not a production default.

Also on that page, and worth pinning from the page itself the week you ship: gemini-3.6-flash, gemini-3.1-flash-lite, image models Nano Banana 2.1 (gemini-nano-banana-2.1), Nano Banana 2 (gemini-3.1-flash-image), Nano Banana 2 Lite (gemini-3.1-flash-lite-image), Nano Banana Pro (gemini-3-pro-image), and video model Gemini Omni Flash (gemini-omni-1.1-flash). Gemini 2.5 Pro, Flash, and Flash-Lite (gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite) stay up for accounts that already used them. They are not deprecated. New projects should use 3.5 Flash-Lite or 3.8 Flash. Several 2.0 ids are marked shut down. Do not revive them from an old gist.

Gemini CLI

The README in google-gemini/gemini-cli still calls it an open-source terminal agent. It also says, in a note at the top, that on June 18, 2026 Gemini CLI was replaced by Antigravity CLI for unpaid-tier and Google One users, and that Gemini CLI has not been optimized for Gemini 4 Argon. Install it only if that note matches your account.

npx @google/gemini-cli
npm install -g @google/gemini-cli
brew install gemini-cli

Stable channel: npm install -g @google/gemini-cli@latest. Preview: @preview. Nightly: @nightly.

Authentication options the README lists:

gemini
gemini -m gemini-2.5-flash
gemini -p "Summarize the failing test in this repo" --output-format json
gemini --include-directories ../lib,../docs

The README’s own -m example is gemini-2.5-flash. Prefer gemini-3.8-flash when the CLI accepts it. --output-format json is for scripts. --output-format stream-json is newline-delimited events. MCP servers go in ~/.gemini/settings.json. gemini extensions install takes a GitHub URL or a local path. Free-tier quotas are not copied here. The README sent unpaid users to Antigravity instead of listing a number.

Patterns you’ll reuse

• New Python app: google-genai, Interactions, gemini-3.8-flash or gemini-4-argon, key only in the environment.
• Repeatable team prompt: a Gem with instructions and a Drive file, until the November 2026 skills move on personal accounts.
• “Read everything and write a report”: Deep Research, then edit the plan before Start research.
• A one-file tool or a short doc: Canvas, then export or copy the result out of the panel so a refresh cannot eat the Docs side-panel history.
• Workspace draft: Ask Gemini in Gmail or Docs, then insert the text. Do not leave the only copy in the side panel.
• Research you must footnote: Deep Research or Perplexity, then open the links.
• Terminal loop on a repo: Gemini CLI or Antigravity CLI, depending on the June 18, 2026 note in the README. MCP tools are the extension point. See MCP servers explained.
• Production pin: a stable model id from the table above, not gemini-flash-latest.

FAQ

How do I call the Gemini API from Python in a new project?

Install google-genai, create an API key in Google AI Studio, and export it as GEMINI_API_KEY. The get-started guide uses the Interactions API: genai.Client() then client.interactions.create with a model id and an input string. As of June 2026 that guide calls Interactions the default. generateContent is the legacy surface.

Which Gemini model should a new app use?

The models page, last updated October 22, 2026, points new work at current models such as gemini-3.8-flash or gemini-4-argon. It says Gemini 2.5 models stay available for people who already used them, and it tells new projects to use 3.5 Flash-Lite or 3.8 Flash instead. Pin a stable id. Do not ship on a latest alias unless you accept silent model swaps.

What is the difference between a Gem and Deep Research?

A Gem is a saved name, instructions, and optional knowledge files you reuse from the Gems list. Deep Research builds a plan, then a report from sources you select, including Google Search by default. The Gems help page says that starting in November 2026, Gems on personal Google accounts move to skills.

Is Gemini CLI still the terminal tool to install?

The Gemini CLI README still documents npm install -g @google/gemini-cli and a GEMINI_API_KEY from AI Studio. The same README says that on June 18, 2026 Gemini CLI was replaced by Antigravity CLI for unpaid-tier and Google One users, and that Gemini CLI is not optimized for Gemini 4 Argon. Read that note before you standardize on it.