Ellipse Gradient for Header

Google Vertex AI Integration Guide

Integrate enterprise applications with Vertex AI through Google Cloud REST APIs, model endpoints, asynchronous jobs, Cloud Storage, and BigQuery.

Google Vertex AI integration options at a glance

Google Vertex AI provides regional REST APIs for models, endpoints, deployed models, datasets, predictions, generative AI requests, and asynchronous jobs. Online prediction supports synchronous model invocation, while batch prediction, training, tuning, and pipeline operations commonly return long-running operation or job references. Cloud Storage and BigQuery provide storage-backed input and output for many batch workflows. Authentication uses Google Cloud OAuth 2.0, service accounts, Application Default Credentials, workload identity federation, IAM, and selected API-key patterns. Martini can consume these APIs, expose controlled REST endpoints, schedule polling workflows, map model-specific payloads, and coordinate downstream processing.

Integration pointSupported by Google Vertex AI?Common use casesHow Martini supports it
REST APIsYesManage Projects, Locations, Models, Endpoints, Deployed models, Datasets, Jobs, predictions, generative AI requests, and long-running Operations through regional Google Cloud endpoints.Martini can consume the Vertex AI REST API, externalize the project and location, map request and response schemas, and expose its own REST API as a controlled façade.
Bulk / async / batch APIsYesCreate BatchPredictionJob resources and coordinate training, tuning, pipeline, and other asynchronous operations.Martini can submit jobs, persist operation references, poll status on a schedule, retrieve results, and route failures or completions.
File / attachment APIsLimitedCloud Storage can provide input and output for batch prediction and other storage-backed workflows; Vertex AI does not provide one universal attachment API.Martini can coordinate object references and file-oriented workflows using API calls and configured storage endpoints, while applying feature-specific validation.
Database / analytics accessLimitedBigQuery can supply batch prediction input and receive batch prediction output; Vertex AI is not a general-purpose database interface.Martini can orchestrate BigQuery-backed workflows and map tabular results into downstream application payloads.
Webhooks / outbound callbacksLimitedSelected Google Cloud or Vertex AI features may support notifications, but there is no general webhook for every model, endpoint, job, or prediction event.Martini can receive documented notifications where available, but its default pattern is scheduled polling of Operations or Jobs with controlled retries.
AuthenticationYesGoogle Cloud OAuth 2.0, service accounts, ADC, IAM, workload identity federation, and selected API-key patterns authenticate access.Martini can store credentials in secrets or environment configuration and apply OAuth or other confirmed authentication settings without embedding keys in workflows.
SDKs and client librariesYesGoogle provides language-specific libraries for several Vertex AI APIs, including Java, Python, Node.js, Go, and .NET depending on the API.Martini can use the documented REST APIs directly; custom JVM-compatible logic can be considered only where REST orchestration is insufficient.

How Google Vertex AI exposes data and business events

Google Vertex AI REST APIs

Vertex AI exposes regional REST APIs for resource management, online prediction, generative AI requests, model deployment, datasets, jobs, and long-running operations. Request schemas vary by model and feature.

Martini implementation pattern

Martini consumes the relevant REST endpoint using configured Google Cloud authentication, constructs the regional host from environment configuration, validates the model-specific payload, and maps the response into a canonical application structure.

Implementation sequence

Receive an application request or scheduled trigger
Resolve the configured project, location, model, and endpoint
Authenticate the REST request with approved Google Cloud credentials
Validate and map the model-specific request
Invoke the Vertex AI REST API
Validate and transform the response before returning or persisting it

Batch and asynchronous jobs

Vertex AI supports BatchPredictionJob resources and other asynchronous operations for batch prediction, training, tuning, and pipelines. These operations commonly return a job or operation reference rather than a final result.

Martini implementation pattern

Martini submits the job, records the reference and source correlation key, then uses a scheduled workflow to poll status. After successful completion, it reads output from Cloud Storage or BigQuery and routes the result to downstream systems.

