llm-core

Appearance

llm-coreBuild AI with Recipes, not Glue.

Define declarative flows using Recipes. Swap providers via Adapters. Stop writing spaghetti code.

Get Started
Core Concepts
llm-core logo

Recipes are Assets

Define flows as named, versioned recipes. Share them across teams like npm packages.

Packs are Logic

Bundle specific capabilities (like "Planning" or "Memory") into reusable internals.

Adapters are Plugs

Swap OpenAI for Anthropic or LangChain for LlamaIndex without rewriting your recipe.

Execution is Deterministic

The engine runs your recipe as a DAG. Every step is traced, resumable, and explainable.

Stop debugging prompts. Start orchestrating logic. ​

llm-core connects your business logic to AI models without gluing them together with fragile scripts. You define the Recipe, plug in the Adapters, and let the Runtime handle the execution.

Quick start (TS/JS) ​

Define a flow, plug in your adapters, and run it.

ts
import { recipes } from "@geekist/llm-core/recipes";
import { fromAiSdkModel } from "@geekist/llm-core/adapters";
import type { Model } from "@geekist/llm-core/adapters";
import { openai } from "@ai-sdk/openai";

// 1. Define your recipe (or load a standard one)
const agent = recipes.agent();

// 2. Plug in your adapters
const model: Model = fromAiSdkModel(openai("gpt-4o"));
const workflow = agent.defaults({ adapters: { model } }).build();

// 3. Run it
const result = await workflow.run({ input: "Build me a React app" });

if (result.status === "ok") {
  console.log(result.artefact);
}

Next ​