VDM AI OS — my system for working with AI
I built an AI system for myself that holds the context of all my projects: rules, tools, agent skills, the current state and the history of decisions. Any AI agent opens the area it needs and continues where it left off, with no explaining from scratch. I use it every day: projects, clients, this website, sport and health.
The task
AI forgets everything between chats. Every conversation starts from zero, decisions and the reasons behind them get lost, and the context has to be retold again and again.
My role
I designed the memory structure and the rules agents follow to read and update it, connected tools and skills, and built a visual interface for browsing the system.
Memory that doesn’t reset
Every area has rules (how to work here), a current state (what’s happening now) and a history (what was decided and why). The agent reads them at the start and adds to them after important decisions on its own, without being told to “save”.
- Rules
- Current state
- Recorded decisions
One system, many areas
Projects and clients, tools, sport and health live in one system and are linked to each other. For a specific task the agent takes only the area it needs from the link map instead of loading everything.
- Link map
- The right area
- Work with context
Not tied to one model
All the memory is plain text files. Agents from different companies work with them: Claude, ChatGPT and others. AI models are updated every few months, but the accumulated context doesn’t disappear when you switch a service or a subscription.
How I use it — and what it means for a business
Clients
I go through client replies with the agent, draw conclusions and improve my approach for the next conversations.
Analysis of enquiries and deals: why clients leave, what works, what the next steps are.
Products
The agent builds products knowing the whole situation: goals, past decisions, constraints. It isn’t a single prompt from scratch.
Projects where AI knows the client’s history, agreements and company standards.
Decisions
The system remembers why a decision was made, even months later.
A history of decisions and procedures instead of “why did we do it this way?”.
Sport
Progress analysis and programme discussions: what works, what doesn’t, nutrition. The charts are a nice bonus.
Regular metrics: the system tracks the trend and suggests what to change.
Health
The agent keeps the full context: past cases, test results, observations. So the answer takes the whole picture into account instead of giving generic advice.
A knowledge base: answers that draw on the company’s whole history, not a generic template.
Who it’s for
For owners and teams whose knowledge lives in people’s heads, chats and a dozen services, and who have to explain everything to AI all over again every time.
How it works
For me: memory and rules in plain files, agents in Claude, ChatGPT and other models, and a local visual interface. The data stays with its owner.
Tools
- Markdown memory
- AGENTS.md rules
- Node.js
- Toast UI Editor
- Document version history
- PowerShell
- PWA
- aTracker API
For you — whatever suits your team
- For those who already work with AI tools — the same way I do.
- For those who don’t want to deal with it — the system runs on a server and you talk to it through a Telegram bot: you write as you would to a colleague, while memory, rules and tools work under the hood.
Want AI to remember your team’s context?
We’ll look at where your knowledge and decisions live and which area to start with.