Self-Hosted Deep Learning Deployment Platform

# Local AI Development Platform – Project Description ## 1. Project Goal I want to build a powerful, locally operated AI development and system administration platform running on my own Linux computer. This is not intended to be just a local LLM, chatbot, or basic OpenClaw installation. The goal is a modular AI-agent system capable of researching, programming, developing WordPress solutions, operating computers, managing remote servers, and continuously expanding its knowledge. The primary use case is professional **WordPress and Gutenberg development**. ## 2. Hardware The system will run on an **Elite ER937 AI Silent Cooling Mini PC** with: - Linux - AMD Ryzen 9 processor - 32 GB RAM The architecture should be optimized for this hardware. If RAM, storage, GPU/NPU, or other upgrades are recommended, they should be specified. ## 3. Internet Research and Vision The AI must have real internet access and be able to: - search the web independently; - open and analyse websites; - follow links and research multiple sources; - use technical documentation, GitHub repositories, forums, and developer resources; - download and process documentation; - search for images; - analyse images and screenshots; - visually analyse websites. It should actively research current information instead of relying exclusively on knowledge contained in the language model. ## 4. Programming The AI should function as a practical software-development agent and support at least: PHP, JavaScript/TypeScript, HTML/CSS, Python, SQL, Bash, REST APIs, Git/GitHub, Docker and Linux. It should be able to: - inspect and understand existing projects; - create and modify files; - write and modify code; - install dependencies; - execute commands and builds; - test its own work; - analyse logs and errors; - debug problems; - correct its implementation and retest it. The goal is not simply to generate code in chat. The AI must be able to work directly in the development environment and execute tasks. ## 5. WordPress and Gutenberg WordPress development is one of the main purposes of the system. The AI should work with: - WordPress administration; - WP-CLI; - WordPress REST API; - filesystem and database; - plugin development; - theme and block-theme development; - Gutenberg and Block API; - custom Gutenberg blocks; - block.json and theme.json; - PHP and JavaScript/React; - CSS and responsive frontend development; - debugging, installation and testing. For example, if asked to create a Gutenberg component, it should be able to inspect the existing project, research current documentation when necessary, design the implementation, create or modify files, write the code, install it, test it, identify errors, correct them and verify that the final result works. ## 6. Computer and Server Control The AI must be able to operate the local Linux computer and remote servers within defined permissions. Required capabilities include: - Linux terminal and shell; - SSH connections; - remote server administration; - filesystem management; - software and package installation; - Docker management; - service configuration; - configuration-file editing; - process and log management; - troubleshooting; - Git; - backup and restoration operations. The AI should not only suggest commands. Within its permissions, it should be able to execute them itself. ## 7. Graphical Computer Use Terminal and API access are not enough. The AI should also be capable of visual computer interaction: - understand screen content; - operate a browser; - click and type; - navigate websites; - operate web administration interfaces; - visually control WordPress administration; - use other graphical applications when required. The system therefore needs both machine-level access through APIs/terminal/SSH and visual computer-control capabilities. ## 8. Persistent Memory and Knowledge Base The AI should have persistent long-term memory and should not start from zero in every session. It should retain and reuse information about: - my projects; - project architecture; - source code; - coding conventions; - development rules; - server configurations; - previous tasks and solved problems; - technical decisions; - documentation and workflows. A searchable knowledge/RAG system should contain WordPress/Gutenberg documentation, project code, technical manuals, research results and other relevant information. The knowledge base should be updateable without retraining the complete LLM. ## 9. Learning Mode A major requirement is a dedicated **Learning Mode**. When I am not actively using the AI for work, I want to be able to assign a learning objective such as: “Study the current Gutenberg Block API documentation in detail.” The system should then autonomously: 1. research the subject; 2. find reliable sources; 3. download and process documentation; 4. compare and verify information; 5. structure the acquired knowledge; 6. store useful knowledge; 7. test its understanding with questions or practical tasks; 8. identify knowledge gaps; 9. research missing areas; 10. update existing knowledge. Background learning should automatically adapt to available system resources and reduce or stop when the computer is needed for productive work. Continuous learning should primarily use memory, RAG, structured knowledge and evaluation rather than constantly retraining the base LLM. Fine-tuning or LoRA can be added later where useful. ## 10. Multi-Agent Architecture The platform should preferably support specialized cooperating agents, for example: - Research Agent - Coding Agent - WordPress/Gutenberg Agent - Server Administration Agent - Computer Use Agent - Learning Agent - Testing/QA Agent - Security Agent - Memory/Knowledge Agent A typical workflow could be: **Request → Research → Planning → Coding → Execution → Testing → Error Correction → Verification** Individual models, agents and tools should remain replaceable. ## 11. Local and External Models Core infrastructure and private information should remain