#Agent SDK overview

@lite-agent/sdk is the batteries-included way to build agents with lite-agent: a working tool set, skills, subagents, persistent sessions, and a permission gate behind a small @anthropic-ai/claude-agent-sdk-shaped API (query / createLiteAgent / tool). Pair it with a @lite-agent/provider for the model — every battery is a toggleable default over the core strategies, so you start productive and opt into control only where you need it.

#SDK vs Core

lite-agent is split into two layers:

• @lite-agent/sdk — the assembled agent. It wires the kernel's strategies into sensible defaults: built-in tools scoped to your workspace, an event-sourced session store, a permission middleware, skills and subagent loading. Use it when you want a working agent with a few lines of config.
• @lite-agent/core — the provider-agnostic kernel underneath: the turn loop, the nine strategy interfaces, the middleware onion, and the AgentEvent stream. It knows nothing about permissions, sandboxing, or skills — those are all plugged in. Use it directly when the SDK's assembly doesn't fit and you want to compose your own agent from kernel primitives.

The SDK re-exports all of @lite-agent/core, so you can drop down a level — write a custom strategy or middleware — without changing packages:

import { createLiteAgent, type Middleware } from "@lite-agent/sdk";
import { anthropic } from "@lite-agent/provider";

const logging: Middleware = {
  name: "logging",
  async wrapToolCall(ctx, next) {
    console.log(`→ ${ctx.call.name}`);
    const result = await next();
    console.log(`← ${ctx.call.name}${result.isError ? " (error)" : ""}`);
    return result;
  },
};

const agent = createLiteAgent({
  model: anthropic(),
  modelName: "claude-sonnet-4-6",
  workdir: process.cwd(),
  use: [logging], // extra middleware around the kernel loop
});

Rule of thumb: start with the SDK; reach for Core when you need to swap a strategy the SDK doesn't expose, reorder the middleware onion, or build a non-agent loop on the same primitives.

#Capability map

This section documents the SDK capability by capability:

#See also

• Getting started — install and run your first agent.
• Core strategies — the nine swappable parts the SDK assembles.
• Model providers — Anthropic, OpenAI, and OpenAI-compatible endpoints.
• CLI example — a full interactive agent built on these APIs.