Harness + Workflow0.2.0-alpha.2

THE LITTLE STACK DOCUMENTATION

Good agents start
with a little stack.

Build with an agent runtime. Add durable workflows.
Keep the tools, files, and traces that make it all work.

Get started Explore the docs

Local first. Bring your own model.

TypeScript
import { createHarness, localHost }
  from "@little-stack/harness";
import { openai } from "@ai-sdk/openai";

export const agent = createHarness({
  host: localHost({ dataDir: ".little" }),
  model: openai("gpt-5"),
  system: "You help people get things done.",
});
Build your first agent
Your model. Your tools. Your workspace.

ONE STACK. CLEAR RESPONSIBILITIES.

Find your building block.

How they fit together

THE AGENT RUNTIME

Little Harness

Give your agent tools, skills, files, and a workspace. Keep sessions and traces together.

Published alpha@little-stack/harness

DURABLE EXECUTION

Little Workflow

Turn bounded jobs into validated workflows. Run, inspect, and replay every step.

Published alpha@little-stack/workflow

TRACING & IMPROVEMENT

LittleDB

Trace what agents do, evaluate the results, and test changes that make them better.

Private sourceAccess required

THE WORKSPACE ENGINE

Little Compute

The self-hosted engine for persistent workspaces and on-demand execution.

Unpublished sourceControl plane frozen

Harness and Workflow run independently. LittleDB is private; Little Compute is unpublished. See release status

FROM AN IDEA TO A FIRST RUN

Start small.
Build from there.

Most projects start with Harness. Add a workflow when a job needs durable, replayable steps.

Choose your starting point
01 / HARNESS

Build your first agent

Bring a model and a few tools. Embed a Harness agent in your own app.

Follow the quickstart
02 / WORKFLOW

Run a durable workflow

Define a job, run it locally, and inspect the steps that got you there.

Follow the quickstart

Your machine is a good place to start.

The published SDKs run locally with Node.js 22+ and AI SDK 7. No Little hosted service required.

Requirements