Not to mention is open source and MIT licensed...!
Looks like another contender has entered the stage!
Report
@chrismessinathings are stepping up with builds so I don't think it will be too long before we start to see a once prompt build platform. This is just the start.
Report
@chrismessina VERY interesting!! The persistent memory is what I'm most keen on - would make life a little easier, lol! And because I'm new to this world - is this something that works for Solos as well as enterprise-level?
Report
Yesterday, while configuring the Hermes agent on my dedicated server, I started wondering if there’s a desktop orchestration tool for managing multiple hermes agents running on different instances.
I tried Hermes Desktop, but it doesn’t seem to support that use case yet.
Has anyone found a good solution for this?
Report
@ilya_makarov2 this is going to be a real problem as more people self-host agents. right now everyone's managing them one at a time through ssh or separate dashboards. the moment you have 3-4 agents doing different things on different machines you need some kind of central control plane. surprised more tools aren't building for this yet
Report
Most local agent setups collapse the moment you point them at a real codebase because the context window fills up and the tool calls start going sideways. Curious how Hermes handles repo-scale tasks where the relevant code is spread across a dozen files. Does it do any chunking or retrieval to stay under the limit, or does it lean on you to scope things down manually before handing off?
Report
Intel Macs are not supported.
Report
Running a capable open-source agent natively on macOS, Windows, and Linux is a meaningful step. Browser-based agent wrappers have too much overhead for tasks that need low-latency local tool access. We've found persistent agent state across sessions to be one of the harder problems when building on top of LLMs. How does Hermes Desktop store and retrieve long-term memory without context window blowup as the agent accumulates history?
Report
people on reddit are already hyped about this and I get why. an open source agent that actually remembers what it learned and gets better the longer you run it is not something you see every day. one thing I kept seeing people ask about though... if you already have hermes running on a server, is there a way to connect the desktop app to that without going through a full fresh setup? would love to just point it at an existing instance and go
Report
Curious where you see the boundary between an agent OS and a collection of specialized agents.
As more teams build dedicated coding, research, and workflow agents, do users seem to prefer one general agent coordinating everything or multiple agents with distinct responsibilities?
Report
I tried using the official download link on Nous website. It downloaded the wrong version. So for Intel Mac users. Use “hermes desktop” and you’re in 😊 works like a charm. And thanks Nous for being so gracious with the whole agent. Using it daily here in 🇿🇼🇿🇦
The agent that grows with you: An MIT-licensed local agent that accumulates knowledge over time is exactly the direction I'm rooting for. How does Hermes represent what it has "learned" — structured memory, embeddings, something else? The memory representation question is the one I find most interesting in this whole category.
Report
"The agent that grows with you" — interesting claim. What does that actually mean in practice — does it log failed tasks and adjust its approach over time, or is it more about the open-source model getting fine-tuned community contributions that surface back into the tool?
Replies
Raycast
On the heels of the demise of Windsurf, clearly the future is agents writing software, not humans.
Which is why Hermes Desktop is very interesting.
It combines elements of @Claude Code or @OpenAI Codex CLI with that of @OpenClaw ... and supports a large number of connectors out of the box:
Not to mention is open source and MIT licensed...!
Looks like another contender has entered the stage!
@chrismessinathings are stepping up with builds so I don't think it will be too long before we start to see a once prompt build platform. This is just the start.
@chrismessina VERY interesting!! The persistent memory is what I'm most keen on - would make life a little easier, lol! And because I'm new to this world - is this something that works for Solos as well as enterprise-level?
Yesterday, while configuring the Hermes agent on my dedicated server, I started wondering if there’s a desktop orchestration tool for managing multiple hermes agents running on different instances.
I tried Hermes Desktop, but it doesn’t seem to support that use case yet.
Has anyone found a good solution for this?
@ilya_makarov2 this is going to be a real problem as more people self-host agents. right now everyone's managing them one at a time through ssh or separate dashboards. the moment you have 3-4 agents doing different things on different machines you need some kind of central control plane. surprised more tools aren't building for this yet
Most local agent setups collapse the moment you point them at a real codebase because the context window fills up and the tool calls start going sideways. Curious how Hermes handles repo-scale tasks where the relevant code is spread across a dozen files. Does it do any chunking or retrieval to stay under the limit, or does it lean on you to scope things down manually before handing off?
Intel Macs are not supported.
Running a capable open-source agent natively on macOS, Windows, and Linux is a meaningful step. Browser-based agent wrappers have too much overhead for tasks that need low-latency local tool access. We've found persistent agent state across sessions to be one of the harder problems when building on top of LLMs. How does Hermes Desktop store and retrieve long-term memory without context window blowup as the agent accumulates history?
people on reddit are already hyped about this and I get why. an open source agent that actually remembers what it learned and gets better the longer you run it is not something you see every day. one thing I kept seeing people ask about though... if you already have hermes running on a server, is there a way to connect the desktop app to that without going through a full fresh setup? would love to just point it at an existing instance and go
Curious where you see the boundary between an agent OS and a collection of specialized agents.
As more teams build dedicated coding, research, and workflow agents, do users seem to prefer one general agent coordinating everything or multiple agents with distinct responsibilities?
Polyvia
The agent that grows with you: An MIT-licensed local agent that accumulates knowledge over time is exactly the direction I'm rooting for. How does Hermes represent what it has "learned" — structured memory, embeddings, something else? The memory representation question is the one I find most interesting in this whole category.
"The agent that grows with you" — interesting claim. What does that actually mean in practice — does it log failed tasks and adjust its approach over time, or is it more about the open-source model getting fine-tuned community contributions that surface back into the tool?