Hermes can get a second opinion before it acts. Or a third. Or a fourth. That’s essentially what Mixture of Agents does. Inside Hermes Desktop, you can choose several reference models to independently think through the same problem. Those models don’t run tools or take over the task. They... act as advisors. Their responses are then passed to the Aggregator, which is the model that actually responds, uses tools, and continues the Hermes agent loop. You can even mix providers across the setup. So instead of relying on one model to reason through a difficult problem alone, your acting model can make the call after seeing several independent perspectives. This video shows how simple that setup is inside: Settings → Model → Mixture of Agentsshow more

don't forget about /review too - its a little more controlled and as-needed then MoA

You know /review is always on deck. I was letting MoA have its moment 🥳

Here's a forgotten feature in Hermes that I never see anyone post about, but I think that's really what it's meant for!

Hermes Achievements is fun for sure. I definitely should cover this.

Please do! I did and 0 people saw it :) lol

One unrelated question. How can I get Kanban in Hermes Desktop? I do not have it in my sidebar.

Kanban ships disabled by default. Go to Settings → Plugins → Kanban and enable it. Once enabled, Kanban appears in the left sidebar.

Thanks for featuring my agents!!! @Teknium

Saw this consumes up to like 20x usage. Is there a mix you can recommend? With limited spend I don‘t imagine putting fable or sol there, but maybe a cheaper model like muse could work well without exploding costs?

If cost is the concern, /review is probably the better fit. I’d save Mixture of Agents for higher-stakes work where several independent opinions are worth the extra usage.

I am certain that it is extremely effective just curious about the inner details

The reference models answer first, then the aggregator gets those responses and uses them to produce the final answer or tool call.

Gonna test with legal queries and repor on the outcome

It's very complicated to mix up models efficiently,after 1 month I think best combo is a thinking aggregator with vision capabilities and just one reference that must be almost same level intelligence as aggregator but very different provider an training data like usa/china model

Thanks, I understand how it works much better now. It's really useful in real life.

It has its places for sure. Best when mission critical.

I've never understood the benefit of doing stuff like this in terms of compute cost efficiency. Adversarial reviews (or other ways of critiquing) using sequential agent handoffs are significantly better ways to catch holes in the plan under budget. Loop engineer the problem away.

the useful part is the separation of roles. advisors can disagree, but one model still owns the tools and final action. that sounds much easier to debug.

long tool calls still die at the client timeout. start+poll with a real workflow id beats hoping the socket stays open.

multiple models arguing before it acts, interesting

hermes is so full of random shit. makes it so hard to use.

The weak link becomes the aggregator deciding which advice deserves weight

How's that different from a conventional orchestration setup in agents.md?