catalystbyzoho/agent-skills/catalyst-quickml

Prerequisites

Before using QuickML, activate it once per project in the console:

Console → your project → QuickML (left sidebar) → click "Start Exploring"

Skipping this step prevents access to Data connectors, Pipelines, Models, Endpoints, LLM Serving, RAG, and Knowledge Base. This is a one-time activation per project.

How It Works

1. Check data-center availability first. ML models: build & publish in US, IN, EU, AU, JP, SA, CA. Generative AI (LLM Serving, RAG, Knowledge Base): US, IN, EU, JP only — not AU, SA, or CA. Zia Model Library inside the QuickML console: IN only. If the feature is regional and the user's DC is unknown, ask before confirming it is available.

1. Classify the task and load only the matching reference (load several if the query spans domains):

• Build, train, evaluate, or call a trained ML model (classification, regression, forecasting, clustering, anomaly detection, text analytics, recommendation, AutoML) → load references/prediction-basics.md.
• LLM chat, vision (VLM), RAG, or Knowledge Base → load references/generative-ai-basics.md.
• Plans, rates, free tier, trial, or billing → load references/quickml-pricing-basics.md.
• A pre-built model with no dataset or training (OCR, face analytics, image moderation, object recognition, barcode, pre-built sentiment/NER/keywords) → STOP and load the `catalyst-zia` skill. The standalone Zia Services APIs have their own DC rules, separate from the QuickML-console Zia Model Library.

1. Tool-first, then SDK, then console. If a connected Catalyst MCP tool fits the action (pipelines, endpoints, datasets, metrics, LLM/RAG configs), discover it at runtime by name/description and read its schema before calling — do not hardcode tool names. Otherwise use the SDK: the JavaScript SDK (new QuickML(app)) is current; Node.js SDK v2 (app.quickML()) is deprecated. In every published SDK the prediction call is `predict(endpointKey, inputData)` — the runInference / run_inference names on the docs pages do not exist. The agent cannot see the console, so ask the user for UI-only values.

1. Never fabricate QuickML specifics. Use only the REST paths, headers, scope (QuickML.deployment.READ), and response shapes in the references (verified live). Do not invent response fields — there is no likelihood_score or confidence — and take anything else from the endpoint's sample in Console → QuickML → Endpoints. Do not invent pricing values the pricing reference does not list.

1. Console steps must be exhaustive. For every step, name the exact field, the dropdown/condition to select, the value, and any checkbox — generic steps cause silent misconfiguration (e.g. a Fill Columns rule needs its condition set to "missing values", not only a fill value). For anything the references do not cover, fetch the page under <https://docs.catalyst.zoho.com/en/quickml/> as Markdown (<page-url>/index.md); if the fetch fails, say so and give the user the URL.

Triggers

Use this skill for: "QuickML", "Catalyst QuickML", "a QuickML endpoint", "training/predicting with a Catalyst model", "no-code ML", "no-code machine learning", "deploy a model", "model endpoint", "MLOps", "AutoML", "ML pipeline", "LLM Serving", "RAG", "pre-built model / Zia model", "OCR / face / image / text model", "QuickML pricing", "QuickML cost", "how much" (for QuickML usage), "billing", "subscription plan", "pay-as-you-go", or "Catalyst plan".

References