Overview

Replicate is a cloud platform that lets developers run, fine-tune, and deploy open-source machine learning models with a single API call, without managing GPUs or model serving infrastructure. Persona's Replicate integration enables you to run any public or private ML model directly from your workflows — incorporating AI-powered analysis, scoring, or classification into your identity decisioning pipeline without building custom model infrastructure.

Benefits

Incorporate Custom ML Models into Identity Decisions: Run any open-source ML model — image analysis, NLP classification, risk scoring — directly from a Persona workflow step and use the result to inform downstream identity decisions.

No GPU Infrastructure Required: Replicate handles all model serving infrastructure. Your team can focus on building the right decision logic without managing compute resources.

Access a Broad Model Catalog: Search Replicate's public model catalog to find the right model for your use case — document classification, face analysis, text extraction, custom risk models, and more.

Support Asynchronous Model Execution: Create a prediction and poll for results asynchronously, allowing long-running model inferences to complete without blocking your workflow.

Cancel and Control Predictions: Cancel running predictions to avoid unnecessary compute charges if a workflow is terminated early or the result is no longer needed.

Integration Features

Persona's Replicate integration enables AI model execution within identity workflows. Key capabilities include:

• Search Models: Search Replicate's public model catalog by keyword.
• Get Model: Retrieve full details of a specific model by owner and name, including the latest version ID.
• Create Prediction: Run a model version with specified input parameters and initiate an asynchronous prediction job.
• Get Prediction: Poll prediction status and retrieve output when complete.
• Cancel Prediction: Cancel a running prediction to stop compute usage.

Setting up the Replicate integration

Prerequisites

To set up the Replicate integration, ensure you have:

• A Replicate account
• A Replicate API token (found in Account Settings > API tokens)

Setting up the Replicate credentials

1. In the Persona Dashboard, navigate to Integrations > Marketplace and click on Replicate.
2. Click on Add Credential at the top of the page.
3. Enter a Nickname for Credential in the modal, along with your Replicate API token, and click Continue.
4. Once complete, the credential will appear in Persona and can be selected in workflows.

Using the Replicate integration in a workflow

1. Create a new workflow, or open an existing workflow you'd like to update.
2. Add a new action step > Integrations.
3. Select the Replicate integration and choose your Replicate credential.
4. Use Search Models or Get Model to identify the model version ID you want to run.
5. Add a Create Prediction step, then a Get Prediction step to poll for results.
6. Map the model output to downstream workflow logic and save and publish the workflow.

Replicate Operations Overview

Persona supports multiple Replicate API operations for discovering, running, and managing ML model predictions. These actions allow you to incorporate any open-source machine learning model into your identity workflows — enabling custom AI-powered signals without managing model infrastructure.

See below for a comprehensive list of available Workflow Action steps and possible configurations for the Replicate integration:

Search Models

Search Replicate's public model catalog by keyword, returning model names, owners, descriptions, tags, and run counts.

Configuration Steps:

• No required fields.
• Optionally add:
  • Query (keyword search string)
  • Cursor (for pagination)

Get Model

Retrieve full details of a specific model by owner and name, including the latest version ID needed to run predictions.

Configuration Steps:

• Provide values for required fields:
  • Model Owner
  • Model Name

Create Prediction

Run a model version with specified input parameters, initiating an asynchronous prediction job. Returns a Prediction ID to poll for results.

Configuration Steps:

• Provide values for required fields:
  • Version ID (from Get Model)
  • Input (JSON object of model-specific input parameters)
• Optionally add:
  • Webhook URL (to receive results via webhook instead of polling)

Get Prediction

Retrieve the current status and output of a prediction by ID. Poll this action until status is succeeded or failed.

Configuration Steps:

• Provide values for required fields:
  • Prediction ID (returned from Create Prediction)

Cancel Prediction

Cancel a running prediction to avoid unnecessary compute charges.

Configuration Steps:

• Provide values for required fields:
  • Prediction ID

FAQs

How long do predictions take?

Prediction time varies by model and input size. Use the Get Prediction action in a loop or with a delay step to poll for completion. Most models complete within seconds to minutes.

Can I use private models?

Yes. Private models you own or have been granted access to can be run using Get Model and Create Prediction with the model's owner and name.

What happens if a prediction fails?

Get Prediction will return a status of failed with an error message. Build conditional logic in your workflow to handle failure cases.