Agent OS Hermes is the layer that finally completed my stack after months of feeling like something was missing, and the moment Hermes slotted into the Research layer was the moment my whole Agent OS stopped feeling like a demo and started feeling like a business. I have built a lot of half-stacks in my time. This is the first time the picture has felt whole, and Hermes is the reason why.
This article walks through why most Agent OS stacks are incomplete without a research layer, what Hermes actually does in that slot, how it wires into the rest of the stack, and how to grab the Hermes-specific bonuses inside the AI Profit Boardroom.
Want the complete Hermes Agent OS as a bonus? Inside the AI Profit Boardroom, I include the Hermes Money Machine, the Hermes Quick Deploy Kit, the Hermes Swarm Playbook, Hermes Agent OS with 10 revenue builds, and the Hermes 30 Day Roadmap. → Get inside
Why Most Agent OS Stacks Feel Incomplete
The pattern I see in nine out of ten founder setups is the same. They have Claude in one tab, OpenClaw on the desktop, maybe an Obsidian vault, possibly OMI capturing screen and mic notes. The pieces are all there, but the stack still feels incomplete. Something is missing, and they cannot quite put their finger on what.
The missing piece is the research layer. Claude is brilliant at thinking. OpenClaw is brilliant at execution. Obsidian is brilliant at memory. None of them are research engines. None of them know how to take a research brief, break it into Kanban tasks, run a list of skills on a schedule, and post the results back to the vault while you sleep.
That is the slot Hermes fills. It is not competing with Claude or OpenClaw — it sits between them, doing the work neither of them is designed for. Once Hermes is in place, the stack stops feeling like four separate tools and starts feeling like one operating system with four specialised layers.
The wider Agent OS thesis lives in What Is Agent OS and Agent OS. If you want the founder-view companion to this article, that is in Hermes Agent OS.
What The Research Layer Actually Does
The research layer is the layer that does the deep work between the planning and the execution. It is not a chat surface — it is an engine. The difference matters because most founders are still thinking about AI as conversation when they should be thinking about it as production.
A chat surface gives you one exchange at a time. You type, the model responds, you read, you type again. That is great for thinking and writing, but it is terrible for the kind of work that takes hours and pulls from multiple sources.
A research engine runs in the background. You give it a brief, it breaks the brief into discrete tasks, it works through them in order on a Kanban-style board, and it posts the output back to your shared memory layer when it is done. You can be doing something else entirely while the research happens.
The four things the research layer does that a chat surface cannot are scheduled runs, multi-source aggregation, structured task tracking, and MCP server exposure for cross-tool calls. Those four together are what make Hermes the completing piece of the stack.
Why Hermes Is The Right Research Engine
There are a handful of agents that could in theory fill the research slot. I have tried most of them. Hermes wins for four specific reasons that matter every day.
First, Hermes is open source and runs locally. The research layer touches sensitive data — your goals, your client notes, your strategic thinking — and you want that data on your machine, not in a cloud database you do not control. Local-first is the only sensible default.
Second, Hermes has a skills and plugins library that you can extend. I run Reddit research, competitor scraping, content drafting, and lead enrichment as my default skills, plus a handful of custom ones I have written. The library grows over time. That is not true of closed agents.
Third, Hermes has Kanban-style task lists native. The research layer needs structure for jobs that take longer than a chat, and the Kanban view is the right abstraction. You see what is in progress, what is blocked, and what is done at a glance.
Fourth, Hermes can run as an MCP server. That means other tools can call it directly — Claude Desktop, OpenClaw, your IDE — instead of being locked inside one dashboard. The same Hermes instance serves the whole stack. The wiring lives in Hermes MCP Server.
The deeper framework view of Hermes lives in Hermes AI Agent Framework 2026.
How Hermes Slots Into The Stack
The cleanest way to think about this is the Goldie Mission Stack — four layers that together form the complete Agent OS, with Hermes sitting in the Research slot at layer three.
Layer one is Intelligence — Claude or Claude Code. This is where the planning, the reasoning, and the strategic decisions happen. Anything that needs founder-level thinking gets routed to Claude first.
Layer two is Execution — OpenClaw. This is the local agent gateway that routes tasks between agents, manages sessions, and handles multi-agent coordination across the whole stack.
