We get a lot of questions about how we run Hivory. What are we actually using? How does the agent setup work? And can we do this for clients? This is my attempt to answer both properly.
The tech stack
We tested a lot before landing on what we run today. After going back and forth on different setups, we're on OpenClaw running on a dedicated Mac mini M4. The Mac mini was a deliberate call. We wanted the technology physically siloed, not living in a cloud environment where our visibility is limited. A dedicated machine gives us full control, a clean security boundary, and a hard killswitch if we ever need one. When agents are touching client data, email accounts, and live business systems, that's not a nice-to-have. OpenClaw connects to Claude. We've written about why we moved to Claude — the reasoning quality is genuinely different, and it's what Artemis runs on. We're always evaluating though. Perplexity Computer is something we're watching closely. From there, everything ties into Notion and Slack. Notion is where tasks, content, and client work live. Slack is the interface for the team, and where agents push outputs, alerts, and updates. It means anyone on the team can interact with the system without touching anything under the hood. One rule we've kept across our entire stack: if a tool doesn't have an API, we generally don't use it. An agent that can't connect to a tool isn't useful. It's also had the side effect of forcing us toward better-built software, which has been its own benefit.
The agent stack
Artemis is the main agent — I covered what it does in that piece and won't repeat myself here. What's changed since then is the workforce around it. Each client now has their own set of specialised agents, each running a specific job. A content agent that researches trending topics, finds the right angle for that client's audience, writes the article, optimises it for search, and posts it. That runs daily. We review and occasionally redirect, but we're not in the production loop. An SEO agent working across each client's site on an ongoing basis. It's not waiting for a monthly briefing. It finds gaps, fixes issues, and builds rankings over time. A reporting agent that tracks what's actually happening — traffic, leads, performance — and surfaces what needs attention. Not a dashboard you have to go and check. Actual analysis, pushed to us when something warrants it. Artemis sits above all of this. It handles strategy, client engagements, and anything that requires real thinking rather than systematic execution. The specialised agents run their own lanes. Artemis runs the operation.
Will we set this up for clients?
Honest answer: not yet, but that's deliberate. We're currently piloting an agent offering with one client to understand what actually delivers value and what's still too fragile to hand over. A few things make me cautious about scaling this before we've got it properly figured out. The technology is changing fast enough that anything we set up today could be outdated in six months. Managing that in our own system, where we control everything, is straightforward. Managing it across multiple client deployments is a different problem entirely. There's also a real safety consideration. Agents that touch live business systems carry real risk. A misconfiguration in our own environment affects us. In a client's environment, with their data and their integrations, the stakes are different. That's not a reason not to do it. It's a reason to get it right first. And honestly, it's still technical. Proper setup requires ongoing maintenance. The tooling isn't mature enough yet to hand off and walk away from. What we're more comfortable offering right now is managing agents on behalf of clients, the same model we use for content and SEO today. We run the system. They get the output. That's something we can stand behind.
The broader offering will come. This industry is moving too fast for it not to. But I'd rather get it right than get it out the door.
