Google Vertex AI

Generate text and multimodal content with Gemini on Google Cloud's enterprise Vertex AI platform, create embeddings, generate images, and browse Model Garden.

7 actions available

New batch of documents lands in the processing bucket

Agent creates embeddings and retrieves relevant context

Agent asks Gemini on Vertex to summarize each document with key figures

Agent enforces a structured summary format

Agent files the summaries alongside the originals

Reviewers get a digest of the processed batch

A summary flagging an unusual figure routes to the owning analyst

What This Integration Enables

Vertex AI brings Gemini and Google's model catalog into a flow without leaving Google Cloud. Content generation covers text and multimodal input, embeddings power retrieval over your own data, image generation produces assets on demand, and Model Garden exposes the broader catalog, all under your project's identity and controls. Running inside GCP settles the data-residency question for a Google-native team: inference happens where the data already lives. The operational value comes from the layer that turns a Gemini call into a repeatable process with a review point. An orchestration layer owns that sequencing and the human gate, and FlowRunner is built for it.

Without FlowRunner

Models outside GCP Inference runs on a vendor outside your Google Cloud project

Documents pile up More reports arrive than anyone can read

No grounding Generated content with no link to source data

With FlowRunner

Inference inside GCP Model calls run under your project, region, and IAM

Documents summarized Every report comes with a structured brief

Retrieval-grounded output Answers built on your own retrieved content

Use Case Scenarios

Document Summarization Pipeline

Reports land in a processing bucket faster than anyone reads them. The agent embeds and retrieves related context, asks Gemini on Vertex for a structured summary with key figures, and files each brief beside its original. Reviewers read short briefs with the source one click away. When the model flags an unusual figure, the summary routes to the owning analyst rather than presenting it as settled.

Grounded Enterprise Q&A

Employees ask questions that only your internal data can answer, and that data must stay in GCP. The agent embeds the corpus, retrieves the relevant passages, and asks Gemini for a grounded answer inside the project. Confident answers return with sources; anything touching sensitive material routes to the owning team.

On-Demand Asset Generation

Teams need first-draft images for content and mockups. The agent generates candidates with Vertex image generation from a brief. The drafts go to a person for selection and refinement rather than publishing automatically, so generation speeds up the work without removing the creative judgment.

Human-in-Loop Highlight

A Gemini summary or a generated asset is a draft the flow produces, not a final artifact. When Vertex output will be acted on, published, or flags something unusual, FlowRunner routes it through a [human-in-the-loop](/concepts/human-in-the-loop/) step: the agent pauses, presents the output with its sources or its brief, and sends it to the right person via Slack. They review and decide. The model produces the draft inside your project; a person owns what leaves it.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via Slack

Human decides

Agent resumes with decision

Agent Capabilities

7 actions

Content Generation

Generate Content Generates text with a Gemini model on Vertex AI from a single prompt. Supports an optional system instruction plus temperature and max output token controls, and returns the generated text together with the finish reason and token usage.
Generate Content (Advanced) Generates content with the full Gemini feature set on Vertex AI: multimodal inputs (images, audio, video, PDFs sent inline from URLs), multi-turn conversation history, Google Search grounding with citations, custom function declarations, structured JSON output via a response schema, thinking budget control for reasoning models, safety settings, and full sampling controls (temperature, top-p, top-k, stop sequences, seed).
Count Tokens Counts the number of tokens a text prompt would consume for a given Gemini model on Vertex AI, without generating content or incurring generation cost.

Embeddings

Create Embeddings Generates dense vector embeddings for one or more texts using a Vertex AI embedding model (e.g. 'gemini-embedding-001').

Image Generation

Generate Image Generates images from a text prompt using an Imagen model on Vertex AI. Supports 1 to 4 images per request, a choice of aspect ratios, and an optional negative prompt (honored by older Imagen models only).

Model Garden

Call Partner Model Calls any partner or open model deployed through the Vertex AI Model Garden (Anthropic Claude, Meta Llama, Mistral, AI21, and others) via the rawPredict endpoint.
Predict Sends a generic prediction request to any Vertex AI predict target: a custom model deployed to an endpoint, or a publisher model that uses the instances/parameters schema.

Start building with Google Vertex AI

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us