Keep exploring · more guides
Grok 4.7 just dropped, and I'm about to show you how to run it inside Hermes Agent in about two minutes.
This gives the newest Grok a real body, so it can build apps, run its own tests, read the web, and fix its own mistakes before it ever says done.
One command creates the profile, one line points it at Grok 4.7, and it shows up inside the Agent OS on its own.
I'll show you five things it did for me this morning, including a full working app it built and checked without me touching a single line.
Stick with me to the end, because there's one flag that runs the same job through two different brains, and it's the most useful trick in this whole setup.
You can. And it will write you code in a chat box that you then copy out, paste in, and run yourself.
Inside Hermes it gets a terminal, your files, your skills, and the job of checking its own work.
What you're watching: the real commands from this morning. One profile create command, cloned from a profile that already had my OpenRouter key, then one line changed in config.yaml to x-ai/grok-4.7.
model: default: x-ai/grok-4.7 provider: openrouter base_url: https://openrouter.ai/api/v1 toolsets: - hermes-cli agent: max_turns: 90 terminal: backend: local cwd: ~/.hermes/profiles/grok-4-7/workspace
It's a text file with one line you change: the model name.
Clone from a profile that already works and you don't even type the keys.
What you're watching: the status command showing Model x-ai/grok-4.7, then a one-line question. The reply came back in 8.4 seconds: “I'm Grok 4.7 via OpenRouter (x-ai/grok-4.7), and 17*23 = 391.”
What you're watching: the real Agent OS Hermes tab. I pick grok-4-7 from the chat profile list, type a question, and the reply streams back: “I'm Grok 4.7 (x-ai/grok-4.7), running inside Hermes Agent…”
No. The Agent OS reads your Hermes profiles folder every time it loads the list.
Make the folder, and grok-4-7 is already in the chat profile menu.
What you're watching: the real tool calls from the session, in order. It read a design skill, looked at the folder, wrote a 20,436-byte file, then ran its own parse check and node --check before replying. 82 seconds, 4 model calls, about $0.13.
What you're watching: me playtesting the file it wrote, untouched. Three cards added with the Add card button, one card dragged into Doing, the timer started. No errors in the console.
Open the app it built →focus-board.html · one file · press N for a new cardMost say it. This one ran the check as a real terminal command, and the output is in the session log.
You can read the exact command and the exact result. It's not a sentence, it's a receipt.
What you're watching: a real run through Hermes' web tools. I asked it to read xAI's announcement and reply with only JSON. It came back in 16.3 seconds with the date, both prices, seven benchmarks, a verbatim quote, and the availability list.
What you're watching: me playing the file it wrote, untouched. Enter to start, arrows to steer, W to boost. The score climbs through the gates, then I clip a car and the game-over screen comes up with the score and a restart prompt. No console errors.
Play the game it built →neon-drift.html · one file · Enter to start · arrows steer · W boost · S drift brakeWhat you're watching: the same broken Python function sent to two profiles. Grok 4.7 fixed it and proved it with a test in 13.9 seconds. My default profile, on GPT-6 Astra, did the same in 35.6 seconds. Both right.
This profile is one brain inside the Agent Operating System, the same setup I run a seven-figure business on. Join the AI Profit Boardroom and you walk out with:
No. The everyday 90% runs on a free local model on your own machine, free APIs slot in as profiles, and for frontier work it drives the CLIs you already pay for. It's a layer on top of what you own, not a new meter.
Inside the Boardroom there are full token-efficiency tutorials, so you learn to cut usage and stop thinking about it.
Members post their wins every day — agency owners, ecom founders, course creators, solo operators across 38 countries.
Read the 158-page wins doc →This page shows you one profile. The Boardroom gives you the whole operating system around it: every model as a profile, the dashboard, the memory, the agents that run overnight, and the people who've already hit your error message. Readers bookmark this page. Operators join, install the Agent OS this week, and have Grok 4.7 building for them by Friday. Decide which one you are tonight.
