Why I'm building Zdravo: a governed context layer so AI agents remember, and so you can prove what they knew when they acted.
Every AI agent you run — Claude, GPT, an internal Ollama model — starts every session from zero. It re-derives context your team already established. It can't tell you why a decision was made last month, or which memory it used to make one today.
That's not a UX problem. It's an operations and governance problem.
Enterprises can't scale AI agents on stateless models with no audit trail — not because the models aren't capable, but because nobody can prove what the agent knew when it acted.
This isn't a model problem. It's an infrastructure gap. We built shared memory layers for people — Notion, Confluence, Git — but nothing equivalent for the AI agents now doing real organizational work.
The data layer should be the product, not the model. Frontier models will keep changing. Your organization's memory, decisions, and audit trail shouldn't have to be rebuilt every time one does.
Zdravo is that layer: model-agnostic, MCP-native, governed by default.
"Zdravo" means "health" in Slavic languages — in Balkan cultures, both a greeting and a wish for wellbeing.
It's a fitting name for infrastructure meant to keep AI systems grounded: healthy context in, auditable decisions out.
Give every AI agent — regardless of vendor or model — persistent, governed context and a provable audit trail.
Not another wrapper on top of one model. Infrastructure underneath all of them.
Our visual language draws from Swiss design, Balkan folk patterns, and the brutalist web movement — bold borders, flat colors, maximalist typography.
Governed infrastructure doesn't have to look sterile. Structure should be visible, not hidden behind soft gradients.
Watching AI agents rebuild the same organizational context from scratch, every session, revealed the actual gap: not memory as nostalgia, but memory as infrastructure.
Started with a simple question: what if an AI agent's context — and the audit trail behind it — persisted and was governed, no matter which model was running?
Zdravo runs as a governed context layer with a live MCP server, REST API, and audit log — in active development, adding enterprise governance features.
MCP support is just the entry point. Next: a policy engine and approval gates for agent actions, compliance mapping (SOC 2, ISO 42001, NIST AI RMF), and decision receipts your auditor can actually read.
The core question: how do enterprises run AI agents they can actually govern?
Built from Lake Ohrid, Albania — infrastructure for the agents doing the work, and the audit trail for everyone who has to trust them.
Shqiprim Balazoski, Founder