Start free

Replicate connector · Prediction reports

Read & reportRead-only source

See which Replicate models you run and for how long.

Bring Replicate predictions and deployments into your reports and AI assistants, with model, status, source and predict time for every prediction.

  • Total predict time by model each week
  • Count failed predictions by model
  • Ask ChatGPT or Claude which models take the longest
ReplicateSheetsSheetsData StudioData StudioChatGPTChatGPTClaudeClaude

Start free Explore the data

Updated November 2026

Replicate reportSample data
ModelStatusSource
black-forest-labs/flux-schnellsucceededapi
meta/meta-llama-3-8b-instructsucceededapi
openai/whisperfailedweb
Refreshes on your schedule in Google Sheets

Replicate reporting for one account

One Replicate account is the data source. Predictions follow your report dates; deployments and the account show it as it is now.

Predictions

Read predictions created in your report dates with model, status, source and predict time in seconds.

Deployments

See each deployment with owner, name and the model of its current release.

Account

Read the account type, username and display name.

Compute time

Add predict time by model to see where your compute goes.

Report datasets and details

Reports you can build

3 reporting datasets, each with its own fields. Pick one in the app, then choose your columns.

  • Model predictions
  • Deployments
  • Account info
What each dataset reports
  • Model predictions

    Replicate predictions created in the report dates with model, status, source, predict time in seconds, and creation date.

  • Deployments

    Replicate dedicated model deployments with owner, name, and active release model.

  • Account info

    Replicate account type, username, display name, and GitHub URL.

Metrics and dimensions

14 report fields across 3 datasets.

Search the available fields. Hover or tap a field to see what it means.

Send Replicate data anywhere

Connect Replicate to ChatGPT or Claude for model usage questions

Use Datapipe’s MCP access as your Replicate AI integration. Ask about predictions, failures and compute time.

Example

Which Replicate models used the most predict time last month?

Add up last month’s predict time by model, then sort from highest to lowest and check their failure counts.

Start with Replicate

What you need

A Replicate API token from Account settings, API tokens in Replicate.

What to know

Read-only. Datapipe never runs a prediction, so nothing is billed. Inputs and outputs aren’t included. Predict time is in seconds. One report reads up to 20,000 predictions.

How it works

  1. Create your free account, then choose Replicate in Data sources.

  2. Paste your Replicate API token.

  3. Choose a report and destination: a sheet, Data Studio, ChatGPT or Claude.

Works well with Replicate

Add both to the same project. Each keeps its own report, so you can check them side by side or ask ChatGPT or Claude a question that needs both.

Questions

Datapipe reads predictions in your report dates, your deployments and your account details.

No. It only lists predictions that already exist, so nothing is billed.

No. Reports cover model, status, source and predict time.

Whether the prediction was started from the API or the Replicate website.

Free to start. Paid plans add more data sources and refresh faster.

Free and Starter refresh once a day. Pro refreshes every hour. Agency refreshes every 15 minutes on the first 50 data sources.

One for each Replicate account you add to a project. Adding the same account to another project counts again. Reports and destinations do not use extra data sources.

Keep model usage in view.

Start free, add Replicate and bring its predictions into the tools you already use.

Start free See setup requirements