Skip to main content
skill CX Codex

ultralytics-platform — Claude Code skill: features and install

ultralytics-platform is a Claude Code skill for managing Ultralytics Platform resources, model training, and hosted inference. It activates on requests to upload datasets or models, train in the cloud, or deploy an endpoint.

Apache-2.0 ★ 1,216 updated 2026-10-24

Install into Claude Code · Plugin marketplace All methods and platforms ↓
/plugin marketplace add fcakyon/claude-codex-settings
/plugin install ultralytics-dev@claude-settings
Open on GitHub

What ultralytics-platform does

ultralytics-platform guides the agent through the Ultralytics Platform Python SDK to upload datasets and models, launch training, export models, and deploy endpoints. The ultralytics package handles local training and inference; the ultralytics-platform SDK manages Platform resources. Training with a Platform dataset or model uses ul://owner/... URIs.

Before API work, the skill directs the agent to consult the current OpenAPI specification. The SDK requires the ultralytics-platform package and an ULTRALYTICS_API_KEY environment variable, which Platform() reads. Cloud training, exports, deployments, and batch image processing can spend credits. Image deletion and clearing the trash are permanent.

What it helps with

  • upload a dataset or model to Ultralytics Platform
  • train on a Platform dataset using an ul:// URI
  • export a model or deploy an endpoint
  • find and download a Platform dataset
  • diagnose why a run is missing from Platform

Example prompts for the agent

  • Upload this model to Ultralytics Platform.
  • Train on the dataset at ul://owner/datasets/dataset.
  • Deploy an endpoint for this model on Platform.

How to install ultralytics-platform

1Plugin marketplace

/plugin marketplace add fcakyon/claude-codex-settings
/plugin install ultralytics-dev@claude-settings

This installs the whole ultralytics-dev plugin with all its skills

2Skills CLI (Vercel)

npx skills add fcakyon/claude-codex-settings --skill ultralytics-platform -a claude-code

3Manually: put the files in place

~/.claude/skills/ultralytics-platform/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/ultralytics-platform/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about ultralytics-platform

How do I install ultralytics-platform in Claude Code?

Add the marketplace with /plugin marketplace add fcakyon/claude-codex-settings, then install the skill with /plugin install ultralytics-dev@claude-settings.

What does ultralytics-platform do?

It helps an agent manage Ultralytics Platform models and datasets, launch training, export models, deploy endpoints, and use hosted inference or AI annotation.

What does ultralytics-platform need to run?

The SDK requires the ultralytics-platform package and an API key in ULTRALYTICS_API_KEY. API work should use the current OpenAPI specification.

Source text

SKILL.md · Apache-2.0
---
name: ultralytics-platform
description: This skill should be used when user asks to "upload my model to Ultralytics Platform", "push this run to the platform", "upload a dataset to platform", "download a dataset from platform", "search platform datasets", "start cloud training", "train on platform GPUs", "export a model on platform", "deploy a model endpoint", "run Moondream on Platform", "auto-annotate a Platform dataset", "run hosted AI inference", "why is my run not showing on platform", or mentions platform.ultralytics.com, ul:// URIs, ultralytics-platform, or ULTRALYTICS_API_KEY.
---

# Ultralytics Platform

Use `ultralytics` for local YOLO training and inference. Use the generated `ultralytics-platform`
Python SDK for Platform resources, hosted inference, and AI annotation. It handles authentication,
typed responses, retries, and errors.

## Read the live contract

Before API work, check the generated [API reference](https://platform.ultralytics.com/api/docs) or
`GET https://platform.ultralytics.com/openapi.json`. Treat the live OpenAPI document as authoritative
when examples disagree. These recipes were checked against API and SDK v0.1.62 on 2026-09-24.
Check the [SDK source](https://github.com/ultralytics/sdk) for generated method signatures.

```bash
uv pip install -U "ultralytics-platform>=0.1.62"
export ULTRALYTICS_API_KEY=ul_... # Settings > API Keys
```

`Platform()` reads `ULTRALYTICS_API_KEY`. The `ultralytics` package also reads the key saved by
`yolo login`. Never print or commit a key.

