CLARYEL

🚀 Bonus Tweet (3/2) — One More Thing! One of the most exciting parts of planning for the future of AI is testing and preparing for transitions between technologies before you actually need to make them. Switching subscriptions, changing entire AI agent ecosystems, moving between local and cloud services, or upgrading, modifying, and completely replacing hardware. The goal isn’t just to build an efficient AI system today. It’s to make sure you can easily adapt it to tomorrow’s breakthroughs. In a world where AI changes almost every day, flexibility is becoming one of our most valuable assets. 🔄

🔬 What I’m Researching — and What’s Coming Next. (2/2) My Three Main Areas of Research 🖥️ 1. Local AI Servers I’m experimenting with hardware starting at just $500, although I’d recommend around $1,000 for serious practical use. I currently have access to systems worth up to $15,000, plus rented, far more expensive GPUs for testing models that wouldn’t make economic sense to run on my own hardware. Raw model performance isn’t my main priority. Hardware and software are just tools. What matters is what these systems can actually accomplish for businesses and homes, how reliably they work, and when local computing makes more financial sense than the cloud. 🌐 2. Cloud Models and AI APIs From cheap APIs, including Chinese models, to frontier AI. I’m comparing capabilities, limitations, real-world costs, and integration with local systems. Subscriptions versus pay-as-you-go APIs are another important factor — the most powerful model isn’t always the smartest economic choice. 🤖 3. AI Agents Beyond their obvious benefits, I’m particularly interested in collaboration between independent multi-agent ecosystems. My Hermes and Grokbot systems are already working together — two separate ecosystems, each with its own agents and tools. 🏠 AI Meets the Physical World One of my biggest priorities is connecting AI to affordable smart homes, offices, servers, computers, and everyday devices. I’m exploring how autonomous, secure, reliable, and affordable these systems can become — especially for individuals and small businesses. 💻 AI Coding Is Moving Into the Background I’ll be focusing less on AI coding. Not because it matters less, but because it’s becoming just another tool. The goal isn’t to write code. It’s to get things done. I’m interested in complete, real-world workflows that AI can handle with minimal human involvement. 💰 Maximum Value, Reasonable Cost I’m not trying to prove that local or cloud AI is better. Sometimes the right choice is a $1,000 PC, a $10,000 server, rented GPU time, a cheap API, or a frontier-model subscription. What matters is the cost of the final result — not the tools used to achieve it. 🚀 What’s Coming Next My posts have mostly been scattered updates from individual experiments, sometimes rough and unstructured. I want to turn them into a coherent, connected picture of AI infrastructure — hardware, models, agents, costs, and real results. Detailed technical specifications and diagnostics will be collected automatically and shared separately, likely through links. I also plan to give you direct access to my AI agents. You’ll be able to ask them anything about my experiments, hardware, configurations, and results. They’ll answer using the collected data — and if they can’t, I’ll step in personally. My goal is to build an open, organized, and genuinely useful picture of what AI can accomplish at different budgets. I’m combining local and cloud AI, planning resources intelligently, and focusing on real results for businesses and homes. That’s what I’m working on — and what I’ll keep sharing with you. Don’t choose sides. Combine the best. 🚀 #LocalAI #CloudAI #AIAgents #AIInfrastructure

