AI-native TypeScript framework · v4.2.11
Fourteen standalone packages — HTTP, ORM, cache, queues, auth, observability, and a first-class AI layer with typed agents, tools, and workflows. Pick what you need; ship faster.
[ Same shape, different layer ]
No bolted-on AI. No different mental model for the AI layer. An agent is just another piece of code with typed inputs and outputs.
import { router } from "@warlock.js/core";import { listProducts } from "./controllers/list-products.controller";
router.get("/products", listProducts);One line registers a route. The controller, validation, response helpers, and module conventions live in core.
import { ai } from "@warlock.js/ai";import { OpenAISDK } from "@warlock.js/ai-openai";
const openai = new OpenAISDK({ apiKey: process.env.OPENAI_API_KEY! });
const supportAgent = ai.agent({ model: openai.model({ name: "gpt-4o-mini" }), systemPrompt: "You are a concise senior engineer.",});
const { text } = await supportAgent.execute("Why use generics?");Pick a provider, define an agent, call it. Tools, retries, structured output, and middleware live in ai.
[ Why Warlock ]
The AI layer isn't a separate world bolted to the side. It's the same shape as everything else you write.
An HTTP controller and an AI agent are both just typed code with typed inputs and outputs. No second mental model for the AI layer.
Seal schemas are validation, TypeScript types, and JSON schema at once. Define a shape once; it holds at the edge, in the compiler, and in tool calls.
Every package is independently installable. Start with Core, add Cascade, cache, auth, or the AI layer only when your app actually reaches for them.
Agents, tools, workflows, and supervisors are first-class — with five provider adapters behind a single API. Swap models without touching agent code.
[ What's in the box ]
Independently installable, designed to compose. Here are six of the headline packages.
The HTTP backbone.
Routing, modules, middleware, mail, and file storage — with the conventions every Warlock app is built on.
Typed database ORM.
Models with relations and migrations, atomic operations, and built-in vector search — all fully typed.
Multi-driver caching.
Tags, locks, stale-while-revalidate, and similarity search behind one consistent API.
Identity and permissions.
Tokens, route protection, and custom user types — plus RBAC checks and ABAC policies when you need them.
One schema, three jobs.
Validation that doubles as TypeScript types and JSON schema — define a shape once and reuse it everywhere.
Agents on the same stack.
Typed agents, tools, workflows, and supervisors — with five provider adapters behind a single API.
[ AI-native ]
Typed agents, tools, and workflows that behave like everything else you write. Swap the model provider without touching your agent code.
[ Pick your path ]
Different goals lead to different first pages. The framework doesn't care which one you take.
Backend developer
Scaffold with `warlock create`, register a route, drop in a Cascade model. Auth, validation, mail, files are already wired.
Start with CoreAI engineer
Pick a provider, define a tool with Seal, call `runAgent`. Same envelope as the rest of the framework, no exceptions to plumb.
Start with AIFramework explorer
How the connector lifecycle boots subsystems, how modules thread through the request, how skills bring AI agents up to speed.
Read the spine[ Ship it ]
Scaffold a project, register a route, run your first agent — all on the same primitives. You'll be productive in an afternoon.