惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

D
Docker
F
Fortinet All Blogs
爱范儿
爱范儿
博客园 - Franky
MyScale Blog
MyScale Blog
罗磊的独立博客
博客园 - 三生石上(FineUI控件)
B
Blog
P
Proofpoint News Feed
IT之家
IT之家
宝玉的分享
宝玉的分享
D
DataBreaches.Net
S
SegmentFault 最新的问题
Microsoft Azure Blog
Microsoft Azure Blog
GbyAI
GbyAI
M
MIT News - Artificial intelligence
L
LangChain Blog
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
月光博客
月光博客
雷峰网
雷峰网
Stack Overflow Blog
Stack Overflow Blog
量子位
V
V2EX
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
Gemma 4 and Pressable MCP: safe AI hosting in Linux terminal
Aribu js · 2026-05-13 · via DEV Community

Transparency: this article contains affiliate links. If you purchase hosting through my link, I will receive a small commission at no additional cost to you.

This is a submission for the Gemma 4 Challenge: Write About Gemma 4.


The Problem: My Tokens Are Flying to Google Cloud

A few weeks ago, I shared how I set up Gemini CLI for managing Pressable servers directly from the Fedora terminal: no browser, no extra clicks - just commands, and the server infrastructure obeys.

But the euphoria quickly turned into a question: "Where are my confidential tokens being processed?"

The access key to the production server was flying to Google Cloud. Logs, PHP versions, site structure - everything was passing through someone else's data center. It's like giving your apartment keys to your neighbors: they might be decent people, but it's more peaceful when the keys are only with you. 😎

Solution: I switched to Gemma 4 - an open model from Google with an Apache 2.0 license, which gives full control over the data. And I found a hybrid approach for those whose laptop physically can't handle 32 GB of RAM: I run Gemma 4 through Google AI Studio, getting all the power of the model without expensive hardware.


Gemini CLI vs Gemma 4: Comparison for DevOps

Before diving into the terminal, let's put the dots over the "i". Here are the key criteria that influenced my choice:

Criterion Gemini (Cloud) Gemma 4 (API / Local)
Confidentiality Data is transmitted to Google servers 😟 Data does not leave the perimeter. Even through the Google AI Studio API, it is not used for training 🛡️
License Proprietary 🔒 Apache 2.0 - free commercial use, modification, distribution 🔓
Context Window Up to 1 million tokens 🚀 128K tokens - enough for analyzing large server logs ✅
Advanced Thinking Yes 💡 thinking: true - models 26B and 31B have a thinking mode, ideal for diagnostics 🧠
Cost Paid after limits are exhausted 💸 Free tier in AI Studio 🆓
Autonomy Requires internet ☁️ Can work fully offline on your own server 🏠
Customization Limited ⛔ Full freedom: fine-tuning, running on your own hardware ⚙️

For managing WordPress hosting through Pressable MCP, I chose Gemma 4 31B (Dense) - the most powerful model in the lineup with thinking: true, capable of analyzing complex API responses and making informed decisions.

We also recommend reading: AI WordPress Hosting Management via Linux CLI


Step 1: Creating an API Key for Gemma 4 in Google AI Studio

Any path begins with the first step. We will receive the access key not in Google Cloud, but in Google AI Studio - this is free and does not require a bank card.

  1. Go to aistudio.google.com
  2. In the left menu, select API Keys
  3. Click Create API key and select a project (or create a new one, for example, Pressable Gemma 4 Agent)
  4. Copy the generated key

⚠️ IMPORTANT: the key will be shown only once. Save it immediately.

Screenshot of Google AI Studio with the selection of the Pressable Gemma 4 Agent project

Now, export the key to an environment variable in the Fedora terminal:

export GEMINI_API_KEY="AIzaSy...your_new_key"

Enter fullscreen mode Exit fullscreen mode


Step 2: First Run of Gemma 4 in the Linux Terminal

The most interesting moment - checking if the model responds. We use gemini-cli, explicitly specifying the model through the --model parameter:

gemini --model models/gemma-4-31b-it -p "Say hello from Gemma 4 31B"

Enter fullscreen mode Exit fullscreen mode

Response:

Hello! Gemma 4 31B here. How can I help you today?

Enter fullscreen mode Exit fullscreen mode

The model works! 🤖

Terminal with a successful first run of Gemma 4 31B


Step 3: Connecting Pressable MCP to Gemma 4

Now, the main thing: connect Gemma 4 to Pressable MCP, so that the model can directly manage hosting through the Model Context Protocol.

3.1. Removing the Old Configuration

gemini mcp remove pressable

Enter fullscreen mode Exit fullscreen mode

3.2. Adding the Pressable MCP Server

gemini mcp add pressable https://mcp.pressable.com

Enter fullscreen mode Exit fullscreen mode

3.3. Specifying the Authorization Token

Open the configuration file:

nano /mnt/DATA/pressable-mcp-project/.gemini/settings.json

Enter fullscreen mode Exit fullscreen mode

Add the Pressable Access Token to the headers section:

{
  "mcpServers": {
    "pressable": {
      "url": "https://mcp.pressable.com",
      "headers": {
        "Authorization": "Bearer YOUR_PRESSABLE_ACCESS_TOKEN"
      }
    }
  }
}

Enter fullscreen mode Exit fullscreen mode

3.4. Checking the Connection Status

gemini mcp list

Enter fullscreen mode Exit fullscreen mode

We see the desired green status - Connected! 🟢

Successful connection of Pressable MCP to Gemma 4 in the terminal


Step 4: Gemma 4 as an AI Administrator of the Server

Launch an interactive chat with Gemma 4:

gemini chat -m models/gemma-4-31b-it

Enter fullscreen mode Exit fullscreen mode

Give a combat task (in English - this is how the API works most accurately):

> Please list all the sites on my Pressable account
  and tell me the PHP version of shchoiyak-sandbox.

Enter fullscreen mode Exit fullscreen mode

Here, the magic of the Model Context Protocol happens: Gemma 4 understands that it needs the search_sites tool and requests permission to execute it. Choose Allow once or Allow all server tools for this session.

Gemma 4 requests permission to use the search_sites tool through Pressable MCP

The model receives raw JSON from the Pressable API, analyzes it thanks to the thinking: true mode, and provides a clear response:

- Domain:     shchoiyak-sandbox.mystagingwebsite.com
- PHP Version: 8.5
- State:       live
- Datacenter:  BUR (Burbank, CA)

Enter fullscreen mode Exit fullscreen mode

Gemma 4's response with detailed information about the site on Pressable

Result: without any browser and login to the control panel - all critical information is right in the terminal.


Conclusion: Why Gemma 4 for DevOps is the Right Choice

Switching from Gemini CLI to Gemma 4 + Pressable MCP is not just a model replacement. It's a fundamental change in approach to security and control over data:

  • ✅ Confidential tokens do not leave your perimeter
  • ✅ Apache 2.0 license - complete freedom for commercial projects
  • thinking: true mode for deep analysis of server logs
  • ✅ Free access through Google AI Studio
  • ✅ Ability to work fully offline on your own server

Yes, running Gemma 4 31B locally requires a powerful server. But even through Google AI Studio, it's free, secure, and preserves the main thing - openness and independence.


Try Pressable for Your WordPress Projects

Want to manage servers like a real AI-ninja? Refuse routine clicking and switch to infrastructure created by the same people who write the WordPress core.

👉 Try AI management from Pressable today

And if you need fast and stylish creatives for your new site's social media - use the free graphic editor SMM Turbo.