The latest articles on DEV Community by Mathew (@mathew_thecoder). https://dev.to/mathew_thecoder https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3934531%2F2b3938b8-8984-4ea9-8474-2c261a28783c.png https://dev.to/mathew_thecoder en Mathew Sat, 16 May 2026 08:56:10 +0000 https://dev.to/mathew_thecoder/i-built-a-cli-that-builds-itself-and-it-outperformed-gemini-pro-using-a-free-model-1g44 https://dev.to/mathew_thecoder/i-built-a-cli-that-builds-itself-and-it-outperformed-gemini-pro-using-a-free-model-1g44 <p>I'm 14, from Romania, and over the past few weeks I built <strong>openmind</strong> — an AI-powered CLI that I used to build itself.</p> <p>Here's the part that got me: I tested the same free model (DeepSeek) on two setups. First on a normal AI coding tool — the results were bad. Then on openmind — in 2 prompts, the output was unrecognizable. Same model, completely different results.</p> <p>Then I asked it to build a landing page. The free model through openmind produced something better than what Gemini 3.5 Pro generated on its own. Clean design, correct branding, bilingual toggle, proper CTAs. Gemini Pro put a football stadium image on a weight loss site.</p> <h2> What openmind actually is </h2> <p>open mind is a that tool that I built using best stuff from opensource CLI's then improved it with:</p> <ul> <li> <strong>120+ tools</strong> including 13 dedicated Windows system tools (registry, services, network, processes, scheduler, security — nobody else has this)</li> <li> <strong>75+ LLM providers</strong> — bring any model, including local ones via Ollama</li> <li> <strong>1,300+ rules</strong> loaded on every response — this is why free models perform like frontier models</li> <li> <strong>Telegram integration</strong> — I control my PC from my phone, tested from 10km away, full CLI parity</li> <li> <strong>Enterprise security</strong> — AES-256-GCM audit logging, TOTP MFA, HMAC sessions, Docker sandbox, EU AI Act compliance</li> <li> <strong>Agent chaining</strong> — no other CLI tool has this</li> <li> <strong>Self-building</strong> — I used openmind to build openmind</li> </ul> <h2> The ruleset is the secret </h2> <p>Most AI coding tools are thin wrappers around a model. The model does all the work. If the model is weak, the output is weak.</p> <p>openmind wraps every model in 1,300+ rules that cover coding standards, security practices, output structure, reasoning patterns, when to ask for approval, how to handle uncertainty. The model almost doesn't matter anymore — the environment carries it.</p> <p>This means you can run a free or local model and get results that compete with paid frontier models on a huge class of tasks. I have the screenshots to prove it.</p> <h2> It builds itself </h2> <p>This week I asked openmind to add two cybersecurity tools — a network anomaly watcher and a file integrity monitor. It read the existing codebase, matched the exact TypeScript patterns, wrote production-quality code, and added itself to the project. I watched it happen in real time.</p> <p>The beta version did all of this. I'm still on a free model.</p> <h2> What's next </h2> <p>The full cyber defense platform — live threat detection for company servers, local and air-gapped, EU AI Act compliant from day one. No data leaves your infrastructure.</p> <p>For now openmind is a personal tool but I'm planning a public release. If you want to follow the build, let me know in the comments.</p> <p>Happy to answer anything — especially skeptical questions.</p> ai webdev productivity programming