Hermes’s research stack just got a pretty serious upgrade. Three recent additions give Hermes a much wider set of sources to investigate when you send it off to research something. Perplexity is now built directly into Hermes. You can select Perplexity in hermes tools, add your API key, and use it as the backend for both web_search and web_extract. So Hermes can use Perplexity to find sources and pull the relevant content from pages. Reddit is now directly readable too. A new bundled Reddit skill lets Hermes search subreddits, read posts and full threads, pull comments, and look through user activity without needing to operate Reddit through a browser. RSS, Atom, and JSON feeds were added alongside it. Hermes can read feeds directly from blogs, publications, release pages, and other sources. Even better, give it a normal website URL and it can try to discover the feed behind the site for you. Put all of that together and a research job can now pull from: Web → broad research and documentation Perplexity → search and page extraction Reddit → community discussion, problems, opinions, and workflows RSS / feeds → blogs, publications, releases, and direct updates Video → demos and discussions Code → what actually changed in the repo So something like: “Research everything happening around [tool] this week. What changed, what are people saying, what problems are coming up, and what are people actually building with it?” can become a real multi-source research job instead of one web search followed by a summary. Hermes can go looking across the web, community discussions, direct feeds, videos, and code, then bring the pieces back together into one picture.

Hermes browser automation just got a seriously useful upgrade. It can now run multiple browser workflows at once without taking over the browser you’re using. Hermes can now do browser work in the background without taking over the browser you are using. And the full setup is actually pretty powerful. For normal browser work, Hermes can launch its own isolated Chromium. No personal profile. No existing logins. No need to attach to your Chrome. So Hermes can be researching sites, scraping pages, testing a web app, filling forms, or running repetitive browser tasks while you keep using your computer normally. But when a job actually needs to be logged in as you, Hermes also has Real Profile mode. That lets it run against a managed copy of your browser login state, including the cookies and sessions it needs for sites you already use. So the mental model becomes: Clean browser → research, scraping, testing, anonymous workflows. Real Profile browser → X, Gmail, dashboards, SaaS tools, or anything else where your logged-in session matters. And it gets even better with multiple browser sessions. `browser_exec` can run named sessions, and each session can get its own separate Chromium instance. So one Hermes agent could be researching in one browser while another is testing something in a completely separate browser at the same time. Your browser. Hermes browser #1#. Hermes browser #2#. Hermes browser #3#. All doing different things without fighting over the same session. The newest piece is that local Browser Use now defaults to Hermes' own packaged Chromium instead of trying to attach itself to your installed Chrome. That removes a lot of the old friction around Chrome being open, remote debugging, browser popups, or Hermes interfering with the browser you are actively using. This is much closer to how browser automation should feel: You keep working. Hermes gets its own browser workspace and gets to work too.

Hermes command cheat sheet. Save this. You’ll probably need it again. UPDATED for the current Hermes command set, including a bunch of genuinely useful commands that have landed recently: • /goal • /loop • /heartbeat • /bg • /btw • /plan • /review • /refine • /moa • /busy The commands are spread across Desktop, CLI, messaging, and the terminal, so I rebuilt the sheet around what you’re actually trying to do. Sessions & context. Active work. Goals, loops & plans. Models & behavior. Skills, memory & tools. Automation & coordination. Recovery & troubleshooting. Plus the Messaging/Gateway and Terminal essentials. Verified against the current Hermes source today. Bookmark it and keep it around.

As part of the Mega Update 🚀 Your Hermes agent team no longer has to live on one computer. You can now spread your Bots across multiple machines and bring them together inside the same Group Chat 💻🤖 → 🖥️🤖 → ☁️🤖 Your laptop. A homelab. A VPS. Hermes Cloud. Each Bot still runs on the machine where it actually lives, with its own profile, memory, tools, model, and environment. So instead of forcing one computer to do everything, you can start building an actual distributed agent team. For example: • Keep a coding Bot on your main workstation where the repo lives. • Run an always-on monitoring or automation Bot from a homelab or VPS. • Put heavier or specialized work on another machine entirely. Then bring them into one Hermes Group Chat and let them work together. Setup is pretty straightforward: 1. Add your machines under Settings → Connections. 2. Create or keep the Bots on the machines where you want them to run. 3. Open New Group Chat and select Bots across those connections. Hermes handles routing each Bot’s turn back to its own machine. And this goes a step further with gateway-hosted Group Chats. The room itself can keep running without Hermes Desktop staying open. So Desktop becomes the place you manage the team, not necessarily the computer doing all the work. Multiple machines. One Hermes agent team. 🤯

Did I fail to mention this was probably the fastest Hermes update I’ve had yet? With everything that just landed, I was expecting the update itself to take a while. Couldn’t have been more wrong. Quick, smooth like butter, and done before I knew it. 🤌🚀

When Hermes starts acting weird, don’t start guessing. Start removing variables. Hermes has a really useful troubleshooting mode for exactly this: `hermes chat --safe-mode` Run the same task again. Safe mode temporarily strips away the major customization layers that could be affecting the result: • your `config.yaml` • project rules and SOUL • persistent memory • preloaded Skills • plugins • shell hooks • MCP servers If the weird behavior disappears, you just learned something important. Now you can narrow it down instead of randomly changing models, prompts, settings, or integrations. Test without your normal config: `hermes chat --ignore-user-config` Test without project rules, SOUL, memory, and preloaded Skills: `hermes chat --ignore-rules` You can combine them too. So the troubleshooting flow becomes pretty simple: 1. Reproduce the problem normally. 2. Try the same thing in safe mode. 3. If it changes, isolate config and rules separately. 4. Add your custom layers back until the problem returns. One small detail: `--ignore-user-config` skips `~/.hermes/config.yaml`, but credentials from `.env` can still load. So this is troubleshooting isolation, not a security sandbox. Safe mode does not tell you exactly what broke. It does something almost as useful. It quickly tells you which side of the fence to start looking on.

