Guides and API

These docs cover the Little Harness alpha line — currently 0.2.0-alpha.0 on npm as little-harness@alpha (the next release moves to @little-stack/harness). The URL path keeps the historical v0.1.0-alpha segment so old links stay stable; it is not the package version. Little Harness is a local-first agent runtime where the developer provides a host, model, system prompt, tools, skills, and optional Persistent Dirs. It requires Node.js 22+ and AI SDK 7 (ai is a peer dependency).

UNRELEASED review preview. This documentation snapshot also covers changes that exist only in the current source and are not in any npm version yet; no version number has been assigned to them. Those sections are labeled UNRELEASED, and Alpha Migrations lists them with their migrations. The site is not an exact catalog of the npm 0.2.0-alpha.0 API.

What This Version Proves

• Auto-created sessions with isolated workspaces.
• Message file staging for raw AI SDK chat input.
• Tools that can return large artifact-backed outputs instead of flooding model context.
• Skills exposed read-only and disclosed progressively.
• Opt-in harness memory for model-authored durable lessons.
• Local Host with .little-harness session folders, traces, and artifacts.
• Persistent Dirs with harnessDir plus developer-owned source logic.
• After-turn commits for Persistent Dirs.
• No parallel turns inside one session; same-session calls queue by default.
• Persistence errors surfaced to the host app and trace, not injected into normal agent context.
• A local runtime path powered by just-bash, run in a subprocess sandbox by default for any turn that can execute code or reach the network.
• Opt-in execution tiers that escalate commands just-bash cannot run to an environment you provide.
• Little Workflow workflows exposed as typed tools, with per-session fan-out budgets.
• Chat SDK and web-rich connectors, with inbound file attachments and reactions captured as run outcomes.
• Outcome capture (outcome.reported) and success rates per prompt from the CLI.

Start Here

• Getting Started — scaffold an agent workspace and chat with it from the CLI in minutes.
• Embedding in Your App — the API in depth: tools, skills, streaming, persistence, and embedding in your own app.
• Foundations — sessions, files, tools, persistence, and traces.
• Build Agents — primitives for defining harness-backed agents.
• Memory — opt-in model-authored memory backed by Persistent Dirs.
• Generating And Streaming — detailed examples for routes, jobs, callbacks, files, and structured outputs.
• Cookbook — example recipes for common agent use cases.
• Runtime Overview — Local Host, just-bash, execution environments and tiers, commits, and isolation.
• Outcome Capture — record whether runs worked and read success rates.
• User Feedback — preview: send your users' thumbs, ratings and surveys to LittleDB.
• API Reference — alpha package API shapes.