catalyst-quickml

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Prerequisites

前提条件

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.

使用QuickML前，请在控制台中按项目完成一次激活：

控制台 → 你的项目 → QuickML（左侧边栏）→ 点击 "Start Exploring"

跳过此步骤将无法访问Data connectors、Pipelines、Models、Endpoints、LLM Serving、RAG及Knowledge Base。此激活操作每个项目仅需执行一次。

How It Works

使用规则

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.
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.
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.
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.
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.

请先检查数据中心（DC）可用性。 ML模型：可在US、IN、EU、AU、JP、SA、CA区域构建与发布。生成式AI（LLM Serving、RAG、Knowledge Base）：仅支持US、IN、EU、JP区域——不支持AU、SA、CA。QuickML控制台内的Zia Model Library：仅支持IN区域。如果该功能是区域限定的，且不清楚用户所在的DC，请先询问再确认功能是否可用。
对任务进行分类，仅加载匹配的参考文档（如果查询涉及多个领域，可加载多个）：
- 构建、训练、评估或调用已训练的ML模型（分类、回归、时序预测、聚类、异常检测、文本分析、推荐、AutoML）→ 加载
```
references/prediction-basics.md
```
  。
- LLM聊天、视觉（VLM）、RAG或Knowledge Base → 加载
```
references/generative-ai-basics.md
```
  。
- 方案、费率、免费额度、试用或计费 → 加载
```
references/quickml-pricing-basics.md
```
  。
- 无需数据集或训练的预构建模型（OCR、人脸分析、图像审核、物体识别、条形码、预构建情感分析/NER/关键词提取）→ 停止并加载 catalyst-zia
  技能。独立的Zia Services API有自己的DC规则，与QuickML控制台中的Zia Model Library相互独立。
优先使用工具，其次是SDK，最后是控制台。 如果已连接的Catalyst MCP工具符合操作需求（pipelines、endpoints、datasets、metrics、LLM/RAG配置），请在运行时通过名称/描述检索到该工具，并在调用前读取其schema——不要硬编码工具名称。否则请使用SDK：JavaScript SDK（
```
new QuickML(app)
```
）为当前可用版本；Node.js SDK v2（
```
app.quickML()
```
）已弃用。在所有已发布的SDK中，预测调用方法均为 predict(endpointKey, inputData)
——文档页面上的
```
runInference
```
/
```
run_inference
```
名称并不存在。Agent无法查看控制台，因此对于仅在UI中提供的值，请向用户询问。
绝不编造QuickML的具体信息。 仅使用参考文档中（已实时验证）的REST路径、请求头、权限范围（
```
QuickML.deployment.READ
```
）和响应结构。不要虚构响应字段——不存在
```
likelihood_score
```
或
```
confidence
```
字段——其他信息请从控制台 → QuickML → Endpoints中的端点示例获取。不要编造定价参考文档中未列出的价格数值。
控制台操作步骤必须详尽。 每一步都要说明确切的字段、需要选择的下拉选项/条件、对应值以及所有复选框——笼统的步骤会导致隐性配置错误（例如，填充列规则需要将其condition设置为"missing values"，而不仅仅是设置填充值）。对于参考文档未涵盖的内容，请以Markdown格式抓取https://docs.catalyst.zoho.com/en/quickml/下的页面（
```
<page-url>/index.md
```
）；如果抓取失败，请告知用户并提供该URL。

Triggers

触发场景

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".

以下场景可使用此技能："QuickML"、"Catalyst QuickML"、"QuickML端点"、"使用Catalyst模型训练/预测"、"无代码ML"、"无代码机器学习"、"部署模型"、"模型端点"、"MLOps"、"AutoML"、"ML pipeline"、"LLM Serving"、"RAG"、"预构建模型 / Zia模型"、"OCR / 人脸 / 图像 / 文本模型"、"QuickML定价"、"QuickML成本"、"多少钱"（针对QuickML使用）、"计费"、"订阅方案"、"按需付费"或"Catalyst方案"。

References

参考文档

Reference	Load when the query is about…
`references/prediction-basics.md`	Trained ML models — pipeline types and when to pick each, data connectors, preprocessing and operations, building and training pipelines, AutoML, evaluation metrics, SHAP, creating/publishing endpoints, prediction SDK calls (Java, Python, JavaScript, Node.js v2), limits
`references/generative-ai-basics.md`	Generative AI — LLM Serving (models, parameters, interaction modes, tool calling), VLM, RAG (modes, retrieval scope, agentic RAG), Knowledge Base uploads (local, WorkDrive, Zoho Learn), GenAI endpoint REST calls (no GenAI SDK methods exist)
`references/quickml-pricing-basics.md`	QuickML pricing — Catalyst subscription plans, pay-as-you-go unit rates, monthly free tier, add-ons, free trial, billing controls, and the rules for answering pricing questions

参考文档	当查询涉及以下内容时加载
`references/prediction-basics.md`	已训练ML模型——pipeline类型及选型场景、数据连接器、预处理与操作、构建与训练pipeline、AutoML、评估指标、SHAP、创建/发布端点、预测SDK调用（Java、Python、JavaScript、Node.js v2）、使用限制
`references/generative-ai-basics.md`	生成式AI——LLM Serving（模型、参数、交互模式、工具调用）、VLM、RAG（模式、检索范围、Agentic RAG）、Knowledge Base上传（本地、WorkDrive、Zoho Learn）、GenAI端点REST调用（无GenAI SDK方法）
`references/quickml-pricing-basics.md`	QuickML定价——Catalyst订阅方案、按需付费单价、月度免费额度、附加服务、免费试用、计费控制，以及定价问题的回答规则