Data Aggregator - API Documentation

App Command (API Call)
POST your request to https://api.aiappsapi.com with your API key and account ID in the POST body. The fields described in each section below go inside the jsonData POST field as a JSON-encoded string. Use this method when your server is making the call and you have your API credentials available.

Mode / Webhook (Direct URL)
POST directly to the endpoint URL shown in each section. The fields go in the POST body as standard form fields (the same fields you would include in an HTML form POST). Use this method for provider webhooks, website forms, or any third-party system that sends form data. No API key is required - the account ID is part of the URL.

Run input data through a trained ML pipeline and get a prediction. The pipeline processes each model step in sequence, passing each step's output as the next step's input, and returns the final result. If the pipeline has live training enabled, each model in the sequence also learns from the prediction automatically. No cloud AI costs - all predictions run locally.

Send any data array to a cloud AI model for analysis. The AI produces a structured summary highlighting patterns, common groupings, outlier data, and key statistics. Results are also stored in your account reports for later retrieval. Use this for ad-hoc analysis of customer feedback, survey responses, transaction records, or any data collection.

Compress a long conversation history into a compact summary while keeping the most recent messages verbatim. The older portion is summarized into a narrative and merged with the recent messages so full context is preserved without the storage and processing overhead. Works with chatbot, SMS, email, and live operator conversations. You can pass messages inline or reference an existing conversation by app and conversation ID.

Run an AI-powered summary on a stored dataset using a named summary configuration. The summary config defines which dataset to analyze, which AI model to use, the maximum data size, and an optional custom prompt. Results are stored in your account reports keyed by date. This runs as a background job.

Train a specific model step within a pipeline. Models that support incremental training (KNN, Naive Bayes, DNN, Logistic Regression, KNN Regressor, MLP Regressor, K-Means) can learn one row at a time or from a batch without losing previous learning. All other models require a full dataset and retrain from scratch each time. Training data can be passed inline as a single row, as a jsonL batch, or pulled from a stored dataset file.

Datasets hold the data entries used for AI summarization jobs. Create named datasets, set their type (text or numeric), add entries one by one or paste bulk jsonL via the Upload page, and optionally add a description. Each dataset can be targeted by one or more summary configs.

Summary configs define how the AI analyzes a dataset. Each config targets a specific dataset and lets you choose the AI model, set a data size limit, and provide a custom analysis prompt. Run the summary through the API as a background job using the summarizedataset command.

The Bulk Upload page lets you paste a block of jsonL data (one JSON object per line) to import many entries into a dataset at once. Existing entries in the target dataset are preserved and new rows are appended. Each upload batch gets a label and status for tracking.

The Pipeline Editor is the main interface for building, configuring, and training your custom ML models. Each pipeline is a sequence of one or more model steps that run in order. A single-step pipeline works like a standalone model. Multi-step pipelines pass each step's prediction as input to the next step. For complex workflows with branching logic, build those in Chain Commands and call this pipeline as one step in that chain.

All models run locally using Rubix ML with no cloud AI costs after training. Models marked with * support incremental training (can learn one row at a time without rebuilding). All other models require a full dataset and retrain from scratch.