🚀 MixAI: I’m Choosing a Mix of Local and Cloud AI — Here’s Why. (1/2) Almost a year of hands-on experiments. Hardware ranging from $500 to $15,000. Cloud APIs, AI agents, and real-world tasks for businesses and homes. From now on, I’ll be sharing more about this combined approach to AI — one I believe has a very promising future. Things are changing almost every day. But AI technology has already reached a point where we can achieve reliable results and realistically plan how to use it over the next year or two. And I’ve come to a simple conclusion: we don’t need to choose between local and cloud AI. We need to learn how to get the best out of both. This goes far beyond simply connecting different models or orchestrating them. It’s about planning the entire system: which tasks should run locally, which should go to external models, how to distribute computing resources, which subscriptions make sense, and what hardware is actually worth buying. We also need to plan our physical infrastructure — computers, servers, GPUs, memory, power consumption, and future expansion. I believe local AI will become an essential part of our lives. It’s already making its way into smartphones, personal computers, and even experimental operating systems. ⚖️ Four Things That Matter Most 1. Real value. The number one priority is getting useful, practical results. 2. Cost. How much are we actually paying to complete a task — not just for hardware, subscriptions, or tokens? 3. Security. This deserves its own category, outside any ranking. Personal data, privacy, and control over our infrastructure require special attention. 4. Ease of use. How easily, quickly, and reliably can we get the job done? These are the criteria I use to compare technologies and build my own setup. ☁️ Subscriptions and APIs: Two Worlds Starting to Merge This is an interesting topic that often gets overlooked. Cloud AI can be accessed through subscriptions with usage limits or through APIs where you pay for what you actually use. Each approach has its own economics, limitations, and advantages. And they need to be evaluated separately. But the line between them is starting to blur. A recent example is OpenAI. Eligible ChatGPT Pro subscribers can receive $100 to $500 per month in API credits, depending on their subscription tier. These credits require separate activation, and API billing remains separate. That’s a pretty interesting example of two pricing models coming together. So I’m not just comparing models. I’m also exploring the economics of subscriptions and APIs, their limitations, and how to combine them efficiently. ➡️ Coming up in Part 2/2: The three areas I’m actively researching, my hardware experiments, AI agents working together, and how I plan to make all this information accessible to you. 🚀

We’re happy to share that Claryel has just been accepted into the OVHcloud Startup Program. 🚀 This is an important step for our project, and we’re excited to continue building with the support of the OVHcloud Startup Program. At Claryel, we are focused on practical AI solutions, local AI infrastructure, and making powerful technology more accessible and useful in real-world work. We’re grateful for this opportunity and looking forward to the next steps. #OVHcloud #StartupProgram #Claryel #AI #LocalAI #Startup

🚀 RTX 3060 12GB. A €200 GPU. Just 48GB of RAM! And this crazy guy @coldniko is preparing something absolutely INSANE for the next Strata release! 🤯🔥 The performance gains on this modest setup look unbelievable! 🚀💥 I’m genuinely blown away! Can’t wait for the release! 🤩🔥 #Strata #LocalAI

Level 2 — Local AI Server up to €3,000 I’m building a three-level range of local AI servers. Level 2 is a modern base system with a target price of up to €3,000, designed with broad possibilities for expanding both computing power and functionality. It is a starting point for those who want new hardware for deploying local AI without a large initial investment, while retaining significant potential for future expansion. 👍 Hardware 👉 Processor: Intel Core Ultra 7 270K Plus — 24 cores / 24 threads, up to 5.5 GHz 👉 Motherboard: ASUS Pro Q870M-C — modern Intel Q870 platform with multiple PCIe slots and Intel vPro support 👉 Memory: 64 GB DDR5 — 2 × 32 GB, with two additional slots for future expansion 👉 GPU: AMD Radeon RX 9070 — 16 GB VRAM 👉 NVMe: Lexar NM1090 PRO 2TB — PCIe 5.0 x4, up to 14,000 MB/s read and 13,000 MB/s write, with professional cooling, a large heatsink and multiple fans 👉 CPU cooling: Efficient air cooling 👉 Case: New, modern design with a ventilation system designed for high loads 👉 Power supply: ASUS Prime 750W 80+ Gold 👉 Case airflow: Three intake fans providing airflow across the entire system, plus one exhaust fan 👉 Wi-Fi: Intel Wi-Fi 7 BE200 + Bluetooth 👉 Intel vPro: Supported by the platform 👉 Recovery USB: Compact 64 GB USB flash drive that does not protrude from the case, containing tools for system recovery, OS installation/reinstallation, software-stack recovery and hardware testing. All components in this system are new. The configuration is based on currently available components and is designed for future scalability. ✌️ Why Level 2? The goal is to provide a modern local AI system without a large initial investment, while leaving significant room for future expansion. The system can be progressively upgraded with more memory, additional storage, a newer GPU or other components as requirements grow. 👀 Software and Performance The software stack is currently being prepared. The main workload will be AI models running on Strata by @coldniko. Performance is still being tested. Real-world results will determine the recommended models and minimum settings for Level 2. More about the software and functionality will be covered in a separate post. 😎 Optional Upgrades The system can be expanded at the time of purchase or later with additional internal and external storage, archiving and backup solutions, additional memory, GPU upgrades and other hardware or software extensions. 🤝 Warranty and Technical Support Warranty support will combine manufacturer warranties with our own warranty, backed by a technical and financial reserve of tested spare components. Some components may have manufacturer warranties of up to 5 years, depending on the specific component and supplier terms. Technical support will be built around voice agents, chat agents and a website communication agent. They will advise users, diagnose problems and, with permission, connect remotely to the computer. Agents will operate around the clock, including weekends. If an agent cannot resolve a problem, or the client says the issue remains unresolved and is not satisfied with the interaction, the ticket will be transferred to a human specialist during working hours. Remote and hardware-level management are included in Level 2. At the user’s request, this can include BIOS access, system recovery or reinstallation, and other functions supported by the configuration. All remote access is performed only with the user’s permission and without access to the client’s personal data. 🚀 Upgrades and Development Level 2 is a starting point for those who want new hardware for deploying local AI with broad possibilities for expansion. The system can be progressively upgraded or completely replaced with a Level 3 system. The option of using Level 2 and Level 3 systems together will be described in the Level 3 post.

