Content Classification

Content classification assigns an input to exactly one of several categories. Because each call is a single fast forward pass, you can classify every message, ticket, or document as it arrives.

This is a multi-class decision. The model returns a probability for each class, and the probabilities sum to 1.

Typical uses

• Content moderation and safety triage
• Intent detection in chat or voice agents
• Sentiment analysis
• Support ticket triage and tagging

note

The snippets below come from the Experiments reference, which covers the decision mode in full.

---

Create the classifier

Each class has a name and a description. Write descriptions so the classes are mutually exclusive.

import json
import requests

response = requests.post(
    'https://api.withemissary.com/v1/experiments',
    headers={
        'Content-Type': 'application/json',
        'X-API-Key': YOUR_API_KEY
    },
    data=json.dumps({
        'name': 'MyDecision',
        'mode': 'decision',
        'classes': [
            {'name': 'positive', 'description': 'The text expresses a positive sentiment, satisfaction, or approval.'},
            {'name': 'negative', 'description': 'The text expresses a negative sentiment, dissatisfaction, or complaint.'},
            {'name': 'neutral',  'description': 'The text is factual or does not express a strong opinion.'}
        ]
    })
)
print(response.json())

{"id": "ex-ahejacuehandhehe", "latest_version": "0.0.0"}

Call the classifier

response = requests.post(
    'https://api.withemissary.com/v1/classification',
    headers={
        'Content-Type': 'application/json',
        'X-API-Key': YOUR_API_KEY
    },
    data=json.dumps({
        'model': 'ex-ahejacuehandhehe/0.0.0',
        'input': 'This product is absolutely wonderful!',
        'data_format': 'probs'
    })
)
print(response.json())

{
    "id": "classify-eac7451c54174ebc8da47be77dfe9581",
    "model": "ex-ahejacuehandhehe/0.0.0",
    "data": [{
        "index": 0,
        "probs": {
            "negative": 0.017637422308325768,
            "neutral":  0.024233082309365273,
            "positive": 0.9581295251846313
        }
    }],
    "created": 1779907078
}

---

Tips

• Add an escape class. Include other or none so off-topic inputs have somewhere to go instead of being forced into a real category.
• Act on confidence. Auto-handle high-confidence results and send the uncertain middle to review.
• Move to a LoRA fine-tune for high-stakes tasks. Moderation and compliance deserve a model trained on your labeled data. Start with quick-train to see how much your labels help, then see LoRA fine-tune and Datasets.