Skip to main content

Google Gemini

Call Gemini models from an automation - generate text, answer with live Google Search grounding, extract schema-shaped data, and generate images - on your own API key.

Connect a Google AI Studio project with an API key to use Gemini models as automation tools: generate text from a prompt, answer a question grounded in live Google Search results with the pages it used, extract data that matches a JSON Schema, generate an image that lands as an attachment on the session, and count a prompt's tokens for a specific model. List models is the source of truth for which model ids this key can reach and what each supports. This is separate from the model that runs your automation - that is configured under Settings → AI settings → AI providers, and per agent under Setup → AI → Model & quality, on the organization's own keys. This integration uses an API key and is unrelated to the Google sign-in behind the Sheets, Docs, Drive and Workspace integrations.

Connect

Credential fieldRequiredWhere it comes from
API key (API_KEY)YesFrom aistudio.google.com → Get API key (starts with AIza). It carries the quota, model access and billing of the Google Cloud project that owns it.
SettingRequiredWhat it is
Default model (DEFAULT_MODEL)NoModel id used when a step names none, e.g. gemini-3.6-flash. Pass it bare, without the models/ prefix. Leave blank to use this integration's built-in default. A step can always override it.

Tools

ToolAccessWhat it does
List models (yekar.google-gemini.list-models)ReadThe Gemini models this API key can reach, with each one's token limits and the methods it supports. This is the source of truth for the model argument on the other tools: availability varies by project and by region, and an embedding-only model will not answer a prompt no matter how it is called. Ids come back bare and ready to use - the models/ prefix Google's API returns is part of the resource name, not the id.
Generate text (yekar.google-gemini.generate-text)ReadSend a prompt to a Gemini model and return its answer. One request, one answer - this tool holds no conversation state, so anything the model needs to know belongs in prompt or system. Use it for a second opinion from a different model family, for work a specific Gemini model is better at, or wherever an automation needs Gemini text specifically. For an answer grounded in live web results with citations, use Answer with search; for a result that must match a JSON schema, use Extract data.
Extract data (yekar.google-gemini.extract-data)ReadAsk a Gemini model for an answer shaped by a JSON Schema, and get back the parsed object rather than prose to hand-parse. Use it to pull fields out of unstructured text, classify against a fixed set of labels, or produce a record another step consumes. Gemini enforces the schema's shape but ignores value constraints such as minimum and maxLength, so validate those yourself if they matter. A blocked or cut-off answer are the two ways to get no data, and both are reported explicitly.
Answer with search (yekar.google-gemini.answer-with-search)ReadAnswer a question with Gemini grounded in live Google Search results, returning the answer together with the pages it is based on. Use it for anything where current information changes the answer - recent events, current prices or versions, anything past the model's training cutoff. Unsearched answers are rejected by default; set require_search: false to knowingly accept a plain Gemini answer with empty citations. Billed as a grounded request, which costs more than a plain generation.
Generate image (yekar.google-gemini.generate-image)ReadGenerate an image from a text prompt with an image-capable Gemini model. The image is saved as an attachment on this session - it is not returned inline, because an image is bytes and the result of a tool call is text. One image per call; any commentary the model wrote alongside it comes back as text.
Count tokens (yekar.google-gemini.count-tokens)ReadMeasure how many input tokens a prompt uses on a given Gemini model, without generating anything or spending output tokens. Use it to size a prompt against a model's input limit before sending it (List models reports that limit), to decide whether text needs splitting, or to estimate cost. Counts are model-specific, so the answer is only valid for the model named.

Notes

  • In Google AI Studio (aistudio.google.com), pick or create the Google Cloud project Yekar.AI should use - a key inherits that project's quota, model access and billing, so pick it deliberately.
  • Go to Get API key → Create API key, and copy it (it starts with AIza).
  • Paste it here. A key on the free tier has both per-minute and per-day quotas; a per-day limit does not clear on retry, so enable billing on the project for production use.
  • Set a default model if you want one model used whenever a step names none; otherwise the tools fall back to a built-in default and every result reports which model actually ran.
  • Call List models to confirm what this key reaches. Pass ids bare - gemini-3.6-flash, not models/gemini-3.6-flash - and note that image generation needs an image-capable model, which is not the same one that answers prompts.
  • Answer with search bills as a grounded request and returns Google's Search Suggestions markup, which the Grounding with Google Search terms require you to display alongside the answer.
  • Add one credential per Google Cloud project - production and evaluation projects can live on the same connection.