Grok Voice 2.0 already impressed me in real-world tests. Now 3.0 is here — time to test it properly. 🚀🎙️

@coldniko releases, Strata gets Mac support — and suddenly, my plans are getting bigger! The plan is to build three different brains for my local AI agent, all running Strata and the same model: Qwen 3.8 Next 3_S. Level 1 — €1,000: DDR4-based system. Level 2 — €3,000: DDR5-based system. Level 3 — €7,000: MacBook Pro M5 Max with 128 GB of unified memory. Same model. Three different hardware levels. I’ll start by installing the same model on all three systems and putting Hermes to work on real-world tasks. One day on the first machine, the next on the second, and the third on the Mac — or perhaps half a day on each. The goal? Compare how the same local AI agent performs across three hardware tiers under real workloads. Looks like I’m going to have a busy time ahead of me!

Bravissimo 👏

Hardware and system setup: done for today. Time to move forward — slowly, stubbornly, and LOCALLY. 😂 These are the two machines I’ll be working with for now. The hardware is only part of the story; the full software stack and system configuration are considerably more involved. I’ll share more details as these systems start delivering real, practical value. The goal? Maximum remote control, voice commands from my phone, and as little time as possible glued to a screen or keyboard. Slowly but surely. No cloud dependency required. Let’s make local AI earn its electricity bill. 😎

I’m taking another step toward large-scale testing and more detailed reports. I’ve already set up 20–30 computers almost from scratch, then handed control to ChatGPT or Grok Bot and managed them entirely by voice from my phone. I’ve used this approach for real tasks, not just benchmarks. One example: I automated an accounting workflow — from installing an ERP and accounting software to retrieving Amazon invoices, organizing them, and entering the relevant invoices into Italy’s tax system. I barely touched the software interfaces myself; I gave instructions by voice and checked the final results. Now I’m trying to reproduce real workflows like these using local AI. For now, I’m testing two machines with Hermes and Strata, almost entirely out of the box. The first trade-off is obvious: tasks that take seconds or a minute in the cloud can take 5–10 minutes locally. I haven’t optimized anything yet. That’s the point: first, see what works out of the box. Then decide whether the agents can optimize themselves — or whether I need to step in.

Scaling up: I’ve bought 10 identical motherboards and CPUs to build 10 matching systems. Why this platform? — ASUS — my favorite brand. — Cheap DDR4 RAM… well, “cheap” 😂 — 8 RAM slots. — BIOS-level remote management — agents can even reinstall the OS remotely. The goal: turn individual test results into a repeatable setup. Less guesswork, more data. P.S. I’m betting on @coldniko Strata. Time to see how far we can push it.

I built this over the summer for internal use. It’s much more informative and more capable in many areas, but I’m not maintaining it for now — I have far more pressing tasks at the moment. investors.claryel.space/syst…