Your NAS shouldn't just store your files.
It should think.
SynapseNAS adds a private brain to the NAS you already have. A local AI operates your server in plain English — files, apps, logs and all — while every change it makes is saved, versioned and easy to undo. It even reads your documents into a private knowledge base only you can search. All on your own hardware, and nothing it can't take back happens without your okay.
A private AI brain for your whole house.
A full local-LLM stack and a real agent runtime, built into the OS. Your server doesn't just hold your data — it works on it. Privately, on your own GPU, by default.
Run any model. Keep everything private.
Download and run any open-source model on your own hardware. Your prompts and data never touch an external server. We've included a playground and OpenAI-compatible API so you can use it like any other service—just locally.
It reads your documents, so you can just ask.
Drop in a file and it's read, sorted and indexed entirely on your NAS — nothing leaves the house. Names, dates and amounts are pulled out by plain, predictable code; the AI only writes the readable summary on top. Your files stay the source of truth, and passwords are pointed to, never copied.
One assistant that actually gets things done.
There's a single built-in assistant. Tell it what you want in plain English and it figures out which action to take, does the safe things on its own, and stops to check with you before anything that can't be undone. It runs on your own hardware by default — and you can hand it a cloud model whenever a job needs the extra muscle.
Safe runs on its own. Risky waits for you.
Everything the assistant can do is a named action with a risk level attached. Reading and checking just happen. Things you can take back run too, but get written down. And anything you can't undo always stops to ask — if an action isn't recognised, it's treated as risky to be safe.
Bring in a cloud model only when you want to.
Out of the box it thinks using a model on your own hardware. If you'd rather hand the harder jobs to OpenAI, Anthropic or Google, there's a clean spot to drop in your key — and it's only used while that task runs. Don't want any of it? Skip it and stay entirely on your own machine.
Run any agent — safely boxed in.
Bring your own agents and run them right here. Deploy any third-party agent — Hermes, OpenClaw, your own LangGraph or CrewAI projects, even plain MCP servers — and SynapseNAS runs it in a locked-down sandbox. A brand-new agent can do nothing at all until you hand it permissions, one piece at a time.
Every agent starts walled off: capped CPU and memory, no network unless you add an address to its allow-list, read-only except for one data folder, and any passwords it needs handed in only as it runs — never saved into the agent.
The same agent, three strengths of containment — a hardened container, a micro-VM with a real kernel boundary, or a full virtual machine for code you trust the least. You approve the deploy, and you approve anything that ever widens its reach. That's the whole deal.
A few simple rules keep it safe.
SynapseNAS works next to the NAS you already run — as a container or right on a Linux box. It never tries to take over your storage, disks or network. Three rules cover the whole thing: reading only reads, every change takes one checked path, and the assistant can only use the tools it's been handed. Anything that can't be undone waits for you.
You can see everything it does.
Every change SynapseNAS makes is written to a running log you can read anytime — what happened, who asked for it, and when. And when something takes a while — a big copy, an import, a model download — you watch it happen live, line by line, instead of staring at a spinner.
Change a setting today, undo it next week.
Every setting in SynapseNAS is kept as a plain, saved file, and each change is quietly versioned — like a document's history. You preview a change before it happens, look back over everything that's ever changed, and roll any of it back to exactly how it was.
If it has an API, it can plug in.
Every SynapseNAS tool speaks MCP — the shared language AI assistants use to call things — and the built-in hub serves them all back out. Connect anything with an API, install your MCP servers on the NAS, and reach the whole lot from one place — from Claude on your phone, ChatGPT, or your own agents. Each connection gets its own key, and that key decides exactly what it's allowed to touch.
Add an integration — it builds its own screen.
Connecting a new service usually means someone has to design a page for it. Here, SynapseNAS does that for you: it looks at what the integration can do and drafts a matching screen — buttons for its actions, tables for its data — in the same look as the rest of the system. You see it, tweak it and approve it before it's saved.
Run Windows and Linux, right on the NAS.
Some jobs only work on a particular operating system — a Windows-only app, a Linux-only utility. Instead of keeping a spare machine around for them, spin up a real virtual machine on your NAS and use it right there. Windows or Linux, as many as your hardware allows.
It's also the strongest box for agents you don't fully trust — the very same VMs do double duty as a hard, kernel-level boundary, with the same approve-before-it-runs rules as everything else.
One process. Runs where you do.
SynapseNAS is a single app — the AI, the screens and the API all in one. Start it however suits you, and point it at your server whenever you're ready.
Six panels read your real server the moment you connect it; the rest run on sample data until you do.
Give your NAS a brain.
SynapseNAS is open source and free. Clone it, run it on your hardware, and let AI agents handle your work—all private, all yours.