Layer three is Research — Hermes. This is the layer that runs the tool calls, the Kanban task boards, the skills and plugins, the scheduled multi-step workflows, and the deeper research tasks. Hermes does the work while Claude plans and OpenClaw routes.
Layer four is Self — the Obsidian vault plus OMI. This is the layer that captures your goals, your daily journal, your screen and mic notes, and feeds context to every agent above it. The vault is what makes the rest personalised.
That is the complete stack. Pull Hermes out and layer three is missing — the research never happens automatically, the Kanban tracking is gone, the skills library is gone, and the scheduler is gone. Hermes is the completing piece.
The Obsidian Vault Connection
The Obsidian connection is the move that turns Hermes from a generic research agent into something that knows your business. This is the wiring most founders do not realise matters until they have done it once.
You point Hermes at the folder where your Obsidian vault lives. Hermes reads the vault every time you give it a research brief. From that moment, every Hermes output is informed by what is already in your notes — your goals, your client list, your offer details, your strategic plans.
The difference shows up immediately. The same prompt produces a very different output. Generic AI gives you a listicle. Hermes-with-vault gives you a strategic recommendation specific to your real business.
The video example sums it up. I asked Hermes "based on my Obsidian vault, give me some ideas on what I should automate today" and it came back with personalised automation ideas pulled from my Goldie Agency notes, my AIPB community notes, and my current Hermes build work. That output is impossible without the vault wiring.
The wider Self Layer write-up lives in OMI Obsidian.
Hermes As MCP Server — The Cross-Tool Bridge
The MCP server mode is the upgrade that makes the research layer accessible to every other tool in your stack. Without it, Hermes is locked inside the Agent OS dashboard. With it, Hermes becomes the research engine the whole stack calls.
When Hermes is running as MCP, Claude Desktop can fire research jobs at it from any conversation. Your IDE can pull context from it. OpenClaw can route tasks to it without bespoke integration code. Any MCP-aware client can hit the same Hermes instance.
That is the difference between a research agent and a research engine. The agent is a thing you use. The engine is a thing the whole stack uses. MCP server mode is the upgrade that flips Hermes from one to the other.
The setup is a config flag plus a port number. Once you have a working Hermes install, adding MCP server mode is a ten-minute job. The full wiring lives in Hermes MCP Server.
The Free $0 Hermes Stack Anyone Can Build
The whole completing-the-stack story would not matter as much if Hermes cost £200 a month. The fact that the entire research layer can run for £0 is what makes this accessible to every founder, not just the ones with a tools budget.
Hermes itself is open source and free. Step 3.5 Flash on OpenRouter gives you a free API tier that handles the model layer without a paid plan. Obsidian is free for personal use, which covers the vault. Your laptop covers the hosting. Total cost to run the full research layer is £0.
That is the answer to the most common objection. People assume a "real" research engine costs hundreds a month. The honest truth is anyone can build the full Agent OS Hermes stack on the hardware they already own, with zero subscriptions.
I unpack the broader cost story in AI Agent OS and the Claude comparison in Agent OS Claude.
A Real Research Workflow That Completes The Day
Here is the workflow that made me realise the research layer was the completing piece. Picture a normal Monday morning before Hermes was wired in. I open my laptop. I have no idea what the competitors did over the weekend, no idea what Reddit said about my niche, no idea what content gaps need filling this week. I start the day cold.
Now picture the same Monday with Hermes wired in. I open my laptop. Sitting in my Obsidian vault are three fresh notes — the weekly competitor briefing Hermes pulled at 5am, the weekend Reddit pain-point summary, and the prioritised content gap list for this week. I start the day with three pages of research that took zero of my time.
That is the difference the research layer makes. Same founder, same hours, same energy — but the day starts with research instead of without it. The leverage compounds because every day starts with more context than the last.
The example I keep coming back to from the video is the "based on my Obsidian vault, give me some ideas on what I should automate today" prompt. Hermes came back with personalised ideas specific to my business — not generic AI suggestions. The research layer is what makes that level of personalisation possible.
The Stack Before And After Hermes
| Layer | Without Hermes | With Hermes |
|---|---|---|
| Intelligence (Claude) | Works fine | Works fine |
| Execution (OpenClaw) | Works fine | Works fine |
| Research | Missing | Hermes |
| Self (Obsidian + OMI) | Works fine | Works fine |
| Scheduled background work | None | Automatic |
| Kanban task tracking | None | Native |
| Skills and plugins library | Custom MCPs only | Open library |
| MCP server mode | Client only | Server too |
| Daily research compounding | No | Yes |
| Vault-aware research | Manual | Automatic |
The right-hand column is the version of the stack that actually completes.
