AutoTrain with Ask AI#
AutoTrain uses Ask AI to help you train, compare, and improve models through a conversation. Start with a baseline, review its results, and ask for the next experiment. Ask AI can also find datasets, annotate images, export models, and deploy them.
Sign in to Ultralytics Platform, open your dataset or project, and click Ask AI.
To use Platform actions, first create an Ultralytics API key in your personal workspace's Settings > API Keys. Ask AI uses your existing personal key automatically; you do not need to paste it into the conversation. Include the project or dataset link when working with a team workspace.
Prompts to Try#
Include your target metric, model, epoch limit, and number of experiments in the request. For comparable results, keep the dataset, splits, and random seed fixed while changing one setting at a time. Cloud runs use your training credits.
Open the relevant dataset or project and ask:
| Goal | Example prompt |
|---|---|
| Start a project | Create a private project called "Wildlife Lab". Find and clone a wildlife detection dataset, then prepare a baseline training run. |
| Compare resolutions | Train three models on this dataset at image sizes 640, 800, and 960. Keep all other settings and dataset splits fixed. |
| Explore a dataset | Find container ships, shipping containers, and cranes in xView. Is it a good starting point for port logistics? |
| Create a baseline | Train YOLO26n on this dataset and name the run "xview-baseline". |
| Compare models | Compare the completed models by validation mAP50-95. Explain the differences and propose the next experiment. |
| Deploy a model | Deploy the completed model with the best validation mAP in this project. Give me its endpoint URL and status. |
| Auto-annotate | Use my best compatible model to annotate unlabeled images in this dataset. Preserve existing annotations. |
| Review classification | Which model has the best classification accuracy? Which breeds remain confusing, and what should we try next? |
| Export a model | Export this model to ONNX and give me the download link when it's ready. |
Use Your Own Coding Agent#
Ask AI uses the Platform CLI, ul cloud, and its companion platform-cli skill. You can use the same commands from Claude Code, Codex, or another compatible coding agent.
Install and Connect#
The setup below is for using the Platform CLI with your own coding agent outside Platform.
The ul CLI comes with the Ultralytics package on Python 3.11+. Install or update it:
pip install -U ultralyticsCreate a Platform API key and set it in the terminal used by your agent. Check the connection with:
export ULTRALYTICS_API_KEY="YOUR_API_KEY"
ul cloud account summaryAdd the Skill#
Follow the Agent Skills installation guide for Claude Code and Codex plugins. For agents supported by the skills CLI, install the Platform skill with:
npx skills add ultralytics/skills --skill platform-cliThen use prompts like the examples above, including the project or dataset link for context.
Run Commands Directly#
Use your project and model URL slugs in commands. These examples target your personal workspace; add owner=TEAM to target a team workspace you belong to:
ul cloud datasets list
ul cloud models list project=license-plate
ul cloud models training project=license-plate model=baseline
ul cloud training start --helpUse ul cloud --help to discover commands and ul cloud <resource> <operation> --help for their arguments. See the Platform API reference for more details.