The Private Sovereign AI Workstation — Why I built Atlarix differently

Historical — the architecture below was removed from Atlarix in v14.9.0 (10 July 2026).

Blueprint, the parser, the SQLite graph, BM25 node scoring and the file watcher were all deleted — roughly 5,246 lines — after the watcher was found to hold one file descriptor per watched file and reach about 12,365 on a large repository.

Atlarix today has no index of any kind. Retrieval is purely lexical: bundled ripgrep (grep and glob), no embeddings, no vector store, no graph, nothing in the background. Current architecture: atlarix.dev/docs.

---

There's an assumption baked into every major AI coding tool right now.

That you have a $20/month subscription to an American AI company. That you have a fast, stable connection. That the models worth using are the ones coming out of San Francisco.

I'm building software from Nairobi. That assumption doesn't hold.

So I built Atlarix differently — and after a year of building, v5.1.0 is the version that finally says out loud what it was always trying to be: The Private Sovereign AI IDE (the name did not last — Atlarix is explicitly not an IDE; it runs beside your editor, and never replaced it).

What that actually means

Sovereign doesn't mean offline-only or anti-cloud. It means you decide.

You decide which model runs your code.

You decide what context gets sent.

You decide whether your keys live on your machine or in a provider's system.

Atlarix is built around three ideas that most AI coding tools don't share:

1. Blueprint — your codebase as a living graph

Instead of sending files per query, Atlarix parsed your codebase into a persistent node/edge graph stored locally in SQLite, and that graph became the AI's long-term memory. (All of it was removed in v14.9.0. Project memory survives as plain notes on disk; the graph does not.)

2. Any model — genuinely

Cloud providers via BYOK. Local models via Ollama and LM Studio. And now Compass — three built-in model tiers (Fast, Balanced, Thinking) that work the moment you open the app. No API key. No configuration. Just open and build.

3. The tool carries the intelligence

This is the core thesis. Atlarix feeds the model exactly what it needs, when it needs it, scoped by your .atlarixignore and Blueprint context tiers. A lighter model with good context beats a frontier model flying blind. That's not a consolation prize — that's the architecture.

Work modes that mean something

• Ask — read-only. No code changes, no terminal. Research, explore, understand.
• Plan — map out what needs to happen before anything gets written.
• Build — full tools, explicit approval queue before execution.
• Debug — focused on what's broken and why.
• Review — read your code the way a reviewer would.

Each mode binds to a Blueprint context tier. The AI isn't just switching personality — it's switching what it knows and what it's allowed to do.

Why sovereignty matters right now

AI coding tools are becoming infrastructure. The question of which model runs your code, who sees your prompts, and whether you can switch providers without rebuilding your workflow — these are not small questions.

Atlarix's answer: your architecture, your model, your sovereignty.

v5.1.0 is live. Free to start. Native desktop — Mac, Windows, Linux.

👉 atlarix.dev

I built this from Nairobi. If you try it, tell me honestly what's missing. That feedback is worth more than any press right now.