Inside The AIPB Hermes Bonuses
If you want to skip the manual build and grab a prebuilt research layer, that lives inside the AI Profit Boardroom as the Hermes bonus stack.
The Hermes Money Machine is the offer template for monetising Hermes-powered research. The Hermes Quick Deploy Kit is the install pack that gets you live in under an hour. The Hermes Swarm Playbook covers the multi-agent patterns when one Hermes instance is not enough. Hermes Agent OS with 10 revenue builds is the pack of real business builds I have shipped on Hermes. The Hermes 30 Day Roadmap takes you from install to a fully working Goldie Mission Stack.
Membership sits at £59 per month locked forever, with the twin guarantee making the trial basically risk-free. Three thousand plus members are inside, with five weekly live coaching calls and daily Q&A.
Want hands-on help completing your stack with Hermes? AIPB runs five weekly coaching calls plus daily Q&A — bring your screen and we will wire the research layer together on the call. → Join here
How To Tell If Your Stack Is Missing The Research Layer
The diagnostic I give members on coaching calls is simple. Ask yourself three questions about your current stack.
First, does any research happen automatically while you sleep? If the answer is no, you are missing the research layer.
Second, can you see your in-flight research jobs on a task board at a glance? If the answer is no, you are missing the research layer.
Third, can other tools in your stack call your research engine as an MCP server? If the answer is no, you are missing the research layer.
Three nos = your stack is incomplete and Hermes is the completing piece. One or two nos = you might be partway there. Three yeses = you already have the research layer wired and you can probably skip this article.
For most founders I talk to, the answer is three nos. That is fine. The fix is one hour of wiring.
FAQ — Agent OS Hermes As The Research Layer
What is agent os hermes in one sentence?
It is the open-source research layer of an Agent OS — Hermes runs tool calls, Kanban tasks, skills, plugins and scheduled workflows that complete a stack Claude-only setups cannot match.
Do I need Claude and OpenClaw as well?
You do not need them to run Hermes — Hermes works standalone. But the full Goldie Mission Stack pairs Hermes with Claude (planning) and OpenClaw (routing) for a complete operating system.
Is the research layer really worth adding?
If you have any work that should run while you sleep, yes. The scheduled background research alone usually saves several hours a week for content-led businesses.
Can Hermes run for £0?
Yes — open-source Hermes plus Step 3.5 Flash on OpenRouter plus Obsidian plus your laptop = £0 monthly cost. I have run the stack for months at that price point.
How does Hermes know about my business?
Through the Obsidian vault connection. You point Hermes at your vault, it reads your notes, and every output is informed by your real context.
What if I am not technical enough to wire this myself?
The Hermes Quick Deploy Kit inside AIPB is the prebuilt version, and the five weekly coaching calls cover the wiring step by step.
About Julian
I'm Julian Goldie — AI entrepreneur, SEO expert, and founder of the AI Profit Boardroom with three thousand plus members. I help business owners scale with AI agents, automation, and SEO.
- Founder of a seven-figure SEO agency, Goldie Agency
- Author of multiple AI automation playbooks
- Five weekly live coaching calls inside the Boardroom
- Free AI training inside the AI Money Lab
→ Get my best AI training inside the AI Profit Boardroom
Latest Updates
- Hermes Agent OS — the founder-view companion piece.
- Hermes MCP Server — the cross-tool bridge wiring guide.
- Hermes AI Agent Framework 2026 — the framework view of the research layer.
Also On Our Network
- 🌐 Read on bestaiagentcommunity.com
- 🌐 Read on aiprofitboardroom.com
- 🌐 Read on juliangoldieaiautomation.com
- 🌐 Read on aisuccesslabjuliangoldie.com
Related Reading
- What Is Agent OS — the beginner explainer.
- Agent OS Claude — the Claude side of the stack.
- AI Agent OS — the wider thesis.
- Hermes Agent Community — where to plug in with other Hermes builders.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
Agent OS Hermes is the research layer that takes a half-built stack and turns it into a complete operating system — wire it in, point it at your vault, and agent os hermes will run the research that makes the rest of your stack worth running.