Hermes Agent v0.15.0 Velocity Update is INSANE

Ajouté : 2026-06-01

[00:00:00]Hermes Agent V0.15.0 velocity update is insane. What if your AI agent could remember every project you've ever worked on? What if it lived on your server and got smarter the longer you used it? Most people are still copy-pasting prompts into a chat box and they have no idea this exists.

[00:00:16]It's free, open source, and it just got way faster. Hey, I'm the digital avatar of Julian Goldie. I help people learn AI tools and use them in their work. Today, I'm walking you through the brand new Hermes Agent V0.15.0 release, the velocity update. Stick with me because the speed jump in this version is something you have to see. First, what is Hermes Agent? Hermes Agent is an open-source AI agent built by Nous Research, free under the MIT license.

[00:00:44]Not a coding assistant stuck inside one app, not a chat bot that forgets you when you close the tab. It lives on your own server. It remembers what it learns and it gets more capable the longer you run it. Think of it like hiring a smart helper. Day one, they know nothing.

[00:00:58]After a few weeks, they know your projects, your files, your workflow, and how you like things done. It connects to apps you already use: Telegram, Discord, Slack, WhatsApp, Signal, email, the command line. Start a chat on your phone and pick it back up on your laptop. Same agent, same memory. Now, here's why it's worth paying attention to. It grows with you. Most AI tools start over every time. Hermes saves what it learns and builds its own skills. It runs on a schedule. Give it a job in plain English like check my reports every Monday and it just does it. It splits work across smaller agents. Big task broken into pieces run in parallel. It's sandboxed.

[00:01:34]Jobs run in safe containers: local, Docker, SSH, Singularity, or Model. And it has full browser and web controls.

[00:01:41]Search the web, browse pages, see images, generate images, talk out loud, all from one agent. Now, let me talk about GitHub because that's where everything lives. The full project is on GitHub under Nous Research/hermes-agent.

[00:01:54]Over 171,000 stars and almost 29,000 forks. Over 321 contributors pushed code into this single release. On GitHub, you'll find the source code, the documentation, the release notes, and the install scripts.

[00:02:08]If you hit a bug, you can open an issue.

[00:02:10]And if you're a developer, you can write your own plugins because the whole thing is plugin based. Now, let's get into what's new in v0.15.0.

[00:02:19]The name is earned. This update shipped 1,302 commits, 747 merged pull requests, and closed over 560 issues. 15 critical and 65 high priority. First, the core agent file got cut down by 76%.

[00:02:37]The main file that runs the conversation loop dropped from 16,083 lines to 3,821 lines. It loads faster, runs faster, and is easier to debug. Second, cold start speed on Termux, the Android terminal cold start went from 2.9 seconds to 0.8 seconds. The version command alone got 63% faster. That benchmark now beats Codex CLI head-to-head. Third, the agent uses 47% fewer function calls inside a conversation. For a 31-turn chat, the calls dropped from 399,000 to 213,000.

[00:03:17]Fourth, session search. The old version took 30 seconds per call. The new version runs in 20 milliseconds, 4,500 times faster. Fifth, Kanban became a full multi-agent platform. Type Hermes Kanban swarm, and Hermes builds a team of mini agents to attack a big task.

[00:03:34]Root agent, parallel workers, a verifier, and a synthesizer. They share a blackboard so they can talk to each other. Sixth, prompt wear defense. This protects your agent from a new kind of attack where bad instructions sneak in through tool output, memory, or installed skills. Hermes scans all of that at three different points and blocks the attack. Seventh, Bitwarden integration. Instead of pasting every API key into a config file, you give Hermes one Bitwarden token and it pulls every credential from there. Rotate a key in Bitwarden and Hermes picks it up automatically. Eighth, skill bundles.

