Important: you bring the model and the security
None of these includes a model. You connect your own API keys or local models. Because these agents can read email, use browsers, and run commands, treat them as powerful software: give them limited credentials, isolate them in containers when possible, and review their actions.
Closest to a personal AI computer
OpenClaw is the closest to an always-on assistant reached through WhatsApp, Telegram, Slack, and other chat apps, handling email and calendar tasks. Hermes Agent adds persistent memory, scheduling, and sandboxed execution across macOS, Windows, and Linux, and ZeroClaw packs 70+ model providers and 30+ messaging channels into a single Rust binary.
Lightweight and isolated options
nanobot is deliberately minimal, running in a web UI, terminal, or chat app with token budgets for long tasks. NanoClaw is designed to be forked and runs each agent group in its own Docker container, with channels installed on demand.
Important: two projects are tied to external platforms
Agent Zero runs in a sandboxed Docker environment, but its governance layer is tied to an A0T token on the BASE blockchain, which may matter to some users. IronClaw runs agents inside Trusted Execution Environments on NEAR AI Cloud, keeping credentials encrypted and hidden from the model, so deployment is tied to that cloud.
License details
OpenClaw, Hermes Agent, nanobot, NanoClaw, and Agent Zero are MIT. ZeroClaw is dual-licensed MIT or Apache-2.0, and IronClaw is Apache-2.0.
Maintenance status
All seven projects had commits within days of this update, and most are young and fast-moving, so expect frequent changes.