NanoClaw joins Open Orchestrators as a lightweight personal Claude assistant for people who like the OpenClaw-shaped workflow but want a smaller, more inspectable codebase and stronger execution isolation.
The project describes itself as an AI assistant that runs agents securely in their own containers. It connects to Telegram, Discord, WhatsApp, Slack, Gmail, and a local CLI; keeps persistent memory per conversation; supports scheduled tasks; and extends through a skills-as-git-branches model. The quick start is deliberately simple: clone the repo, run bash nanoclaw.sh, and walk through setup for an agent and first channel.
This matters for Open Orchestrators because NanoClaw is not just a prompt wrapper. It is an assistant runtime with messaging channels, memory, schedules, credentials, skills, and a container execution model around agent work. That puts it in the same personal-operator branch of the category as OpenClaw, Hermes Agent, and SwarmClaw, while making a different trade-off: fewer moving parts and a tighter security boundary.
The security model is the clearest differentiator. NanoClaw’s docs frame container isolation as the primary boundary: agents execute in Linux containers with process isolation, explicit filesystem mounts, non-root execution, and ephemeral container lifetimes. The project explicitly contrasts this with relying only on application-level allowlists and permission checks.
For builders, the practical signal is that the “personal assistant with hands” category is splitting into distinct operating models. OpenClaw emphasizes a broad assistant surface. Hermes emphasizes persistent skills, memory, subagents, schedules, and messaging gateways. NanoClaw emphasizes a small codebase and OS-level isolation around the agent’s execution environment.
Source confidence: High for the public GitHub README, package metadata, docs, MIT license, and repo activity. Medium for downstream adoption claims because this directory entry does not benchmark reliability, security hardening, or channel coverage in production.
Evidence:
- NanoClaw GitHub repository
- NanoClaw website
- NanoClaw docs
- NanoClaw security model
- NanoClaw skills system
Explicit non-claims:
- This entry does not claim NanoClaw is more secure than OpenClaw in audited production deployments.
- This entry does not claim every listed channel has identical maturity or reliability.
- This entry does not claim container isolation removes the need for credential, mount, network, and approval policy design.