Data Aggregator lets you build and run custom machine learning models directly on the platform. Choose from 18 algorithms across four categories: classifiers that assign category labels like spam detection, sentiment analysis, and intent sorting; regressors that predict numeric values like prices, scores, and demand forecasts; clusterers that discover natural groupings in unlabeled data; and anomaly detectors that flag unusual inputs after training on clean data. After training, all predictions run locally with zero per-request AI costs.

Models are organized into pipelines. A single-step pipeline works like a standalone model. Multi-step pipelines chain models together, where each step's output feeds directly into the next step's input. Enable live training and every model in the sequence learns from real predictions automatically. Models that support incremental training can learn one example at a time through the admin panel or the API. Batch-trained models use a managed dataset file and retrain on demand.

The app also includes two AI-powered data analysis commands. The aggregate data command accepts any data array through the API and returns structured analysis with patterns, groupings, and outlier detection. The conversation consolidation command compresses long conversation histories from chatbot, live operator, SMS, and email channels into compact summaries while keeping the most recent messages verbatim, preserving full context without the storage and processing overhead.

Everything is accessible through the API and Chain Commands. Call predict through a pipeline, send data for AI analysis, consolidate a conversation, or train a model, and wire any of them into automated workflows. Classify incoming messages with a pipeline, branch on the result in Chain Commands, and trigger different follow-up actions per category.

• 18 local ML model types across classifiers, regressors, clusterers, and anomaly detectors running on Rubix ML.
• Build pipelines that chain multiple models together, passing each prediction as input to the next step.
• Incremental training for supported models, train one row at a time without rebuilding from a full dataset.
• Batch-trained models use a managed dataset stored in S3 with retrain on demand through the admin panel.
• Live training mode trains every model in the sequence automatically from real predictions.
• Send any data array to AI for analysis and get structured results with patterns, groupings, and outlier detection.
• Conversation consolidation compresses long chatbot, SMS, email, and live operator histories while keeping recent messages verbatim.
• All functions available as API commands: predict, aggregatedata, summarizeconversation, trainmodel.
• Full Chain Commands integration for building automated workflows that combine ML predictions with AI analysis and other app commands.