All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
ServerManager.ai: SaaS Linux fleet management (open source agent)
Along side https://hypervisor.io/ we have been building this for a while, thought the LET crowd might find it useful. If you manage more than a handful of VPSes, this saves a lot of SSH hopping.
How it works:
Install our open source Rust agent on any Linux box with one curl command. Everything else is managed from our dashboard. No self-hosting required.
What you can do from the dashboard:
Real-time telemetry (CPU, RAM, disk, load), remote command execution with allowlist enforcement, and an interactive browser shell with a full PTY terminal. Every action is logged with an audit trail.
The agent:
Written in Rust. Sub-50ms command dispatch. Self-updating with Ed25519 signature verification and staged rollouts. mTLS connections, JWT auth, certificate fingerprinting, rate limiting. Open source.
AI features (optional):
Connect an LLM provider and you unlock structured plans for fleet operations, bulk doc generation per server, predictive monitoring with anomaly forecasts, and an AI chat that executes fleet commands inline with approve-all. Bring your own keys. Supports 15 platforms.
Pricing:
Free tier available. No credit card to start. Deploy agents in minutes. But agents start at $2/mo per agent
Happy to answer questions. The agent is open source if anyone wants to poke around. Will post github link soon.
Comments
how secure is it compared to those ai prompt "Make My VPS server Secure" type tutorial?
Basically you will get commands twith explanation to what its suggesting and which ones you wanna approve, apart from that we have loaded it with alot of finetuned vectorless rag structured information layer aligning towards hosting/devops/security as well so no matter which llm you use it will utilise our layer
Looking forward of release source code
>
The interface currently looks unimpressive: unintuitive buttons, hard-to-read text. I understand there's no functionality yet, other than viewing the status. And I'd like to see the IP addresses of each server. The free mode only offers 5 agents, with each additional agent costing $2, which is a bit much for me, especially if I have my own AI. It's unclear how it differs from other open-source free metrics tools, like Kuma. And what's the load on the VPS?
UI was intentionally made to look monochrome, it was never the target, alot of things are coming soon which will make it worth your time, give the project some time before you make a decision is all I can say at the moment. @Snyppi also its not only about having your own AI, it's about what information is driving that AI to give you a better experience.
What could go wrong?
At this point, you probably could install llama.cpp, take a decent LLM like Qwen 3.6.
Give it a decent prompt, to do that, on your system.
The idea of some company, getting root shell access, to a bunch of my machines, naaaaah bro.
When are you sharing the GitHub link?
https://github.com/hypervisor-io/smagent
Totally fair instinct, honestly if you want to wire up llama.cpp + Qwen yourself, feel free to do so, the LLM is the least interesting part of this product. The best part is the plumbing around it: fleet-wide telemetry, alert state machines, patch compliance across 50 boxes, and getting an agent architecture going.
On the root access point, a few specifics:
The AI provider is pluggable. Point it at any OpenAI-compatible endpoint, including your own llama.cpp server. Local Qwen analyzing your fleet, our orchestration. Best of both worlds.
What's the default black/whitelist?
I assume its by default, pretty open?
The thing I don't like is the centralized architecture.
Lets say, your guys get hacked and since you essentially have nearly unrestricted shell access to a few hundreds if not thousands of machines, at some point.
Makes you a juicy target.
How do you protect yourself against this?
Do you put each account on their own instance?
Never going to deny that it's not a possibility, and yes by default the allow list is wide open unless you configure it on your own, I can add nothing more constructive or defensive to this argument at this point of time. That being said, we have taken as many precautions as possible to eliminate vectors of getting attacked. In near future we will most likely release a self hosted licensed version similar to https://hypervisor.io/ just for peace of mind I guess.