Predictions

A live, unified log of every inference request across your API keys, SDKs, CLI, and Playgrounds — with status, latency, cost, source, and full output inspection.

The Predictions page is your real-time audit log of every model inference request associated with your account — aggregated from all clients into a single searchable list.

In the Skytells Console, go to AI → Predictions in the left sidebar.

Predictions list

The main view is a table of predictions in reverse-chronological order (most recent first). Each row represents a single model execution.

List columns

Column	Description
Status	The current lifecycle state of the prediction.
ID	A truncated unique identifier for the prediction (e.g., `05bd8fe1-7646-4b3d-8...`). Click to inspect the full record.
Model or Deployment	The model namespace that was invoked — e.g., `FLUX.2-pro`, `truefusion-pro`, `flux-fast`, `beatfusion-2.0`.
Source	Where the prediction originated: `API` (direct API call via key), `WEB` (from a Playground session in the Console).
Queued	Time the prediction spent in the queue before execution began. `—` if it started immediately.
Running	Actual execution time once the model started processing, in seconds (e.g., `5.51s`, `1.37s`).
Total	End-to-end elapsed time from submission to completion.

Prediction statuses

Status	Badge color	Meaning
`Succeeded`	Green	The model ran successfully and output is available.
`Processing`	Blue	The model is currently running.
`Queued`	Grey	The prediction is waiting for capacity.
`Failed`	Red	The model returned an error. Open the detail to see `error`.
`Cancelled`	Orange	The prediction was cancelled before it completed.

Filtering and search

The search bar at the top of the list filters predictions in real time. You can search by:

Model name or namespace — e.g., flux, truefusion, beatfusion
Prediction ID — paste the full UUID or a prefix
Status — filter by succeeded, failed, processing, etc.

Use the refresh button (↻) to reload the list and pull the latest status for in-flight predictions.

Select Load more at the bottom of the list to paginate through older predictions.

Predictions from all clients — API, SDK, CLI, and Playground sessions — appear in this list. The Source column shows where each prediction originated.

Inspecting a prediction

Click any row to open the prediction detail sidebar on the right side of the page. The sidebar provides a quick summary without leaving the list.

Field	Description
Prediction ID	The full UUID for this prediction, shown at the top. Copyable.
Model	The model namespace that was used.
Created at	The exact timestamp when the prediction was submitted.
Status	Current or final status of the prediction.
Predict time	The measured execution duration (e.g., `5.51s`).
Source	`API` or `WEB`.
Cost	The dollar amount charged for this prediction (e.g., `$0.02`).

Below the metadata, the sidebar shows two views:

JSON — The Prediction object as returned by the Predictions API.
Preview — A visual preview of the output (rendered image, playable video, or playable audio).

Opening the full prediction view

Select View full prediction at the bottom of the sidebar to open the dedicated prediction detail page.

Full prediction detail page

The dedicated prediction detail page gives you deep access to the complete prediction record.

Header bar

The header displays key metadata in a concise summary row:

Status badge — Succeeded, Failed, etc.
Source — API or WEB
Output type — Image, Video, Audio, Text
Predict time — measured execution duration
Date — when the prediction was created
Cost — the dollar cost charged

An action button — Try in Playground — appears on the right, loading this exact prediction's input into the Inference Playground so you can re-run or iterate on it.

Input panel (left)

The input panel shows every parameter submitted for this prediction:

Tab	Content
Input	All input fields as key-value pairs, as submitted at creation time.
JSON	The `input` field of the Prediction object as raw JSON.
Create	Opens the Playground pre-loaded with this model to create a new prediction.

Output panel (right)

Tab	Content
Preview	Rendered output — image inline, video with playback, audio with playback.
JSON	The `output` field of the Prediction object, including delivery URLs for generated assets.

Output files

Below the output panel, all generated files are listed with filename and file size. Each file is downloadable directly from the prediction detail page.

Source column

The Source column tells you where each prediction originated:

Label	Meaning
`API`	Created via the Predictions API, TypeScript or Python SDK, or CLI.
`WEB`	Submitted from the Inference Playground in the Console.

You can also ask Eve to open a specific prediction or summarize your recent inference history. Say "Show me my last prediction" or "What did prediction 05bd8fe1 return?" and Eve will navigate directly.

Inference Playground — run new predictions interactively; every run appears here.
Model Card — view model details and jump back to re-run from a prediction.
API: List Predictions — programmatic predictions list.
API: Get Prediction — retrieve a single prediction by ID.
TypeScript SDK — Predictions — client.predictions namespace reference.
Eve — ask Eve to surface predictions, navigate to a specific record, or summarize your inference activity.

How is this guide?