<p>I've been deep in the AI agent space for the last few months. Organizing meetups, testing frameworks, talking to builders. At some point I realized I was spending more time evaluating agents than actually using one that worked for me.</p><p>So I built Eva. She's a personal AI assistant that lives in Telegram. No app to download, no tab to keep open. Just a contact in a messaging app I already use every day, on every device.</p><p>Here's how I got there.</p><h3>Phase 1: the open framework route</h3><p>My first instinct was to go with open-source tooling. I started with <a href="https://github.com/openclaw" target="_blank">OpenClaw</a>, an agentic AI framework we've been exploring in the Porto community. The appeal was obvious: full flexibility, no vendor lock-in, community-driven.</p><p>And it delivered on flexibility. But flexibility without direction is just complexity. I found myself spending more time configuring the framework than getting actual work done. For someone building products and running a company, the setup cost was too high relative to the output.</p><p>OpenClaw taught me what I valued: modularity, the ability to add new capabilities on demand, and transparency into what the agent is doing. I carried those lessons forward.</p><h3>Phase 2: pre-built agents</h3><p>Next, I tried the opposite end of the spectrum. Off-the-shelf AI assistants. Claude, dedicated agent platforms, various "personal AI" products.</p><p>They work. For generic tasks, they work well. But the moment I needed something specific to my workflow, I hit walls:</p><ul><li>I wanted a daily briefing at 08:00 Lisbon time with my calendar, news, and market updates. Some platforms could do parts of this, but at high cost. If you wanted to send more than a few messages a day you quickly ran into a token wall within their subscription constraints.</li><li>I wanted the agent to remember context across days and devices. Most products either don't persist memory or do it poorly.</li><li>I wanted to add new tools without waiting for a product roadmap. No chance.</li></ul><p>The core problem: pre-built agents are designed for the average user. I'm not the average user. Neither are most founders, operators, or power users who actually need an AI assistant to do real work.</p><h3>Phase 3: building Eva</h3><p>The decision to build my own agent came down to a simple realization: <strong>it's actually cheaper and more contextualized to build exactly what you need.</strong></p><p>Not cheaper in the "I'll spend 200 hours and save $20/month" sense. Cheaper because:</p><p>1. <strong>No wasted features.</strong> Every capability Eva has exists because I needed it. No bloat.<br />2. <strong>Deep context.</strong> Eva knows my schedule, my tools, my preferences. She's not a general-purpose assistant trying to be everything to everyone.<br />3. <strong>Compound value.</strong> Every new tool I add makes the whole system more valuable. Pre-built agents don't compound the same way because you don't control the architecture.</p><h3>What Eva actually does</h3><p>Eva is a Telegram bot. Under the hood, she connects to any LLM via OpenRouter, stores conversation history in Supabase, and has a growing set of tools.</p><p><img src="https://storage.googleapis.com/taikai-storage/images/569379e0-39b7-11f1-9e83-1ba9aa6e7f04screenshots.png" style="width:949px" class="fr-fic fr-dib" /></p><h4><strong>The basics:</strong></h4><ul><li>Text, voice notes (with voice replies), images (OCR), and PDF processing. She's multimodal in and out.</li><li>Persistent memory per user. Context carries across days and devices.</li><li>Web search via Tavily. Google Calendar and other apps via Composio.