ZenBrain memory as core
The full seven-layer architecture carries every conversation. Working memory for active tasks, episodic memory for history, procedural memory for learned routines.
Track C
„What does an AI platform look like that supports knowledge work — modular, with real memory, and reasoning steps organisations can inspect?"
ZenAI is the production research application of the ZenBrain architecture. It shows that the memory research holds up in practice — not only on a benchmark, but as a working tool.
Motivation
The production application is not a by-product of the basic research — it is the stress test. A memory architecture that only works on benchmarks has limited value. An architecture that supports day-to-day knowledge work proves itself on a different level.
ZenAI is therefore both: a production AI operating system for the enterprise — it consolidates data streams, generates insights and alerts automatically and carries day-to-day knowledge work — and the test vehicle for the ZenBrain research. What the papers describe has been running in production here since early 2026.
Methodological approach
Three core decisions shape the platform. First, ZenAI is modular — each functional unit (memory, tools, vision, code execution) can be used independently or replaced. Second, real memory instead of simulated persistence — the full seven-layer memory architecture carries every conversation. Third, transparency — tool use, reasoning, and memory access are visible to the user.
This is complemented by a process-atlas methodology for AI adoption in organisations: a systematic mapping of existing processes prior to the AI implementation, combined with a three-year stepwise adoption framework.
7-layer memory architecture, open-source core (Apache 2.0).
AI operating system for the enterprise — built on the ZenBrain core.
Core components
The full seven-layer architecture carries every conversation. Working memory for active tasks, episodic memory for history, procedural memory for learned routines.
Code execution (Python, Node.js, Bash, sandboxed), vision API, web search, GitHub, calendar, email intelligence, maps — each addressable as a separate module.
Systematic mapping of existing processes before AI implementation. In research — methods paper in preparation.
Stepwise adoption methodology for organisations — from the first pilot application to AI-augmented work throughout. In development.
Anthropic Claude (primary), Mistral (cloud fallback), Ollama (local inference). Researchers and authorities can choose the provider based on data-protection requirements.
The entire application layer can be fully self-hosted. LLM providers are configured separately; local inference is possible.
Current state
Get involved
Application studies with universities, pilot projects with research institutions, and methodological discussion on AI adoption in organisations are explicitly welcome. So is scientific leadership or accompanying research for real AI-adoption projects, including in the public sector — a reciprocal loop between theory and practice whose shared knowledge gain allows a resource-efficient arrangement.
research@zensation.aiMore from this research
Core and application fields, on a shared ethics foundation.