--- title: 'Hermes + Grok: Three New Superpowers That Change the Workflow' summary: >- If you already pay for X Premium, you already have Grok. Connect it to Hermes with one OAuth login — no API key — and the agent reads X for you, runs browser tasks, and executes multi-skill playbooks from a single slash command. A tour of X Search, Browse.sh, and Skill Bundles. author: babyape113 authorUrl: 'https://x.com/babyape113' category: Integrations difficulty: Intermediate readingTime: 5 date: '2026-06-17' tags: - grok - x-search - browser-automation - skill-bundles - cron integrations: - Hermes Agent - Grok - Browse.sh - X - DeepSeek --- If you already pay for X Premium, you already have Grok. Connect it to Hermes with one OAuth login — no API key — and the agent reads X for you, runs browser tasks, and executes multi-skill playbooks from a single slash command. That's the news. Three drops landed this month: **X Search** (with video gen), **Browse.sh**, and **Skill Bundles**. The stack stopped being a research tool and started being a real agent. Here's what's running on my machine. ## Grok app vs. Grok in Hermes Grok in the app is fine until you close the tab. Then it forgets you exist. Hermes wraps Grok. Same model, same reasoning — but now it remembers. Context accumulates. The agent on day 30 isn't the agent on day 1, because it's been listening the whole time. That's the whole pitch. ## Just using X vs. X + Hermes | Task | Just X | X + Hermes | | --- | --- | --- | | Finding content | Scroll, hope | Agent surfaces on schedule | | Reading X articles | One tab at a time | Full text pulled and summarized | | Monitoring accounts | When you remember | Cron runs daily, dedupes | | Bookmarks | Graveyard | Nightly digest with full content | | Memory across days | Yours, if you're lucky | Cross-referenced against past reports | | Cost | Your time | ~$0.10/day | The X API can only pull a headline and a few lines of any X article. `x_search` reads the whole thing. That's the move. ## The stack ```text Hermes (orchestration) ├─ x_search → Grok 4.3 → X posts + full article content ├─ Browse.sh → hundreds of browser skills via @browserbase ├─ Skill Bundles → one slash command, one full playbook ├─ Video gen → text/image-to-video, up to 7 reference images └─ Base model → DeepSeek v4 (not Grok 4.3 — it dies on multi-turn) ``` Set Grok 4.3 as your base model and you'll learn why I lost two evenings. Don't. ## The three latest superpowers ### 01 · X Search Your X Premium sub = your Grok access = your agent's research feed. No API key. Any Grok tier works, including the X Premium you already pay for. One OAuth login, then cron monitors your account list, reads the whole article (not headlines), and DeepSeek prioritizes. ~$0.10/day. Nothing to argue with. > **Heads up:** `x_search` is OFF by default. After OAuth: `hermes tools` → CLI → toggle X Search on → restart. People miss this. ### 02 · Browse.sh Hundreds of browser skills via @browserbase — pull or contribute. Browser agents always broke the same way: each one rediscovered the web from scratch, ate the same captchas, failed the same logins. Skill reuse fixes it. Reliability comes from the catalog, not the model. The unlock: Hermes now finishes workflows on the internet. Fills the form, completes the checkout, monitors the dashboard until something fires. That's a different category of tool than "reads X for you." ### 03 · Skill Bundles One slash command, the whole playbook runs. Wrap skills that chain — each output feeds the next. The unlock here isn't new capability; it's compression. Once you've got 20 skills, orchestration cost dominates run cost. Bundles fix the orchestration tax. ## Video generation, in action Bundled with X Search is video gen — text/image-to-video with up to 7 reference images. ```text Prompt: generate a short video of a dragon fighting with an ape → 8 seconds. 720p. 93.5 seconds to render. ``` ## Four workflows I'm actually running **1. Daily brief.** Hermes runs in the background with my thesis and preferences loaded. Every morning: a brief on macro, geopolitics, tech, AI, and crypto. Each report feeds Hindsight, so the brief sharpens because it stops repeating what it already told me. Most people don't believe this until they see it. **2. Account tracker.** Four AI accounts I refuse to miss: @gregisenberg, @milesdeutscher, @AlexFinn, @JulianGoldieSEO. Cron runs daily. The algorithm doesn't get a vote. **3. Bookmark digest.** Cron pulls the last 24 hours of bookmarks, dedupes, `x_search` reads the full articles, and DeepSeek summarizes. My bookmarks stopped being a graveyard. **4. /post-maker.** My Skill Bundle for shipping content — one command runs four skills in sequence: ```text /post-maker write a post about why AI skill bundles are a game changer bundle loads: · concept-synthesis → pull the angle from notes + wiki · writing-plans → draft the structure, not just an outline · article-enrichment → add evidence, examples, sources · humanizer → strip AI patterns, sharpen voice ``` Before bundles: five separate commands, five separate prompts, context loss between each. Now: one line. The output comes out coherent because each skill saw what the last one did. That's the orchestration tax disappearing. **Why this works: linear composition.** Each skill's output feeds the next. Bad bundles fight each other (research + outbound + bug-fix in one shot — the agent picks the wrong path, output drifts, you lose precision). Good bundles chain. The rule: bundle what you run more than twice a week. Lower frequency than that, keep them separate. Bundling something you only do monthly is overhead disguised as productivity. ## The shift in month one You stop delegating the thinking. You form the hypothesis yourself and use the agent to test it. Hermes catches what you'd miss, compresses monitoring into a single brief, and remembers what it told you three weeks ago when fresh data lands. It can't decide what matters. That's still you. If you outsource the take, you don't have a take. ## What you're actually building X owns the data. Grok has access. Browserbase owns the browser layer. Hermes is the orchestration on top. If any of those layers changes terms tomorrow, your workflow goes dark. That's not pessimism — that's the stack. What you're building is **operational leverage, not a moat**. An agent that processes the real-time town square for $10/month plus $0.10/day. Useful. Cheap. Not yours. What's yours: your thesis, your judgment, your audience, the wiki where the thinking compounds. The agent is rented infrastructure. The take is the asset. Just know what you're renting. ## Run it yourself - Install: [hermes-agent.nousresearch.com](https://hermes-agent.nousresearch.com) - Browser skills: [browse.sh](https://browse.sh) Follow [@babyape113](https://x.com/babyape113) for more workflows from inside the stack. Stay curious. Stay humble. Invest responsibly.