From OpenClaw to Hermes: giving an NZ business an AI employee that actually remembers

The brief

A local business came to us frustrated. They had paid another company to set them up with OpenClaw — an AI assistant that, on paper, did everything they wanted. In practice, it did two things very reliably: it timed out, and it forgot.

The gateway in front of the assistant was unstable. Long requests died with timeout errors, often right in the middle of a task. Worse, there was no memory built in. Every conversation started from zero. The team found themselves explaining the same context — who the client was, what the project was, what they'd asked yesterday — over and over again. It felt less like an employee and more like talking to someone with no short-term memory.

They didn't want to throw out the idea of an AI assistant. The idea was right. The implementation was letting them down.

What we moved them to

We migrated them to Hermes — our AI personal assistant platform, built to behave like a real virtual employee rather than a stateless chatbot. The move was deliberate and staged, so the team never lost access to the help they'd come to rely on.

The core of the new setup:

• Persistent memory. Hermes remembers. Context from past conversations, decisions, and recurring tasks carries forward, so the team stops repeating themselves. The longer it runs, the more useful it gets.
• A real SQL database and structured logs. Instead of an opaque black box, Hermes stores its state in a proper database, and every action is logged. That makes it stable, auditable, and debuggable.
• A stable gateway. The timeout problems that plagued the OpenClaw setup were largely a gateway and architecture issue. Hermes runs behind a properly configured, health-checked gateway, so long-running tasks complete instead of dying halfway.
• A library of prebuilt apps and skills. Hermes ships with a growing set of ready-made integrations and tools, so common jobs don't need to be built from scratch.
• Docker-based security isolation. Hermes runs inside a Docker environment, sandboxed away from the rest of the company's systems. If something goes wrong, it's contained.

Why it gets better over time

The thing the team noticed first was that OpenClaw never improved. It was as clueless in month three as it was on day one.

Hermes is the opposite. Because it has memory and a database underneath it, every week of use compounds:

• It learns the business's recurring questions and context.
• It builds up a working knowledge of the team's projects and people.
• We connect it to more tools and data sources as trust grows.

A standout example: we connected Hermes to the team's Obsidian vault through MCP (the Model Context Protocol). Now the assistant can read and reason over their own notes and knowledge base — and it keeps learning more about how the business actually works the more they write. Their documentation became something the assistant could use, not just somewhere notes went to die.

Security and operations — the honest version

We're upfront with clients about what they're getting, so here's the straight version.

• Run it behind a private network (VPN). Hermes is powerful, which means access needs to be controlled. We configure a virtual private network so the team can connect to their assistant safely from anywhere — without exposing it to the open internet.
• It's isolated by design. The Docker environment keeps Hermes sandboxed from the rest of the company's infrastructure. Blast radius stays small.
• Everything is logged. Because actions go through a database and structured logs, unusual activity is visible rather than hidden.
• It's still beta software, and it needs daily updates. Hermes is evolving fast. That's a strength — it improves constantly — but it means it needs updating regularly, effectively every day, to stay current and secure. We handle that maintenance as part of the service so the client doesn't have to think about it.

We'd rather a client know it's beta and improving daily than oversell it as finished. The team understood the trade-off and were happy to make it, because even in beta it was already far ahead of the stable-but-useless setup they came from.

The result

The repetition stopped. The timeouts stopped. The team now has an assistant that remembers what they told it last week, pulls from their own knowledge base, runs safely in isolation, and gets a little more capable every week.

The difference wasn't a smarter model. It was giving the AI a memory, a database, a safe place to run, and someone to keep improving it.

What started as a rescue job — getting them off a broken OpenClaw install — turned into something they now describe as a member of the team.

Want an AI employee for your business?

We help businesses across New Zealand set up their own virtual employee with Hermes — from the first conversation to a secure, memory-backed assistant connected to your real tools and data. We handle the architecture, the security isolation, the VPN, the integrations, and the ongoing updates, so you get the upside without the maintenance burden.

If you've got a single point of knowledge in your team, a pile of repetitive questions, or an existing AI setup that isn't delivering, that's exactly the problem Hermes is built to solve. Get in touch and let's turn AI into measurable business outcomes.