# Introducing Potato 2.0: AI-Powered Annotation

Source: https://www.potatoannotator.com/blog/introducing-potato-2-0

> **Note:** This post describes Potato 2.0 as it was at release. Some configuration keys and features have been updated in later versions. See the [current documentation](/docs) for up-to-date configuration syntax.

Potato 2.0 is here, and it is the biggest update we have shipped. The headline additions are AI-powered features, broader multimedia support, and an active learning module. Together they change how you build training data.

## What's new in Potato 2.0

### AI-powered annotation assistance

Potato 2.0 connects to OpenAI GPT-4, Anthropic Claude, and Google Gemini. Once you wire one up, the model can suggest annotations that the annotator is free to accept, edit, or throw out. It can highlight important terms and phrases so attention lands where it should. And it can flag potential problems with an annotation as the annotator works, without taking the decision out of their hands.

```yaml
# Enable AI assistance in your config
ai_support:
  enabled: true
  endpoint_type: openai

  ai_config:
    model: gpt-4
    api_key: ${OPENAI_API_KEY}

  features:
    hints:
      enabled: true
    keyword_highlighting:
      enabled: true
    label_suggestions:
      enabled: true
```

### Multimedia support

Potato started as a text tool, and that is still the core. Now it also handles images (classification, bounding boxes, polygons, and keypoint detection), audio (waveform display, transcription review, speaker diarization), and video (frame-by-frame navigation, temporal event marking, object tracking).

### Active learning

The new active learning module can cut annotation effort by up to 50%. It plugs into any sklearn classifier to surface the items the model is least sure about, retrains as new annotations come in so the sampling keeps improving, and lets you set the classifier and feature extraction in YAML.

## Migration from Potato 1.x

Upgrading from Potato 1.x is painless. Your existing YAML configs still work, and there is a migration tool that helps you pick up the new features:

```bash
# Upgrade your installation
pip install --upgrade potato-annotation

# Run the migration helper
potato migrate config.yaml --to-v2
```

## Performance improvements

Page rendering is faster, large datasets load more smoothly, and session management holds up better when several annotators are working at once.

For the full rundown of what changed, see the [2.0 release notes](https://github.com/davidjurgens/potato/blob/master/docs/releasenotes/v2.0.0.md).

## Getting started

Installation is one line:

```bash
pip install potato-annotation
potato start your_config.yaml
```

The [Quick Start Guide](/docs/getting-started/quick-start) has a full walkthrough, and the [Showcase](/showcase) has example configs to copy from.

## What's next

Potato has kept moving since 2.0:

**Potato 2.1** added instance display, visual AI support (YOLO, Ollama Vision, OpenAI Vision), span linking, and layout customization.

**Potato 2.2** (now available) adds 9 new annotation schemas (event annotation, entity linking, triage, pairwise comparison, coreference chains, conversation trees, and more), a pluggable export CLI, MACE annotator competence estimation, 55 validated survey instruments, and remote data sources.

[Read about Potato 2.2 →](/blog/potato-2-2-release)

Thanks to the researchers and practitioners who have shaped Potato along the way. Your feedback is what drives the roadmap.

---

*Have questions or feedback? Join our [GitHub Discussions](https://github.com/davidjurgens/potato/discussions) or reach out on Twitter [@PotatoAnnotation](https://twitter.com/PotatoAnnotation).*