Implementation sequence

Identify a new input partition or processing request
Validate storage, dataset, region, and model configuration
Submit the Vertex AI job
Persist the returned job or operation reference
Poll status with bounded backoff
Read completed output from Cloud Storage or BigQuery and update downstream systems

Feature-specific notifications

Some Google Cloud or Vertex AI features may support event-driven notifications, but Vertex AI does not provide universal webhook coverage for every resource or operation. Notification availability must be confirmed for the selected feature.

Martini implementation pattern

Where a documented notification exists, Martini can receive the event and use it as a signal to retrieve the authoritative Vertex AI resource. Otherwise, Martini uses scheduled polling rather than assuming a callback exists.

Implementation sequence

Verify notification support for the selected Vertex AI feature
Receive the documented event or start a scheduled polling workflow
Retrieve the authoritative job or operation resource
Map the completed result to the downstream model
Apply idempotency and business rules
Persist the outcome and operational correlation data

Cloud Storage and BigQuery exchange

Cloud Storage and BigQuery commonly provide input and output locations for Vertex AI batch prediction and related data workflows. Their use is feature-specific and requires aligned permissions and regional configuration.

Martini implementation pattern

Martini coordinates source data availability, Vertex AI job creation, output completion checks, and downstream distribution. It treats storage references and query results as integration data rather than assuming Vertex AI offers a general file attachment service.

Implementation sequence

Validate the source object or BigQuery input
Check permissions, format, region, and retention requirements
Create the relevant Vertex AI operation
Poll until the job completes
Read the configured output location
Transform and distribute the result

Common Google Vertex AI integration patterns

Pattern 1: Invoke Vertex AI for online prediction

When to use this pattern

Use this pattern when an internal application needs a synchronous classification, extraction, generation, document analysis, or custom-model prediction response.

Integration direction
Application
Martini
Google Vertex AI
Application
Example Mapping
Google Vertex AI FieldCanonical FieldTarget Field
request.textinput.contentVertex AI model input
request.sourceIdcorrelation.sourceIdrequest metadata
prediction.resultenrichment.valueapplication response
Martini implementation pattern

Martini exposes or consumes an application API, validates the request, selects the configured region and model, invokes the appropriate Vertex AI endpoint, validates safety and response structure, and returns or persists the approved result. Transient failures receive bounded retries; invalid payloads and authorization failures are routed without blind retry.

Martini capabilities used
  • REST API exposure
  • API consumption
  • data mapping
  • validation
  • business rules
  • error handling

Pattern 2: Process batch prediction results

When to use this pattern

Use this pattern for large data volumes or asynchronous inference where input and output are stored in Cloud Storage or BigQuery.

Integration direction
Cloud Storage or BigQuery
Martini
Google Vertex AI
Cloud Storage or BigQuery
Example Mapping
Google Vertex AI FieldCanonical FieldTarget Field
input.uribatch.inputLocationBatchPredictionJob inputConfig
input.modelmodel.identifierBatchPredictionJob model
job.nameprocessing.correlationIdMartini checkpoint
output.uribatch.outputLocationdownstream result source
Martini implementation pattern

A scheduled Martini workflow discovers eligible input, creates a BatchPredictionJob, stores its reference, and polls until completion. It checks for an equivalent active or completed job before submission, reads the configured output, transforms results, and records failures for replay.

Martini capabilities used
  • scheduler triggers
  • workflows
  • API consumption
  • mapping
  • state and correlation handling
  • retry and error handling

Pattern 3: Enrich enterprise records with model output

When to use this pattern

Use this pattern when Salesforce, ServiceNow, Jira, or another application needs summaries, classifications, extracted fields, or suggested responses.

