UNPKG

synthex

Version:

Type-safe LLM response simulation with streaming & error injection

225 lines (174 loc) 7.81 kB
<br /> <div align="center"> <a href="https://github.com/nick-vanduijn/synthex"> <img src="images/logo.png" alt="Logo" width="80" height="80"> </a> <h3 align="center">Synthex</h3> <p align="center"> Type-safe Mocking for LLMs & Structured APIs </p> ![npm version](https://img.shields.io/npm/v/synthex?color=brightgreen&label=npm) ![types](https://img.shields.io/npm/types/synthex?label=types) ![downloads](https://img.shields.io/npm/dt/synthex?color=blue) ![license](https://img.shields.io/npm/l/synthex?color=orange) ![build](https://img.shields.io/github/actions/workflow/status/nick-vanduijn/synthex/ci.yml?branch=main&label=build) </div> ## Table of Contents - [Install](#install) - [Features](#features) - [Example Usage](#-example-usage) - [Sample Output](#-sample-output) - [API Overview](#api-overview) - [Streaming / LLM Simulation](#streaming--llm-simulation) - [Why Synthex?](#why-synthex) - [Documentation](#documentation) - [More Examples](#more-examples) - [CLI Demo](#cli-demo) - [FAQ](#faq) - [Contributing](#contributing) - [License](#license) ## Documentation For a comprehensive usage guide, advanced schema patterns, and API reference, see: - [docs/USAGE.md](docs/USAGE.md) --- ## More Examples - `node examples/form-demo.cjs` — Interactive form mocker - `node examples/llm-demo.cjs` — LLM-style streaming and error demo - `node examples/advanced-demo.cjs` — Advanced: nested, error, streaming, and conditional fields --- ## The Problem When building AI or LLM-powered applications, testing and prototyping can be painfully slow. Every time you want to check a new feature, you have to wait for your LLM to respond—sometimes for seconds or even minutes. This slows down your feedback loop, makes debugging tedious, and can rack up unnecessary API costs. **Synthex** solves this by letting you instantly simulate type-safe, realistic LLM or API responses. You can test your code, validate edge cases, and iterate rapidly—without waiting for a real LLM or burning tokens. Save time, move faster, and focus on building, not waiting. ## Install ```bash npm install synthex ``` ## Features - **Type-safe schema builder**: primitives, objects, enums, unions, intersections - **Realistic mock data** for testing, prototyping, or LLM scaffolding - **Conditional fields** with probabilities - **Simulated errors** for edge-case testing - **Streaming mock generation** — mimic LLM token flow - **Context-aware** field templates (e.g. IDs, tokens, slugs) - **Test-friendly metadata**: timestamps, token usage, finish reasons - **Composable** schema API, like Zod but mock-first - **Markdown & JSON formatters** for quick debugging - **Plugin system**: extend or override field generation logic - **Schema import/export**: JSON/YAML, CLI utilities - **LLM simulation**: hallucination (for all types, including enums), function-calling, streaming, error injection - **Performance profiling utility**: measure mock generation speed - **Lite entry point**: minimal bundle for browser or edge - **CI/CD ready**: robust tests, linting, and GitHub Actions ## Example Usage ```ts import { s, MockGenerator } from 'synthex'; const userSchema = s .object({ id: s.uuid().required(), name: s.string().min(2).max(50).required(), email: s.email().required(), age: s.number().min(18).max(99), isActive: s.boolean(), role: s.enum(['user', 'admin', 'moderator']), profile: s .object({ bio: s.string().max(160), website: s.url(), }) .optional(), }) .build('UserSchema'); const generator = new MockGenerator({ seed: 42 }); const mock = generator.generate(userSchema); console.log(mock.data); ``` ## Sample Output ```json { "id": "8c5d3a91-14b4-4c5b-a301-cb837b66f0a1", "name": "Ava Jackson", "email": "ava.jackson@example.com", "age": 35, "isActive": true, "role": "admin", "profile": { "bio": "Developer. Writer. Tinkerer.", "website": "https://ava.dev" } } ``` ## API Overview | Type | Builder | |--------------------|------------------------------------| | String | `s.string()` | | Number | `s.number()` | | Boolean | `s.boolean()` | | Date / Time | `s.date().format()` | | Email, URL, UUID | `s.email()`, `s.url()`, `s.uuid()` | | Arrays | `s.array(s.string())` | | Objects | `s.object({...})` | | Enums | `s.enum(['a', 'b'])` | | Conditional Fields | `.probability(0.5)` | | Error Simulation | `.simulateError(true)` | | Streaming Output | `generator.streamGenerate()` | ## Streaming / LLM Simulation & Hallucination ```ts const generator = new MockGenerator({ hallucinate: true, // Enable hallucination for all fields hallucinationProbability: 0.5, // 50% chance per field seed: 123, }); const mock = generator.generate(userSchema); console.log(mock.data); // Streaming (mimic LLM token flow) const stream = generator.streamGenerate(userSchema, { chunkSize: 10, delayMs: 50 }); for await (const chunk of stream) { process.stdout.write(chunk); } ``` **Advanced:** - Hallucination works for all field types (string, number, boolean, array, object, enum, etc.) - Simulate OpenAI-style function-calling, error injection, and role-based responses ## Why Synthex? | Feature | Synthex | Zod | Faker | json-schema-faker | |------------------------------|:--------:|:---:|:-----:|:-----------------:| | Type-safe schema builder | ✅ | ✅ | ❌ | ❌ | | Realistic mock data | ✅ | ❌ | ✅ | ✅ | | Field-level error simulation | ✅ | ❌ | ❌ | ❌ | | Streaming LLM-like output | ✅ | ❌ | ❌ | ❌ | | Composable API | ✅ | ✅ | ❌ | ⚠️ Partial | ## CLI & Utilities - **Schema import/export:** ```bash node bin/schema-io.js import ./schema.yaml node bin/schema-io.js export ./schema.json ``` - **Performance profiling:** ```ts import { profileMockGeneration } from 'synthex/perf'; profileMockGeneration(userSchema, 1000); ``` - **Lite entry point:** ```ts import { s, MockGenerator } from 'synthex/lite'; ``` - **Examples:** - `node examples/form-demo.cjs` — Interactive form mocker - `node examples/llm-demo.cjs` — LLM-style streaming and error demo - `node examples/advanced-demo.cjs` — Advanced: nested, error, streaming, and conditional fields ## FAQ **Q: What makes this better than Faker?** A: Synthex combines type-safety, streaming support, error injection, hallucination, and realistic test metadata. Ideal for LLMs and structured APIs. **Q: Can I use this to test OpenAI function-calling?** A: Yes! It’s designed to simulate real-world outputs for tools/functions, including function-calling mocks and streaming. **Q: Does it support nesting?** A: Yup. Nest objects, arrays, even deeply nested enums and conditional fields. **Q: How do I handle type-only tests and linting?** A: Type-only tests (e.g., `test/types.test.ts`) may trigger `@typescript-eslint/no-unused-vars` for schema variables used only for type inference. Suppress these with `// eslint-disable-next-line @typescript-eslint/no-unused-vars` or use a dummy test to satisfy Jest. **Q: How do I extend Synthex?** A: Use the plugin system to override or extend field generation logic. See [docs/RECIPES.md](docs/RECIPES.md) for advanced patterns. ## Contributing We welcome PRs, ideas, and issue reports. Clone the repo and run: ```bash npm run dev ```