Infrastructure that understands itself.

A suite of small, sharp tools that give your self-hosted stack a memory of its own normal — running entirely on your hardware, with every claim one click from proof.

Explore the tools → See the code
4 packages shipped ◇ Apache-2.0 & MIT Verified by running, not asserting
The suite

Five tools, one bet

The value lives in the substrate, not the model that happens to be running. Each tool ships and stands on its own.

🗂 folder-nature → Your folders, self-aware. A tiny convention that tags each directory with machine-readable context, so config travels with the service — not the admin. PyPI v0.2.1Apache-2.0120 tests 🧭 advisor → Monitoring that refuses to cry wolf. Learns your system's normal and surfaces real drift — with reliability measured in the open, n disclosed. aura-pceApache-2.0runs live 🧪 whypass → Catch the claims that outrun their evidence. A claim-discipline linter for text — the same one that checks this very site. PyPI v0.1.0Apache-2.0 🔗 copresence → Memory for your event streams. Link moments by what was there together, not just what they resemble. PyPI v0.1.0Apache-2.0 🧵 aura-router → Continuity for AI agents. Re-lights an agent's working thread every time a fresh instance wakes — it picks up your sentence mid-stride. GitHubMIT 🧠 aura-mnemos → Memory that refuses to lie about being down. A persistent-memory MCP server for AI agents — when the store dies, health says so instead of returning silence. PyPI v0.1.0Apache-2.022 tests
The gap

You find out something's wrong the same way every time: a user complains.

You run your own stack — a Mastodon instance, a Nextcloud, a mail server, a few things on a Pi in the closet. You react after the fact. Scaling means manual tuning. Moving hosts means rebuilding config from scratch.

Enterprise tools exist — Datadog, New Relic — but they're expensive, centralized, and overkill for one VPS. Nobody builds lightweight, predictive operational intelligence for small self-hosted deployments. That's the gap AURA fills.

Meet the advisor →
How it works

Observe · learn · warn — on your hardware

AURA reads your infrastructure's own signals, learns what normal looks like for your system, and warns you when something drifts — before your users notice. Every learned pattern is inspectable. Nothing leaves your hardware.

AURA architecture: infrastructure is tagged by folder-nature and observed via telemetry; the advisor learns each system's normal and writes a transparent log; the operator is warned before users notice — all on your own hardware.
Why it's different

Auditable, not "trust me"

🔒Local-firstZero outbound network calls. No SaaS, no third-party telemetry. Run it offline and it works exactly the same.
🔍Open & inspectablePermissively licensed (Apache-2.0 & MIT). Read every line. The privacy isn't a promise — it's verifiable.
🪶LightweightBuilt for single-board and ARM hardware. Scales horizontally across a mesh.
📦Ships cleanYou receive the learning mechanism — never someone else's accumulated data. Your patterns stay yours.
The differentiator

Every number on this site carries its source.

When we quote reliability, we disclose the sample size. When we say "shipped," it's on PyPI and you can install it. When a feature is still research, we name it as research instead of shipping the word without the work. That honesty isn't a footnote — it's the one thing a competitor can't fake.

Start where it's easiest

Install a tool in one line, or read exactly what's proven and what isn't. No signup, no sales call.

Install folder-nature → Work with us
Backed by
△Illia Hladkyi
Join them ❤