[00:04:10]Group skills together and load them all with one {slash} command. If you have a writing workflow that uses four tools, you type {slash} writing day and they're all active. Ninth, NTFY support as the 23rd messaging platform. NTFY sends push notifications without an account or an API key. Your agent can ping your phone the second a task finishes. 10th, a curated MCP catalog. Run Hermes MCP and you get an interactive picker for trusted MCP servers. Now, here are two things you could actually build with Hermes once it's installed. The first is a single HTML page that takes inputs you choose, runs the math, and shows the result. You describe what you want in plain English. Hermes writes the HTML, the CSS, and the script. You get a finished page you can drop on any website in minutes. The second is a small Python or React app you can run locally. Same flow. You describe the inputs, the logic, and the output.

[00:05:07]Hermes uses its terminal and patch tools to write the file, run it, fix any errors, and even host it locally for you to test. The pattern is simple. You describe the tool. Hermes plans the build, writes the code, runs the code, and hands you a working file. Quick story. A lot of people watching this are sitting on a tool like Hermes and they have no clue where to start. They install it, they poke at it, they get stuck on one config issue, and they give up. That's why we built the AI Profit Boardroom. Inside the Boardroom, we have full walk-throughs on agent setups exactly like this one. Step-by-step videos that show you how to install Hermes from scratch, configure it for your work, and connect it to the messaging apps you actually use. We run live coaching calls every week. So, if you get stuck on something specific like setting up the Bitwarden integration or running your first Kanban swarm, you can hop on a call and get unstuck in minutes. There's also a 30-day road map built around getting you from zero to running automations with agents like Hermes, plus prompts and templates you can copy and use the same day. If you're serious about adding agents to your workflow, this is where the heavy lifting happens. All right, let's get you installed. Installation is one line.

[00:06:14]Open your terminal and paste this: code-fs-sl-https://hermes-agent.nausresearch.com/install.sh.

[00:06:23]Then pipe that into bash. That pulls the install script and runs it. Hermes sets itself up on your machine. Once it finishes, type hermes setup. You pick which models to use, whether you want messaging gateway features, and it asks for your basic credentials. Now, if you want to connect Hermes to a web chat interface like Open Web UI, here's what you do. Open the file at tilde /.hermes/.env.

[00:06:47]Set API_SERVER_ENABLE to true. Set API_SERVER_KEY to a secret value. Save the file, then run hermes gateway. That starts the local server on your own machine. You'll see a line that says API server listening on port 8642.

[00:07:07]Then launch Open Web UI in Docker, point it at the same port and the same key, and you have a clean web chat front end for your agent, conversation history, user accounts, the works. If you want to run more than one agent at a time, Hermes supports profiles. Each profile gets its own port, its own memory, and its own skills. So, you can keep your work agent and your personal agent completely separate. If you're on Linux without Docker Desktop, the host.docker.internal address won't resolve by default. You either add a host mapping flag, use host networking, or point at the Docker bridge IP directly. The docs walk through all three options for you. Once it's running, type Hermes in your terminal to chat with it directly or message it from Telegram, Discord, or any of the other 23 messaging platforms after you connect them. A few pro tips for beginners. Start with one messaging platform. Don't connect everything on day one. Pick Telegram or Discord. Get comfortable, then add more. Turn on memory early. The longer Hermes runs with memory on, the better it gets. Use skill bundles. Group your common workflows with one {slash} command and save hours every week. Move your API keys into Bitwarden Secrets Manager, a one-time setup that pays off forever.

[00:08:18]And try Kanban for big projects, even if you're not a coder. Type Hermes Kanban swarm and watch Hermes break the task down for you. That's the full velocity update. And one more thing, when you install Hermes today, you're going to run into the same questions everyone does. Which model do I use for what task? How do I set up the Kanban swarm for my workflow? Where do I store my keys safely? How do I connect a second messaging platform without breaking the first? That's why the AI Profit Boardroom exists. Full Hermes setup tutorials, live coaching calls every week, a 30-day roadmap from install to real automations, and a library of prompts and skills you can use straight away. If you're ready to actually use this thing, come join us inside. Head to process, SOPs, and 100 plus AI use cases like this one, join the AI Success Lab. Links in the comments and description. You'll get all the video notes from there, plus access to our community who are crushing it.