Hermes Agent Just Got 5X Faster!

**Julian Goldie** (0:00)
There's a brand new update from Hermes that makes it way more powerful for agent swarms for using different tools across all the parallel agency have. And this makes it way more powerful. So you can see the announcement here, it literally just dropped a few hours ago. And it's called Segmented Tool Batch Dispatch.
I'll explain what that means in a second, but essentially this is a more powerful version of the Hermes Swarm System, which means your agents can run in parallel and Hermes have just changed how your agent runs tool calls. So before they only went in parallel, every single one of those tool calls was safe. Now every safe call fires at the same instant. So only the ones that must wait, wait, which means you can get way more done. So you can ask, and you can get multiple different tools running at the same time with Hermes agents. So for example, you could give it one task and it can read a file, search, fetch, check calls and list all at the same time, which means number one is faster, and number two, that your agent swarms work better together as well. So prior to today, tool calls would only work in parallel if all were safe to paralyze. Now you gain big speed ups for parallel tool calls, so long as any subset are paralyzable, according to Teddy. I'll explain in simple English what that means in a second. But here's the whole change. Before today, Hermes would only run if all of them were safe to run together.
If there was like one sort of a slightly questionable tool call inside the whole batch, then it all dropped back to each tool call being done in different steps. And that means that your tool calls are way slower. So now it runs everything together, and only holds ones that can't run, which means that you get a lot of big speed ups when it comes to using the system. So I would call this the one at a time problem, and here's what this solves. So picture asking your agent to do five things, like check your calendar, read three files, and pull the latest information in. None of them depend on each other. They could all happen at once, but your agent does them at one at a time. And then the first finishes, and then the second finishes, then the third, and you sit and wait whilst it just plugs down the same list. And so what happens with Hermes before is that five quick jobs would take like five times as long as one. The Hermes swarm system actually fixes out for good. So you can run multiple things at once. Now, you might say, okay, that sounds like it could break something. But the one, and if there's anything that shouldn't work or shouldn't be run inside the tool core system, that one is held back. The rest of the tool cores run at the same time. And that means that you get the speed of out the best.
And here's an example.
So if you look in terms of speed, this will be the previous one at a time system, and it will be super slow. This is if you're running everything at once. So it could be the difference between like 10 seconds, if you got two seconds per task, versus two seconds because every task is running. It's the same five jobs, but it's just the example of like a queue versus a swarm.
And that's really the difference between, you know, imagine if you can move five times faster when it comes to building, when it comes to shipping. Well, that's what you can do now because you got the system. Obviously, two calls don't take two seconds each, but that's just an example of how it can run. So you want to think of your agent as a worker with a to-do list. The old way was a queue. You do one task, you finish it, you start the next, you finish that one, down the line and the line again. The new way is a swarm. So every task that doesn't step on another one's toes goes out in the same seconds. Reading a file doesn't hurt reading another one. So both can go at once. The only thing that has to wait is a task that changes something the other is counting on. The one that one holds back, everything else already went. So you get the same list, the same correct result, but a fraction of the weight.
So you can ask Hermes to do something and do several things at once. You know, that could be, for example, like create an image, do a video, research this, read that file, edit this, et cetera. And it can do it all at once. And it looks at each call and ask one question, is it safe to run alongside the others? A read is safe, a write that the next step depends on is not. Every call that is safe fires in the same instance, and the one call that must wait for you, well, that's just held back.