Google AI Platform Training & Prediction API
Browse all Google APIs✓ Official Vendor SpecAI/MLMl Inferenceoauth229 EndpointsREST
For Agents
Train custom ML models, deploy versions, and serve online predictions or explanations from the original AI Platform Training & Prediction service.
Use for: Run an online prediction against an AI Platform model version, Get feature attributions for a prediction via the explain endpoint, Set a model version as the default for a model, Submit a hyperparameter trial measurement to an existing study
Not supported: Does not handle Vertex AI Pipelines, Matching Engine, or generative AI endpoints — use for AI Platform Training & Prediction models, versions, jobs, and hyperparameter trials only.
AI Platform Training & Prediction is Google Cloud's earlier-generation managed service for training custom machine learning models and serving online and batch predictions. The API exposes models, model versions, jobs, online predict and explain calls, hyperparameter trial measurement, and study management. Most teams new to Google Cloud now use the Vertex AI API for the same workloads, but AI Platform remains supported for existing pipelines.
Install Jentic One Beta
Connect the AI Platform Training & Prediction API to your agent
Jentic One is a self-hosted execution layer for AI agents. It lets your agent call the AI Platform Training & Prediction API, or any other public or private API you need. You set the rules, the agent never sees your credentials, and every call is logged.
Two steps, two machines. Install the instance in a safe environment, then register your agent from wherever it runs.
1
Step 1: Jentic One Host machine
# On the machine that will host your Jentic One instance:
curl -fsSL https://raw.githubusercontent.com/jentic/jentic-one/main/tools/install.sh | sh2
Step 2: Agent machine
# On the machine where your agent runs (keep this separate from the instance):
curl -fsSL https://raw.githubusercontent.com/jentic/jentic-one/main/tools/install.sh | sh
jentic register # connects your agent to your Jentic One instanceJentic One is in public beta. The setup above keeps your agent separate from the instance, which is what you want before using real credentials: an agent running as the same OS user as Jentic One can read its stored keys directly. Just evaluating? A single local install is fine to start. See the secure deployment guide for the tiers.
Capabilities
What an agent can do with AI Platform Training & Prediction API.
Run online prediction calls against deployed AI Platform model versions
Run online explain calls to surface feature attributions for a prediction
Manage AI Platform models and their versions with default-version selection
Submit and cancel training and batch prediction jobs
Manage hyperparameter studies and individual trials including measurements
Mark a trial complete or check the early-stopping state of a hyperparameter trial
Read AI Platform configuration for a project via getConfig
Use Cases
Patterns agents use AI Platform Training & Prediction API for, with concrete tasks.
★ Online prediction serving
Deploy a trained model to AI Platform and call the predict endpoint to score requests in real time. The endpoint accepts JSON instances and returns the model's predictions, with autoscaling and version routing handled by the platform. Useful for teams maintaining existing AI Platform deployments rather than migrating to Vertex AI.
Call predict on projects/acme/models/churn/versions/v3 with a JSON payload of customer feature instances and capture the prediction array
Hyperparameter tuning trials
Drive a hyperparameter study by submitting trial measurements via addMeasurement, marking trials complete, and consulting checkEarlyStoppingState to abort underperforming trials. The study and trial endpoints support custom search algorithms running outside AI Platform that still want to use the platform's coordination primitives.
Add a measurement to trial projects/acme/locations/us-central1/studies/lr-search/trials/12 with stepCount 50 and metrics validation_loss=0.42, then check early stopping state
Model explainability for compliance
Use the explain endpoint alongside predict so that customer-facing decisions can be paired with feature attributions, useful in regulated domains where every model decision must be auditable. Attributions reflect the explainability config attached to the model version.
Call explain on projects/acme/models/credit/versions/v2 with the same instance payload used for predict and return both prediction and attributions
Agent integration via Jentic
AI agents that orchestrate ML pipelines on legacy AI Platform deployments can use this API through Jentic to run predictions, manage versions, and drive hyperparameter searches without holding the OAuth refresh token in their prompt.
List versions of model projects/acme/models/churn and set the version with the highest deployedAt timestamp as default via setDefault
Key Endpoints
29 endpoints — ai platform training & prediction is google cloud's earlier-generation managed service for training custom machine learning models and serving online and batch predictions.
METHOD
PATH
DESCRIPTION
POST
/v1/{+name}:predict
Run an online prediction
POST
/v1/{+name}:explain
Run an online explain call
POST
/v1/{+name}:setDefault
Set a model version as default
POST
/v1/{+name}:addMeasurement
Add a hyperparameter trial measurement
POST
/v1/{+name}:complete
Mark a hyperparameter trial complete
POST
/v1/{+name}:checkEarlyStoppingState
Check trial early stopping state
POST
/v1/{+name}:cancel
Cancel a job or operation
GET
/v1/{+name}:getConfig
Read AI Platform config for a project
POST
/v1/{+name}:predict
Run an online prediction
POST
/v1/{+name}:explain
Run an online explain call
POST
/v1/{+name}:setDefault
Set a model version as default
POST
/v1/{+name}:addMeasurement
Add a hyperparameter trial measurement
POST
/v1/{+name}:complete
Mark a hyperparameter trial complete
POST
/v1/{+name}:checkEarlyStoppingState
Check trial early stopping state
POST
/v1/{+name}:cancel
Cancel a job or operation
GET
/v1/{+name}:getConfig
Read AI Platform config for a project
Why Jentic?
Three things that make agents converge on Jentic-routed access.
Credential isolation
Google Cloud OAuth 2.0 client credentials and refresh tokens are stored encrypted in the Jentic vault. Agents receive short-lived bearer tokens scoped to cloud-platform; the refresh token and client secret never enter the agent's context.
Intent-based discovery
Agents search Jentic by intent such as run online prediction or hyperparameter trial measurement and Jentic returns the matching operations with input schemas, so the agent does not have to read Google Cloud's discovery documents.
Time to first call
Direct integration with AI Platform takes 1-2 days for OAuth setup, version targeting, and trial coordination logic. Through Jentic the same workflows are under an hour: search, load schema, execute.
Related APIs
Alternatives and complements available in the Jentic catalogue.
FAQs
Specific to using AI Platform Training & Prediction API through Jentic.
What authentication does the AI Platform API use?
The API uses Google OAuth 2.0 with the https://www.googleapis.com/auth/cloud-platform scope. Through Jentic, OAuth credentials are held in the encrypted vault and the agent receives only a short-lived bearer token, never the refresh token.
Can I get feature attributions alongside predictions?
Yes, the explain endpoint returns attributions according to the explainability config attached to the model version. It accepts the same instance payload as predict, so you can call both for the same request when you need an auditable decision.
What are the rate limits for the AI Platform API?
Online prediction has its own per-project queries-per-minute quota separate from training and batch jobs, listed under the AI Platform service in the Cloud Console quotas page. High-throughput serving usually requires requesting a quota increase.
How do I run an online prediction through Jentic?
Search Jentic for run online prediction on ai platform, load the predict operation that maps to /v1/{+name}:predict, and execute with the model version resource name and a JSON instances payload. Jentic injects the OAuth bearer for the call.
Should I be using Vertex AI instead?
For new workloads, Google recommends Vertex AI, which supersedes AI Platform with a unified API for training, prediction, pipelines, and matching engine. AI Platform remains supported for existing deployments and is the right choice when you are not yet ready to migrate.
Is the AI Platform API free?
The API itself is not separately billed; you pay for prediction node-hours, training compute, and Vizier study usage. See the AI Platform pricing page for current rates by region and machine type.
GET STARTED