CandleKeep
Anthropic

NanoClaw: The Complete Agent Developer's Guide

nano-claw
Pages35
Formatmarkdown
ListedMay 17, 2026
UpdatedJune 14, 2026
Subscribers36

About

A definitive reference for building, customizing, and operating NanoClaw v2 agents. Covers architecture, entity model, Claude integration, multi-agent orchestration, memory, security, and the full skills ecosystem.

35Chapters
241Topics
35Pages

Preview

NanoClaw: The Complete Agent Developer's Guide

Load this book when:

  • You are building, wiring, or debugging a NanoClaw v2 agent (containers, sessions, channels, providers).
  • You are configuring an agent's model/effort, prompts (CLAUDE.md), memory, or approvals.
  • You are writing a NanoClaw skill, channel, provider, or patch and need the customization contract.
  • You are upgrading a NanoClaw fork (/update-nanoclaw, migrations, the startup tripwire).
  • You are diagnosing session corruption, stuck containers, credential/egress, or prompt-injection issues.

Last updated: 2026-06-14 · Scope of validity: NanoClaw v2 (≈ v2.0.64, May 2026); Anthropic model/pricing facts current as of 2026-06-14. Review cadence: model, pricing, and breaking-changes pages quarterly or on any Anthropic model release; architecture/skills pages on each NanoClaw minor version. Maintainer: see CONTRIBUTING.md in the NanoClaw repository.

Updated 2026-06-14 — model landscape moved above Opus 4.8. Anthropic now ships Claude Fable 5 (claude-fable-5) as its most capable widely released model, positioned above Opus 4.8. Where this book calls Opus 4.8 "the current flagship," read that as "the current flagship of the Opus tier" — Fable 5 sits above the Opus family. NanoClaw still defaults agents to Sonnet 4.6 and reserves Opus 4.8 for direct/long-horizon work; Fable 5 is an opt-in tier for the hardest long-horizon agentic runs (see Ch 23 §23.1). This supersession note follows RULE 14.2 (flag, don't silently overwrite).

Upstream: NanoClaw is open source (MIT). Latest release at time of writing: v2.0.64 (2026-05-18). Repos: github.com/qwibitai/nanoclaw / github.com/nanocoai/nanoclaw. NanoClaw is a deliberately small (~700-line core) container-isolated alternative to OpenClaw, built on Anthropic's Agent SDK. Verify the current release before relying on version-specific behavior in this book.

A definitive reference for building, customizing, and operating NanoClaw v2 agents. This book is written for developers who know TypeScript and want to build production-grade Claude-powered agents on top of NanoClaw. It covers architecture, the entity model, Claude integration, memory, multi-agent orchestration, security, and the full skills ecosystem — all grounded in the real v2 codebase.


Add to library to read more

Table of Contents

NanoClaw: The Complete Agent Developer's Guide
Chapter 0 — Decision Matrix

What sessions exist?
What's in this session's inbound queue?
What's the container state?

Add to Library

Free · Live updates included

36 readers subscribed