Guides · field notes from production
Guides
Practical write-ups on running private AI agents: what breaks, what to check, and what I do about it on the machines I'm responsible for.
AI chief of staff: what it does and who needs one
An operator's guide to the category: what an AI chief of staff does, the four properties that separate it from an assistant, and when you don't need one.
Read →OpenClaw vs Hermes Agent: when to choose which, and why I run both
An operator's comparison of OpenClaw and Hermes Agent: what each one is for, when to pick which, and what running both on one server looks like in practice, including one agent repairing the other.
Read →Why Do Multi-Agent LLM Systems Fail?
An operator's read of MAST: the 14 failure modes behind multi-agent LLM failures, why more agents isn't better, and the design controls that reduce them.
Read →How to update OpenClaw (and when to leave it alone)
An operator's guide to updating OpenClaw safely: the openclaw update command, release channels, pinning and rollback, what actually breaks on update, and when not to update at all.
Read →OpenClaw Telegram bot not working: connect it right and fix it when it breaks
An operator's guide to running OpenClaw on Telegram: the real setup (there is no channels-login step), recovery after an update, fixing group silence and 401 tokens, and diagnosing polling that flaps.
Read →OpenClaw gateway "connect failed: pairing required": what it means and how to fix it
What the OpenClaw gateway "pairing required" (1008) error means and how to fix each cause: device approval, scope and role upgrades, the token-mismatch lookalike, and remote-gateway pairing.
Read →OpenClaw Security: a self-hoster's checklist
A practitioner's checklist for securing a self-hosted OpenClaw: gateway exposure, auth, CVE-2026-25253, version pinning, and third-party skills.
Read →How to run OpenClaw on a VPS: system requirements, hosting, and locking the box down
An operator's guide to running OpenClaw on a VPS: system requirements and RAM for one agent or several, where to host it (paid, free Oracle ARM, or a box at home), the install, and locking the box down before you expose it.
Read →OpenClaw dashboard not connecting: how to open the Control UI, fix a 401, and recover from a gateway error
An operator's guide to the OpenClaw dashboard and gateway when they won't connect: opening the Control UI on loopback, reaching it remotely over an SSH tunnel, fixing a 401 invalid bearer token, restarting the gateway, and recovering from a gateway error.
Read →Private AI Assistant: How to Tell If a Personal AI Assistant Is Actually Private
An operator's honest guide to what makes a personal AI assistant actually private: the two questions that decide it (whose cloud holds your context, whose keys it thinks on), why 100% private can't be verified, and the trade-offs of cloud, local, and dedicated-server setups.
Read →