</li><li>A daily 08:00 briefing delivered to my Telegram every morning.</li></ul><h4>The interesting part:</h4><ul><li>Eva can write her own new tools at runtime. I describe what I need, she generates the tool code, and there's a sandboxed approval flow before it goes live. Dynamic skills.</li><li>She can switch her own underlying LLM. If I want to test a new model, I tell her to switch.</li></ul><p>The stack is Python, deployed on Railway. The whole thing runs for a few dollars a month.</p><h4>Why Telegram?</h4><p>This was a deliberate choice. Most "AI assistants" are web apps you have to navigate to. That's a destination model. You go to the AI when you remember to.</p><p>Eva also works on WhatsApp, which matters if that's where your life happens. But Telegram has fewer constraints for bot development and richer API capabilities, so that's where I run her daily.</p><p>Telegram inverts the destination model. Eva is a contact in the app I already have open. She's ambient. When I need something, she's right there between my group chats and channels. No context switching, no new tab, no app to open.</p><p>Voice notes make this even more natural. I can talk to Eva while walking, driving, or cooking. She responds with voice. It feels like texting a very capable friend, not using a product.</p><h3>What I took from each phase</h3><table><tr><td><strong>Phase</strong></td><td><strong>What worked</strong></td><td><strong>What didn't</strong></td></tr><tr><td><strong>OpenClaw</strong></td><td>Modularity, transparency, community ethos</td><td>Setup cost too high for personal use</td></tr><tr><td><strong>Pre-built agents</strong></td><td>Polished UX, zero setup</td><td>No customization, no real memory, high monthly cost</td></tr><tr><td><strong>Eva (custom)</strong></td><td>Deep context, extensible, ambient interface</td><td>Requires building and maintaining it yourself</td></tr></table><p>The key insight: I didn't throw away what I learned. Eva uses the flexibility principles from OpenClaw (modular tools, transparent architecture) with the safety and reliability of managed integrations like Composio. Best of both worlds.</p><h3>Who should build their own agent?</h3><p>Not everyone. If Claude or ChatGPT handle your needs, use them. They're great products.</p><p>But if you find yourself repeatedly hitting the same limitations, if you need deep integration with your specific tools, if you want an assistant that gets better over time because you control the architecture, then building your own is worth considering.</p><p>The barrier is lower than you think. With Python, a Telegram bot token, an LLM API, and a database, you can have a working personal agent in a weekend. Making it great takes longer, but the foundation is surprisingly simple.</p><h3>What's next</h3><p>I'm dogfooding Eva daily. Every morning starts with her briefing. Every time I hit a limitation, I either add a tool or improve an existing one.</p><p>When the architecture stabilizes and I stop restructuring the core, I'll keep improving it in the open. The repo is live at <a href="https://github.com/hypermari0/eva" target="_blank">github.com/hypermari0/eva</a> if you want to check it out, fork it, or contribute.</p><p>If you're interested in this space, I talk about AI agents and personal AI at events like <a href="https://agentsday.org/" target="_blank">Agents Day</a> in Portugal. Come say hi.</p><p>────────</p><p><em>Mario Alves is CEO &amp; Co-founder of </em><a href="https://layerx.xyz" target="_blank"><em>LayerX</em></a><em>, an AI studio based in Porto, Portugal. He builds AI-powered products and occasionally writes about what he learns along the way.</em></p>