## Choose the interface

| Goal                                                       | Interface                        |
| ---------------------------------------------------------- | -------------------------------- |
| Track a run that has not started                           | `ultralytics` training callback  |
| Train with a Platform dataset or model                     | `ultralytics` with a `ul://` URI |
| Manage datasets, models, training, exports, or deployments | `ultralytics-platform` SDK       |
| Predict with model weights or a dedicated endpoint        | `client.models.predict` / `client.deployments.predict` |
| Preview Moondream or other AI labels on a stored image    | `client.images.predict`          |
| Save AI labels across a dataset                          | `client.datasets.create_batch`   |
| Use another language or inspect a new field                | Live OpenAPI                     |

### Live training and `ul://` URIs

Pass an owner-qualified project to stream a run:

```python
from ultralytics import YOLO

YOLO("yolo26n.pt").train(data="coco8.yaml", epochs=100, project="owner/project", name="run1")
```

`project=` is required. Without it, the callback exits before creating a Platform run. Use the
owner prefix for a team workspace.

```python
YOLO("ul://owner/project/model").train(data="ul://owner/datasets/dataset", epochs=100)
```

### SDK

Use a context manager and owner/name paths. Keep returned IDs for operations that require them,
including image operations, upload `assetId`, training `modelId`, and export IDs.

Responses have resource-specific shapes, not a generic envelope. Create calls return `id`, `owner`,
and the URL name at the top level. Detail calls wrap the resource under its type, such as `dataset`.
A rename changes the URL name, so use the name returned by the update response.

Read [references/recipes.md](references/recipes.md) for live-run diagnosis, finished-run upload,
dataset upload, hosted inference, Moondream and other AI annotation, and billable jobs.

## Invariants

- Confirm the target workspace with `client.account.summary()` and read the exact resource before a
  mutation. Team work requires an API key created in that workspace.
- Upload with a signed URL and `PUT` using the returned `headers`. Dataset ingest now verifies and
  completes the upload itself, so `upload.complete` is optional for datasets. Models still require it.
- Dataset ingest accepts one source: `sessionId`, `sourceUrl`, or a connected-storage `reference`.
  Set `targetSplit` when every incoming image must enter one split.
- Top-level model `metrics` accepts only the contract's named summary metrics. Per-epoch
  `trainResults[].metrics` accepts numeric metric names from `results.csv`.
- Hosted AI annotation uses a stored image ID and dataset class names. It is not a free-form chat or
  caption API. Single-image predictions are unsaved, while batch annotation writes labels.
- On `429`, wait for `Retry-After` before retrying. Do not invent fixed sleeps.

## Cost and destructive actions

Cloud training, model exports, deployments, and batch image processing can spend credits. Confirm
the requested scope and cost before an unapproved billable launch. Do not ask again when the user
has already authorized it. Check `client.billing.usage_summary()` for current usage and plan limits.
Training returns cost estimates, but not every create response includes a price.

Get approval for deletes outside the user's authorized scope. Project, dataset, and model deletes
move resources to 30-day trash. Image deletion and `client.lifecycle.delete_trash` are permanent.

© Fatih C. Akyon · Apache-2.0 · fcakyon/claude-codex-settings

Similar skills

All skills for AI and ML →

From the same set113

The whole set →
simplify simplify runs four parallel agents to inspect and clean up recent diffs across reuse, complexity, and performance before committing. CX Codex +1 ★ 1,216repo stars navigation-native-navigators navigation-native-navigators is a Claude Code rule that steers React Native navigation toward native components. It covers stacks, tabs, and screen headers. CX Codex +1 ★ 1,216repo stars frontend-testing-debugging frontend-testing-debugging guides an agent through visual debugging: reproduce the interaction, inspect console and layout, fix the root rule, rerun the journey that broke. CX Codex +1 ★ 1,216repo stars programmatic-seo programmatic-seo plans scalable template-driven page generation from data sets and protects your project from thin-content deindexing penalties. CX Codex +1 ★ 1,216repo stars ultralytics-branding ultralytics-branding loads Ultralytics brand rules: colors, typography, logo specs and tone for PDFs, slides, documents, marketing emails and social posts. CX Codex +1 ★ 1,216repo stars email-design email-design — a skill for Claude Code and Codex that handles transactional email layout. Styles account confirmations, password resets and billing receipts to match your application branding. CX Codex +1 ★ 1,216repo stars commit-creator commit-creator reviews staged changes, plans a single commit or a logical series, and writes concise commit messages. Available for Claude Code and Codex. CX Codex ★ 1,216repo stars site-architecture The site-architecture skill for Claude Code and Codex plans or restructures site structure, including page hierarchy, navigation, URL patterns, breadcrumbs, and internal links. CX Codex +1 ★ 1,216repo stars

Project sponsors

Companies that keep this analytics open