Integration direction
Enterprise application
Martini
Google Vertex AI
Enterprise application
Example Mapping
Google Vertex AI FieldCanonical FieldTarget Field
record.idsource.recordIdcorrelation key
record.descriptionai.inputTextmodel prompt or instance
model.outputvalidated.enrichmentrecord summary or classification
Martini implementation pattern

Martini retrieves selected fields, keeps the source identifier and model metadata, invokes Vertex AI, treats generated output as untrusted input, applies validation and business rules, and writes only approved results back to the application. Duplicate updates are prevented through source and model-version correlation.

Martini capabilities used
  • API consumption
  • data mapping
  • validation
  • business rules
  • workflow orchestration
  • idempotency handling

Pattern 4: Orchestrate model and endpoint lifecycle

When to use this pattern

Use this pattern for controlled deployment checks, endpoint status monitoring, model rollout coordination, traffic allocation changes, or operational notifications.

Integration direction
Deployment process
Martini
Google Vertex AI
Operations platform
Example Mapping
Google Vertex AI FieldCanonical FieldTarget Field
model.versiondeployment.modelVersiondeployed model
endpoint.statusdeployment.stateoperations status
traffic allocationdeployment.routingrelease configuration
Martini implementation pattern

Martini checks endpoint and deployed-model state, applies environment-specific rules, and performs permitted lifecycle calls. Administrative permissions are separated from prediction permissions, and failed operations generate correlated operational records rather than repeated uncontrolled changes.

Martini capabilities used
  • workflows
  • API consumption
  • business rules
  • environment configuration
  • error handling
  • monitoring

Applications commonly integrated with Google Vertex AI

Vertex AI can be coordinated with Google Cloud services and enterprise applications to invoke models, process batch data, enrich business records, and distribute results. The exact implementation depends on the APIs exposed by each adjacent application and the Vertex AI feature being used.

Application Scenario Direction Martini Pattern
Google Cloud Storage Store documents, model artifacts, batch prediction inputs, and generated outputs. Google Cloud Storage → Martini → Google Vertex AI Martini detects or receives references to new objects, validates storage and regional configuration, submits an online or batch Vertex AI request, and routes completed results back to Cloud Storage or downstream systems.
BigQuery Supply structured batch prediction data, store prediction results, and support analysis of model outputs. BigQuery → Martini → Google Vertex AI A Martini workflow retrieves or coordinates a BigQuery source, creates a Vertex AI BatchPredictionJob, polls its status, and reads or distributes the resulting BigQuery output after completion.
Pub/Sub Distribute selected Google Cloud or application events that initiate asynchronous processing. Pub/Sub → Martini → Google Vertex AI Martini consumes an event or scheduled work item, verifies that the relevant Vertex AI feature supports notifications, invokes the appropriate REST API, and publishes or persists the result through a controlled workflow.
Cloud Run Host application services that invoke Vertex AI or coordinate runtime-specific processing. Cloud Run → Martini → Google Vertex AI Martini provides orchestration around Cloud Run and Vertex AI calls, externalizing credentials, model configuration, validation, retries, and correlation identifiers rather than embedding these concerns in each service.
Salesforce Enrich Leads, Cases, or Knowledge content with classification, summarization, extraction, or generated text. Salesforce → Martini → Google Vertex AI Martini retrieves Salesforce data through its APIs, maps selected fields to a model-specific Vertex AI request, validates the response, and writes approved enrichment back to Salesforce with source identifiers for replay and audit.
ServiceNow Summarize Incidents, classify requests, and generate suggested knowledge or response content. ServiceNow → Martini → Google Vertex AI A Martini workflow reads ServiceNow records, invokes Vertex AI with controlled prompts or prediction payloads, applies validation and business rules, and updates the originating record or a related knowledge workflow.
Jira Classify or summarize Issues to support triage, routing, and project reporting. Jira → Martini → Google Vertex AI Martini consumes Jira issue data, applies a model-specific mapping and safety checks, invokes Vertex AI, and writes classification or summary fields back only when the response passes configured rules.