INSANE!!!! This might be the wildest Hermes update I’ve seen yet. 🤯 @Teknium basically unleashed an agent swarm on the entire Hermes codebase. One massive PR: • 1,652 subagent tasks • 111,352 tool calls • ~110 agents running concurrently • 4,271 commits • 2,655 files changed • More than 1.2 MILLION lines added or removed The result? Hermes cut more than a THIRD of its source code while preserving behavior. Same Hermes. Massively cleaner underneath. And dramatically easier for agents to work on. Absolutely ridiculous scale. 😂 Want these changes? Make sure Hermes is up to date: - Terminal / CLI: `hermes update` - Messaging Gateway: `/update` - Slack / Matrix: `!update` - Desktop: Settings → About → Check now → Update now

This is one of those Hermes use cases that feels straight out of the future 🤖🏡🚀 Hermes can become the manager for your entire smart home. The bridge is Home Assistant. Home Assistant is free, open source, and built to bring smart-home devices from more than 1,000 brands into one system. -Lights. -Locks. -Garage -Thermostats. -Sensors. -Cameras. -Switches. -And a whole lot more!!! Connect Hermes to Home Assistant and suddenly you do not have to manage all of that device by device. You can just tell Hermes what you want. “Is the house secure?” Hermes can check the relevant locks, door sensors, and alarm state. “Set the thermostat to 72 and turn off everything downstairs.” Hermes can inspect the devices involved and make the supported Home Assistant service calls. “Which lights are still on?” Hermes can check. “What can I control in the living room?” Hermes can find out. And it can go beyond commands. Through the Hermes Gateway, Home Assistant can also feed selected live state changes back to Hermes. A door opens. A sensor triggers. The HVAC mode changes. You choose what Hermes should watch, so it does not need to react to every tiny event your house generates. Setup starts with a Home Assistant Long-Lived Access Token: `HASS_TOKEN` And `HASS_URL` if your Home Assistant instance uses a different address. Home Assistant remains the system connecting and controlling your devices. Hermes becomes the part you actually talk to. Instead of managing a collection of smart-home apps and devices, you can give the job to Hermes.

Okay, this is a big update and shifts how Hermes learns from experience, not just how it stores memory 🤯 Hermes just changed what it does with the things you teach it while it works. Hermes is moving from storing lessons to organizing them by context, so what it learns can come back in the right place, for the right kind of work, at the right time. Task-specific lessons are now meant to improve the Skill for that work instead of filling up global memory. So if you correct Hermes while it is doing something you repeat often, like writing, coding, research, or running a workflow, Hermes is now instructed to put that reusable lesson into the relevant Skill. Think: Reusable lesson for a type of task → Skill Stable fact that should matter everywhere → Memory Temporary detail for this conversation → Session history For example: “Always format these reports with a summary first.” That belongs with the reporting Skill. “I prefer concise responses.” That can belong in persistent memory if it applies everywhere. “Use this temporary staging URL today.” That can stay with the session. The reason this matters is simple. Skills only need to load when they are relevant. Global memory follows Hermes much more broadly and has a limited budget. So instead of every task-specific correction competing for space in memory, the lessons Hermes learns from doing a specific kind of work can make that Skill better the next time it needs it.

Start a Hermes session in one place. Continue it somewhere else. You can begin working with Hermes in one place and continue the same session somewhere else. Start in Desktop, then pick it up from the CLI or TUI. Start from the terminal, then hand the live session over to Telegram, Discord, Slack, WhatsApp, Signal, or Matrix. Hermes can also be running remotely on another computer, home server, or VPS while you connect to it from somewhere else. The useful part is that you are not rebuilding the agent every time you change interfaces. The session history and Hermes state stay with the underlying Hermes profile. There are two basic ways this works: Resume the existing session when moving between interfaces that share the same Hermes session store. Or hand a live terminal session to a messaging platform with: `/handoff ` For example: `/handoff telegram` `/handoff discord` `/handoff slack` Same command. Different destination. For messaging handoffs, you first set the destination once from that platform with: `/sethome` Then Hermes knows where that handoff should go. Coming back is just a resume. From the terminal: `hermes --resume ` Or use `/resume ` inside Hermes. Desktop can open saved sessions from the interface. So Hermes does not have to be: “the agent on my desktop” or “the bot in my Telegram.” It can be the same Hermes you work with from different devices, interfaces, and locations. Start where the work makes sense. Continue where you happen to be.

Want to run Hermes locally but have no idea which model your computer can handle? Already running local models but want an easier way to choose, download, and manage them? Hermes Desktop can now help with both. Local models are becoming a first-class provider inside Hermes. Hermes maintains a curated catalog of supported local models and can compare those options against the hardware actually sitting in your machine. Launch Desktop with: `hermes desktop --local` Then go to: Settings → Providers → Local models Hermes checks things like your: • GPU • VRAM • RAM • unified memory Then it can rank the available models by what your machine can realistically handle, recommend a model and the right build for your hardware, and explain why it picked it. If you are starting from zero, there is a quickstart that can handle the rest: • install the llama.cpp runtime • download the recommended model • start the local server • make the model available inside Hermes No hunting for GGUF files, guessing which quantization you need, or manually wiring everything together. And if you already have models downloaded through Hermes' local setup, Desktop can also show what is installed, what is loaded into memory, what is active, and let you manage them from the same place. Downloaded local models can then appear alongside your other providers in the normal Hermes model picker. So the new flow is much closer to: Your hardware → Hermes recommends what fits → download → run locally. And for people already running local models: Your models → Hermes helps manage them.