Echostash SDK

Universal prompt SDK with messages, tools, and 5 provider converters

Echostash SDK

The Echostash SDK (@goreal-ai/echostash) provides a fluent API for fetching prompts and converting them to any LLM provider format. Available for JavaScript and Python.

Installation

npm install @goreal-ai/echostash

Quick Start

import { Echostash } from "@goreal-ai/echostash"

const es = new Echostash("https://api.echostash.app", {
  apiKey: process.env.ECHOSTASH_API_KEY
})

// One-liner: fetch → set variables → convert to OpenAI format
const result = await es
  .prompt("welcome-email")
  .vars({ name: "Alice", tier: "pro" })
  .openai()

// result.messages = [{ role: "system", content: "..." }, { role: "user", content: "..." }]
// result.tools = [{ type: "function", function: { name: "...", ... } }]
// result.model = "gpt-4o"
// result.temperature = 0.7

Fluent API Chain

The SDK uses a chainable pattern: prompt() → version() → vars() → provider()

.prompt(id) — Fetch a prompt by slug or numeric ID
.version('published') — Pin to published, staging, or specific version number
.vars({...}) — Set template variables for rendering

Provider Converters (Terminal Methods)

Each method has two overloads: with no arguments it returns the full prompt result (messages, tools, model config) ready to spread into the provider API. With options, it returns a single message in that provider's format.

Method	Provider	No-arg Result
`.openai()`	OpenAI	`{ messages, tools?, model?, temperature?, max_tokens? }`
`.anthropic()`	Anthropic	`{ system?, messages, tools?, model?, max_tokens? }`
`.google()`	Google Gemini	`{ contents, tools?, generationConfig? }`
`.vercel()`	Vercel AI SDK	`{ messages, tools? }`
`.langchain()`	LangChain	`{ messages, tools? }`

Server-Side Render

// Render on the server — returns resolved content
const rendered = await es.prompt("greeting").render({ name: "Alice" })
// rendered.content = "Hello Alice! Welcome to our platform."

Batch Render (up to 50 prompts)

const batch = await es.batchRender([
  { promptId: 1, variables: { name: "Alice" } },
  { promptId: 2, variables: { name: "Bob" } },
])
// batch.results = { "1": { content: "...", versionNo: 3 }, "2": { content: "..." } }
// batch.successCount = 2

Messages & Tools

const loaded = await es.prompt("support-agent").vars({ tier: "pro" }).get()

loaded.messages  // [{ role: "system", content: [...] }, { role: "user", content: [...] }]
loaded.tools     // [{ type: "function", function: { name: "search", ... } }]
loaded.renderMeta // { provider: "openai", model: "gpt-4o", temperature: 0.7 }

Rate Limiting

The SDK automatically retries on 429 responses (up to 3 times) with exponential backoff. EchostashError includes isRateLimited and retryAfter for custom handling.