How to build a Google Vertex AI integration in Martini

Objective

Configure the Google Cloud project, location, API host, model or endpoint identifiers, and approved authentication method for each environment.

Instructions in Martini

  • Store OAuth, service-account, workload identity, or API-key configuration in secrets and environment settings
  • Use least-privilege IAM permissions appropriate to prediction or administration
  • Keep project, region, model, and endpoint values configurable

Objective

Select an API request, application event, storage or data availability signal, or scheduled workflow based on the Vertex AI operation.

Instructions in Martini

  • Use a Martini API for synchronous application requests
  • Use a scheduler for polling and batch coordination
  • Use feature-specific notifications only when documented

Objective

Collect the source payload, Cloud Storage object, BigQuery data, or application record required by the selected model operation.

Instructions in Martini

  • Validate required fields and source availability
  • Capture source identifiers and processing version
  • Check storage permissions, format, and regional alignment

Objective

Coordinate request construction, Vertex AI invocation, long-running operation tracking, and downstream processing in a maintainable workflow.

Instructions in Martini

  • Call the appropriate regional REST endpoint
  • Persist job or operation references
  • Separate model-specific mappings from reusable orchestration logic

Objective

Transform business data into the selected model schema and validate model output before it reaches operational systems.

Instructions in Martini

  • Apply model-specific request and response mappings
  • Validate generated content, safety conditions, required fields, and output types
  • Use canonical fields and retain correlation metadata

Objective

Apply business, permission, idempotency, and routing rules around model invocation and result handling.

Instructions in Martini

  • Avoid duplicate job creation and duplicate downstream updates
  • Route authorization, validation, quota, and safety failures differently
  • Restrict sensitive prompt and response logging

Common Google Vertex AI data objects used in integrations

ObjectTypical UseCommon target systemsMartini handling
ProjectsContain Vertex AI resources, quotas, IAM policies, and billing configuration.Google Cloud administration, governance platforms, deployment toolingMartini keeps project identifiers environment-specific and uses them in authenticated API requests and operational checks.
ModelsRepresent foundation, publisher, custom-trained models, and model versions used for prediction or generative AI.Model registries, deployment workflows, business applicationsMartini validates model identifiers, applies model-specific request mappings, and records model metadata with correlation details.
EndpointsExpose deployed models for online prediction and manage deployment configuration.Application APIs, release workflows, monitoring systemsMartini invokes configured endpoints, checks status, and can coordinate deployment or undeployment where IAM permissions allow.
Deployed modelsAssociate models with endpoints, traffic allocation, and machine configuration.Release management, operations, capacity managementMartini retrieves deployment state, applies environment-specific rules, and routes lifecycle failures for review.
BatchPredictionJobRepresent asynchronous batch inference using Cloud Storage or BigQuery input and output.Cloud Storage, BigQuery, data platforms, reporting systemsMartini creates the job, stores its reference, polls status, retrieves results, and prevents duplicate submissions with correlation keys.
OperationsTrack long-running training, tuning, pipeline, and other asynchronous API activities.Schedulers, monitoring systems, operational notificationsMartini persists operation names, polls status, distinguishes transient from permanent errors, and starts downstream processing after completion.

Authentication and security considerations

Google Cloud identity and access

Vertex AI uses Google Cloud authentication and authorization. Common options include OAuth 2.0, service accounts, Application Default Credentials, workload identity federation, IAM roles, and selected API-key patterns.

Credential protection

Martini should store credentials in environment configuration or secrets rather than workflow definitions. Use least-privilege permissions and separate prediction access from administrative permissions such as model deployment.

Data protection

Prompts, documents, prediction inputs, and generated responses may contain sensitive information. Limit logging, control retention, validate generated content, and configure regional resources consistently with data residency requirements.

Operational considerations for Google Vertex AI integrations

Quotas and regional configuration

Google Cloud applies quotas and model-specific limits for requests, tokens, concurrency, job submission, payload size, and regional capacity. Keep project, location, model, endpoint, and API host configurable.

