@agent-layers/core

Ports, adapter registry, composition pipelines and stack config for agent-layers. Zero runtime dependencies.

The core defines one port per layer of an agent. Adapters map a standard or a technology onto a port; a stack config (agent-layers.json) picks one adapter per slot, per profile; compose turns the config into a running stack.

npm install @agent-layers/core @agent-layers/adapters

import { compose, loadConfig } from '@agent-layers/core';

const stack = await compose(await loadConfig('agent-layers.json', { profile: 'prototype' }));
try {
  const run = await (await stack.agent('assistant')).run('Hello');
  console.log(run.status, run.output);
} finally {
  await stack.close();
}

The JSON Schema of the stack config ships with the package: @agent-layers/core/schema.json, or ./node_modules/@agent-layers/core/schema/agent-layers.schema.json as $schema in the config file.

Start a project with @agent-layers/cli (npx @agent-layers/cli init my-agent). Concepts, ports and the adapter catalog are in the documentation.

Requires Node >= 22.18. License: MIT.