Examples Overview

This section contains complete, real-world examples showing Nuabase in action. Each example includes:

• A problem description
• Complete, copy-paste ready code
• Expected output
• Variations

Available Examples

Lead Enrichment

Classify and prioritize sales leads based on notes and context. A classic B2B use case demonstrating nua.list() with complex schemas.

View Example

Sentiment Analysis

Analyze customer feedback at scale. Shows batch processing with nua.list() and grouping results.

View Example

Data Extraction

Extract structured data from unstructured text. Demonstrates schema design and validation patterns.

View Example

Running the Examples

Install Nuabase and Zod:

npm install nuabase zod

The examples use whichever coding agent is installed and logged in on your machine:

import { Nua, z } from 'nuabase';
import { findLocalAgent } from 'nuabase/local-agent';

const nua = Nua.direct({ localAgent: await findLocalAgent() });

To run them against an LLM API instead, swap the client:

import { Nua, z } from 'nuabase';

const nua = Nua.direct({
  providers: { openrouter: { apiKey: process.env.OPENROUTER_API_KEY! } },
});

The rest of each example is the same with either client.

Schema Design Tips

When designing schemas for your use cases:

Keep Schemas Simple

// Good: focused schema
const SentimentSchema = z.object({
  sentiment: z.enum(['positive', 'neutral', 'negative']),
  confidence: z.number().min(0).max(1),
});

// Avoid: overly complex schemas
const OverlyComplexSchema = z.object({
  sentiment: z.object({
    primary: z.enum(['positive', 'neutral', 'negative']),
    secondary: z.array(z.string()),
    scores: z.object({
      positive: z.number(),
      neutral: z.number(),
      negative: z.number(),
    }),
  }),
  // ... many more nested fields
});

Use Enums for Classification

// Good: constrained options
const CategorySchema = z.enum(['bug', 'feature', 'question']);

// Less reliable: free-form string
const CategorySchema = z.string();

Make Optional Fields Explicit

const ContactSchema = z.object({
  name: z.string(),
  email: z.string(),
  phone: z.string().optional(),  // Explicitly optional
  company: z.string().optional(),
});

Add Validation Constraints

Every answer is parsed with your schema, and an answer that breaks a constraint is retried:

const MetricsSchema = z.object({
  score: z.number().min(0).max(100),
  count: z.number().int().nonnegative(),
  percentage: z.number().min(0).max(1),
  zip: z.string().regex(/^\d{5}$/),
});

.refine() checks, string formats such as z.email(), .default()s, and .transform()s all apply to result.data. See How It Works.

Keep the Prompt Constant

Put everything that varies in input, not in the prompt string. Providers cache the unchanging start of each request, which makes repeated calls cheaper and faster. See prompt caching.

Prompt Writing Tips

Be Specific

// Good: clear instructions
await nua.get("Classify the customer support ticket into one of the following categories: bug report, feature request, billing question, or general inquiry", options);

// Vague: leaves too much interpretation
await nua.get("Classify this", options);

Provide Context

await nua.get("You are analyzing sales leads for a B2B SaaS company. Classify each lead by their likely company size and purchase intent.", {
  input: leadNotes,
  output: { name: 'classification', schema: LeadSchema }
});

Match Prompt to Schema

const OutputSchema = z.object({
  category: z.enum(['high', 'medium', 'low']),
  reason: z.string(),
});

// Good: prompt matches schema fields
await nua.get("Rate the priority (high, medium, or low) and explain the reason", {
  output: { name: 'rating', schema: OutputSchema }
});