Asynchronous processing

Training, tuning, batch prediction, and other operations may return references instead of immediate results. Persist references and poll with bounded backoff, or use a documented feature-specific notification mechanism.

Reliability and idempotency

  • Use correlation keys based on the source, model, input version, and processing attempt.
  • Handle pagination and returned page tokens for list operations.
  • Retry transient failures and quota responses, but not invalid requests or permission failures.
  • Check for equivalent jobs before creating new asynchronous work.

Schema and testing

Model schemas vary across generative, custom, embedding, multimodal, online, and batch operations. Keep mappings model-specific, validate responses, and test region, permissions, safety behavior, storage access, and downstream failure paths.

Why use Martini instead of scripts or point-to-point integrations?

Centralized orchestration

Martini coordinates Vertex AI calls, enterprise application APIs, Cloud Storage, BigQuery, scheduled polling, and downstream updates in reusable workflows instead of duplicating logic across scripts.

Controlled transformation

Mappings, validation, business rules, model-specific schemas, correlation identifiers, and idempotency handling can be maintained as explicit integration logic.

Operational reliability

Martini provides structured error handling, retries, scheduling, monitoring, and environment configuration for long-running and quota-sensitive workloads.

Reusable API assets

Martini can expose a controlled REST API façade for internal applications while keeping Google Cloud credentials, regional configuration, model selection, and Vertex AI implementation details behind the integration boundary.

Frequently asked questions

How can Google Vertex AI be integrated with enterprise systems?

Vertex AI can be integrated through Google Cloud REST APIs for online prediction, generative AI requests, model and endpoint management, datasets, and asynchronous jobs. Cloud Storage and BigQuery commonly provide batch input and output, while long-running Operations or Jobs support status tracking.

Can Martini integrate with Google Vertex AI?

Yes. Martini can consume the documented Vertex AI REST APIs, authenticate with configured Google Cloud credentials, map model-specific requests, invoke endpoints, coordinate asynchronous jobs, poll status, and route validated results to enterprise applications.

Do I need a connector to integrate Google Vertex AI with Martini?

No. A dedicated Google Vertex AI connector is not required. Martini can use Vertex AI's native REST APIs, Google Cloud authentication methods, Cloud Storage and BigQuery workflows, and feature-specific notification mechanisms where confirmed.

Is there any extra Lonti cost to integrate Google Vertex AI with Martini?

Lonti does not charge an additional per-connector or per-vendor fee to integrate Google Vertex AI. Integrations are subject to the provisioned capacity of the Martini environment. Separate costs may apply from Google Cloud, infrastructure, or other third-party systems based on subscription, usage, and deployment model.

Which Google Vertex AI integration methods should be used?

REST APIs are the primary integration method. Use online prediction APIs for synchronous inference and batch or asynchronous APIs for large-volume, training, tuning, and pipeline workloads. Cloud Storage and BigQuery are commonly used for batch data exchange.

Does Google Vertex AI provide webhooks or callbacks?

Vertex AI does not provide universal webhook coverage for every model, endpoint, job, or prediction operation. Some feature-specific Google Cloud notification patterns may exist, but the dependable general pattern is to store an Operation or Job reference and poll it with a scheduled Martini workflow.

How does synchronization with Google Vertex AI work?

Martini can synchronize through API-led requests, scheduled polling, and batch workflows. It captures project, region, model, endpoint, source identifiers, and job references, then retrieves results from Vertex AI, Cloud Storage, or BigQuery and applies idempotent downstream updates.

How does Martini handle Vertex AI mapping, errors, and retries?

Martini maps each model or operation with its own request and response schema, validates generated or predicted output, and applies business rules before writing results. It can retry transient failures and quota responses with backoff, while routing invalid requests, permission failures, and safety or validation failures without blind retries.