The story of building Eva, a personal AI assistant that lives where my conversations already happen.

Why I Stopped Using Other People's AI Agents and Built My Own

<p>Eight years ago, I started building LayerX with a simple belief: the best ideas do not always come from inside the building.</p><p>The companies with the hardest problems and the builders with the sharpest skills were living in completely different worlds. They rarely met. And when they did, the process was slow, expensive, and built for a different era. I wanted to fix that.</p><p>So we built TAIKAI, a platform to run hackathons and open innovation challenges. We connected organizations with builders. We ran programs for the European Commission, for startups, for enterprises trying to solve problems they could not solve alone. Over eight years, thousands of builders came through. Hundreds of challenges ran on the platform. We learned, slowly and sometimes painfully, what it actually takes to move ideas from concept to something real.</p><p>That work taught me one thing above everything else: the bottleneck is almost never the idea. It is the gap between thinking and building.</p><p>I just did not know yet how completely that gap was about to close.</p><h3>The wall I lived with for years.</h3><p>For most of my career as a non-technical founder, I had an idea and then I waited.</p><p>Waited for the technical team to have capacity. Waited for the sprint to open up. Waited for someone to tell me whether what I was imagining was even possible. I had the vision, the context, the business instinct. But the moment I needed something built, I hit a wall.</p><p>That was the tax of being a non-technical founder. I paid it for years without questioning it.</p><p>Then something shifted.</p><p><img src="https://storage.googleapis.com/taikai-storage/images/9a863190-2870-11f1-a311-d3fea339c98fAI-setup-3.png" style="width:932px" class="fr-fic fr-dib" /></p><p>A few months ago I <a href="https://blog.layerx.xyz/vibecodingceo" target="_blank">started building internal products</a> at LayerX using Claude Code. Not prototypes. Not wireframes to hand off to someone else. Actual working products, built by me, deployed and used by our team. A financial dashboard. An automation pipeline. A personal assistant.</p><p>I am not a developer. I do not write production code. But I shipped those products.</p><p>That experience broke something open in my thinking. The question I had been asking for years was completely wrong. The question was never "how do I get more engineers?" The question was always "how do I close the gap between thinking and building?"</p><p>AI closed it.</p><h3>And it was not just me.</h3><p>What happened to me as a founder happened to our technical team in a completely different way.</p><p>Our engineers did not just get faster. They became a different kind of team. Products that used to take months now take weeks. Features that would have required a specialist we did not have, they figure out and ship. The ceiling on what a small, focused team can deliver has moved so dramatically that I sometimes have to remind myself we are still the same company we were two years ago.</p><p>We are not. And that is exactly the point.</p><p>My wife tells me I have become obsessed. She is right. I see an inefficient process and I immediately start thinking about the agent that could fix it. I walk into a company and within ten minutes I am mentally mapping where intelligence should be sitting and is not. It has become the way I see the world.</p><p>Once you experience what it feels like when the wall comes down, you cannot stop seeing where the walls still are.</p><h3>What I saw next.</h3><p>Once I felt that freedom myself, I started seeing the same gap everywhere. In every company we worked with at LayerX. Across every industry.</p><p>Healthcare teams sitting on workflows that could be automated in a week. Finance teams manually reconciling data that AI could process in seconds. Retail businesses losing customers to friction that an intelligent system could eliminate overnight. Public institutions running procurement processes that belong in a different century.</p><p>The gap was not about awareness. Every one of these teams knew AI existed. They had read the same articles. Some had even run pilots.</p><p>The gap was about confidence. About knowing where to start. About having someone in the room who had actually done it, not just presented about it.</p><p>That is the gap LayerX is built to close.</p><h3>The applied layer is the real frontier.</h3><p>There is a version of AI progress that happens in research labs. Bigger models, better benchmarks, new architectures. That work matters and I have deep respect for it.</p><p>But there is another frontier that gets far less attention. The frontier of actually deploying intelligence inside a real business, with real constraints, real data, and real people who need to trust it before they will use it.</p><p>That frontier is harder than it looks. And it is where most companies are stuck right now.</p><p>At LayerX, we have made a decision to live on that frontier. Not to talk about AI transformation but to do it. To sit inside companies, understand how they actually run, and build intelligence into the parts that matter most.</p><p>Not because it is a good consulting opportunity. Because we have felt what it is like when it works. When a process that was eating 20 hours a week disappears. When a product becomes genuinely smarter. When a non-technical founder ships something real on a Tuesday afternoon and realizes the wall was never as solid as it looked.</p><h3>What I believe is coming.</h3><p>We are at the beginning of a decade where every company in the world gets rebuilt from the inside out. Not the ones that buy the right software licenses. The ones that develop the genuine capability to wire intelligence into how they operate.</p><p>The companies that move early will build an advantage that is almost impossible to catch. Not because of the tools they use but because of the institutional knowledge they accumulate. Every workflow they automate teaches them something. Every intelligent product they ship makes the next one faster. The compounding starts on day one.</p><p>Eight years ago I built LayerX to close the gap between builders and companies. The tools have changed completely. The mission has not.</p><p>The gap between what is possible and what most companies are actually doing has never been wider. And we have never been better equipped to close it.</p><p><em>If you are a founder, an operator, or a team that is ready to stop waiting, </em><a href="https://www.linkedin.com/in/hypermari0/" target="_blank"><em>I want to talk.</em></a></p>

From running hackathons across Europe to deploying AI inside companies, this is what eight years of building actually taught me.

Why I Stopped Using Other People's AI Agents and Built My Own

Why I Stopped Using Other People's AI Agents and Built My Own

Open innovation was the journey. AI was the destination.