NanoClaw is a lightweight open-source personal AI assistant framework built directly on the Anthropic Claude Agent SDK, with 26,700+ GitHub stars. Conceived as a secure alternative to OpenClaw, it runs Claude agents inside isolated Linux containers (Apple Container on macOS, Docker elsewhere) so that no agent can access files, credentials, or system resources beyond what is explicitly mounted. A single Node.js host process manages multiple container instances via a polling loop, routing inbound messages from WhatsApp, Telegram, Slack, Discord, and Gmail to the appropriate container and returning responses. Each group maintains its own CLAUDE.md-based context — making per-group memory independent and persistent across sessions through SQLite. Credentials are routed through OneCLI's Agent Vault rather than being exposed to the agents themselves. The project's extensibility model is deliberately unconventional: instead of merging feature PRs, new capabilities are contributed as Claude Code skills — prompt-based transformations applied to each user's own fork. Built-in skills include /setup, /customize, /debug, and channel connectors such as /add-telegram, /add-discord, and /add-gmail. With 685 commits and an active Discord community, NanoClaw targets developers who want a minimalist, fully auditable AI assistant backbone they can understand end-to-end rather than a black-box platform.