The latest articles on DEV Community by Max Velloc (@maxbuilds). https://dev.to/maxbuilds https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4088821%2Fc4cf78aa-a125-4bfc-b204-9531be05e7b6.png https://dev.to/maxbuilds en Max Velloc Fri, 21 Aug 2026 20:31:53 +0000 https://dev.to/maxbuilds/i-built-velloc-code-an-open-source-desktop-coding-agent-with-multi-model-workflows-39k0 https://dev.to/maxbuilds/i-built-velloc-code-an-open-source-desktop-coding-agent-with-multi-model-workflows-39k0 <p>I've been using coding agents like Claude Code, Codex, and OpenCode for a while now. They've genuinely changed how I work — but almost all of them live in the terminal.</p> <p>I really like the openness and model independence of tools like OpenCode. What I wanted was that same flexibility inside something that feels like a real desktop application: install it, add an API key, open a project, and start working.</p> <p>So I built <a href="https://velloc.app" rel="noopener noreferrer">Velloc Code</a> — an open-source, Windows-first desktop coding agent.</p> <h2> Why a desktop UI? </h2> <p>Terminal/TUI agents are great, and I'm not trying to replace them. The argument for a desktop app is simple:</p> <ul> <li> <strong>No environment setup</strong> — install it, add an API key, open a project folder. No shell config, no dependency bootstrapping.</li> <li> <strong>A UI you can reason about</strong> — sessions, project memory, and agent configuration are visible and inspectable, not hidden in config files.</li> <li> <strong>One app for your whole workflow</strong> — reading code, running agents, reviewing output, and checking what the agent actually did all live in the same window.</li> </ul> <p>The philosophy is: <em>make the default experience easy, and don't lock anyone in.</em></p> <h2> Multi-model workflows </h2> <p>The core idea is that one model doesn't have to do everything.</p> <p>You can start with a single model and a single agent. When you outgrow that, you can assign different models to different agents or tasks:</p> <ul> <li>a cheaper/faster model for research and planning</li> <li>a stronger model for implementation</li> <li>another model for review and testing</li> </ul> <p>The same project can split a larger task across agents, each using the model that fits its role — without changing your workflow or your provider.</p> <p>It's provider-agnostic: bring your own providers or custom model endpoints, and switch models without touching your setup.</p> <h2> What else it does </h2> <ul> <li> <strong>Project memory across sessions</strong> — context survives, so you're not re-explaining your project every time.</li> <li> <strong>Extensible</strong> — skills and Claude Code-compatible plugins.</li> <li> <strong>Inspectable</strong> — you can inspect and customize the agent itself, instead of treating it as a black box.</li> </ul> <h2> The direction </h2> <p>The openness and model independence of open-source coding agents, with a polished desktop UI and a much easier starting point.</p> <p>It's Windows-first right now and still early.</p> <p>I'd really value feedback from people already using Claude Code, Codex, OpenCode, Cursor, or similar:</p> <ul> <li>Would you actually prefer a desktop UI, or is terminal/TUI still the ideal interface?</li> <li>Would you use different models for different agents or tasks?</li> <li>What would make you switch from your current coding agent?</li> <li>What feels unnecessary or over-engineered?</li> </ul> <p>Check it out: <a href="https://velloc.app" rel="noopener noreferrer">velloc.app</a> — and the source is on GitHub.</p> <p>I'm the developer, so feel free to be critical — that's mainly why I'm posting it here.</p> showdev opensource ai tooling