@atisse/core

A lightweight, production-grade execution kernel for managing the LLM interaction lifecycle.

---

The Problem

Most AI integrations fall into one of two anti-patterns. Raw SDK calls have no retry, no fallback, no memory, and no observability — they break in production. Heavy frameworks like LangChain introduce opaque abstractions, hidden behavior, and vendor lock-in that make debugging and testing painful.

The gap between these two extremes is a production-ready, minimal orchestration layer that manages the lifecycle of an LLM call without becoming a framework. That is what @atisse/core provides.

---

How It Compares

	LangChain	Vercel AI SDK	@atisse/core
Weight	Heavy	Light	Minimal
Focus	Framework	Frontend	Backend kernel
Hidden behavior	High	Medium	None
Vendor lock-in	Yes	Partial	No
Production retry/fallback	Complex	None	First-class
Tool lifecycle	Opaque	Limited	Explicit
Streaming	Yes	First-class	First-class
Testability	Hard	Medium	MockProvider

---

Quick Start

pnpm add @atisse/core @atisse/provider-openai

npm install @atisse/core @atisse/provider-openai

import { Orchestrator } from '@atisse/core';
import { OpenAIProvider } from '@atisse/provider-openai';

const apiKey = process.env.OPENAI_KEY ?? '';
if (!apiKey) throw new Error('OPENAI_KEY environment variable is required');

const orchestrator = new Orchestrator({
  provider: new OpenAIProvider({ apiKey }),
  retry: { maxAttempts: 3 },
});

const result = await orchestrator.run({ prompt: 'Hello' });
console.log(result.text);

Behind the scenes: context was loaded, the prompt was composed, the provider was called with retry policy enforced, and the result was returned with usage and timing data — all in a deterministic, observable pipeline.

---

Packages

Each package has its own README.md with detailed API documentation, installation guides, and code examples.

Package	Description
@atisse/core	Kernel, frozen interfaces, MockProvider
@atisse/provider-openai	OpenAI adapter (GPT-4o, GPT-4o-mini)
@atisse/provider-anthropic	Anthropic adapter (Claude 3.5 Sonnet, Haiku)
@atisse/memory-inmemory	Reference in-memory MemoryAdapter
@atisse/memory-redis	Redis-backed MemoryAdapter with TTL support
@atisse/context-rag	RAG ContextProvider — pluggable vector store

---

Links

• Getting Started
• Writing Adapters
• API Reference — Run pnpm run docs to generate and view API documentation (will be listed under docs/api/)
• Examples
• GitHub Discussions