███╗ ██╗ ██╗ ██████╗ ██╗ ██╗ ████████╗ ███████╗ ████╗ ██║ ██║ ██╔════╝ ██║ ██║ ╚══██╔══╝ ██╔════╝ ██╔██╗ ██║ ██║ ██║ ███████║ ██║ ███████╗ ██║╚██╗██║ ██║ ██║ ██╔══██║ ██║ ╚════██║ ██║ ╚████║ ██║ ╚██████╗ ██║ ██║ ██║ ███████║ ╚═╝ ╚═══╝ ╚═╝ ╚═════╝ ╚═╝ ╚═╝ ╚═╝ ╚══════╝
An AI agent that lives on your VPS. It watches your services, catches crashes, writes real fixes, deploys them, and remembers everything — forever. You do nichts. It does everything.
Used in production · 24+ services managed · Running 24/7 autonomously right now
Service crashed at 3 AM? Nichts already fixed it. It reads the logs, diagnoses the root cause, writes a real code fix, deploys it, and verifies the fix is live. You see a notification. The service sees zero downtime.
Every fix, every preference, every decision — stored in a hex-indexed knowledge tree that never forgets. Other AI tools start from zero every session. Nichts picks up exactly where it left off, with full context.
At your desk? Use the dashboard. On the train? Open Telegram. In a terminal? Use the CLI. Same agent, same memory, same capabilities. "Hey, restart the docs service" — done in 3 seconds, from your phone.
Not just maintenance — evolution. A self-improvement loop runs autonomously: scan for issues, analyze root causes, implement fixes, verify impact. Your infrastructure gets measurably better every week. You do nichts.
See something broken in your dashboard? Shift+right-click on it. Nichts sees the component, its data, the surrounding context — and opens a chat about exactly that element. Comes as a browser extension too: select text on any website, hit the shortcut, Nichts acts on it with full context. One keystroke away, everywhere.
After every session, Nichts scans its chat logs for errors, corrections, and missed context. Found a pattern? It writes a new rule for itself. Got corrected twice on the same thing? It updates its own personality file so it never happens again. This isn't fine-tuning — it's self-authored behavioral evolution, in plain text you can read and edit.
Other tools answer questions. Nichts reads your code, writes fixes, commits to git, deploys to production, monitors the result, and saves what it learned. While you're asleep.
An autonomous loop scans for issues, writes real fixes, deploys them, and measures the result. It eliminated an entire class of port collision bugs by creating a structural config file — no human involved. Your platform gets better while you're not looking.
Fixed a tricky bug 3 months ago? Nichts remembers exactly how. Your preference for direct fixes over explanations? Stored. That weird port conflict on the office server? Documented, categorized, searchable. Every session makes it smarter.
Multi-stage pipelines that mix shell commands with AI reasoning. Cron triggers, crash recovery, stepping mode. The pipeline engine runs independently — even if the dashboard goes down, your automations keep running.
Real-time SSH intrusion detection. GeoIP tracking of every attacker. Automated fail2ban management. UFW firewall control from the dashboard. A live threat timeline that shows you exactly who's knocking on your door — and blocks them automatically.
This isn't a one-shot fixer. It's a continuous cycle that runs day and night — each iteration building on everything it's learned before.
Watches every service. Reads logs. Checks health. Scans for patterns other tools miss. Its memory means it knows what "normal" looks like for your system.
Doesn't just read error messages — reads your source code. Traces the actual call stack. Cross-references against hundreds of past fixes in its memory tree. "I've seen this before" isn't a guess — it's a database lookup.
Writes real code. Commits to git with clear messages. Builds before deploying. Stops services before builds so nothing crashes. Reviews its own diffs. It follows rules it wrote for itself — because it learned the hard way.
Hits every endpoint. Reads the post-deploy logs. Confirms the fix is actually live — not just committed. Measures impact against previous cycles. Then saves what it learned so next time is even faster.
24+ services, battle-tested in production. Every piece is independently deployable, version-controlled in its own git repo, and managed by the agent.
The brain. Real-time monitoring, AI chat, one-click deploys, log viewer, security dashboard, job pipelines. Everything from one screen.
Plug in Claude, Codex, or both. Provider adapter pattern means adding a new AI is just a new file. Sessions persist across restarts. Jobs queue and run concurrently.
Cron triggers. File-watch triggers. Multi-stage execution. Crash recovery. Runs independently of the dashboard — your automations never stop.
SSH intrusion logs. GeoIP attacker maps. Fail2ban management. UFW rules from the UI. Live threat timeline via WebSocket. Know who's attacking and stop them.
Hex-indexed knowledge tree. Usage tracking. Compaction rules. A "memory gate" that forces the agent to save knowledge before responding. It literally can't forget.
Headless browser for scraping. Speech-to-text for voice commands. Collaborative docs. A browser IDE. Telegram bot. Uptime monitor. All wired into the agent.
Every service is a git submodule. The agent has full read/write access. It can edit code, deploy, rollback, and create new services. Your entire stack, orchestrated by AI.
Three files define Nichts. CLAUDE.md controls how it works. SOUL.md controls who it is. The memory tree stores everything it's ever learned. Together, they create an agent that develops over time — not just a model with a prompt.
CLAUDE.md — the rules it follows
SOUL.md — the personality it develops
memory/ — everything it's ever learned
Before every single response, Nichts asks itself: "Did I learn anything?" If yes, it writes to memory before it replies. It's physically incapable of forgetting what it learned.
The agent notices patterns in how you work and proposes personality changes. "You prefer directness — should I stop asking permission and just fix things?" It becomes your agent, not a generic one.
Not a demo. Not a pitch deck. This is the real infrastructure behind a real business, managed by Nichts 24/7.
A single 4 vCPU / 8GB server runs 24+ services via PM2. The dashboard, agent API, pipeline orchestrator, security suite, Telegram bot, uptime monitor — all on one box. Nichts manages every process, deploys code, and restarts services autonomously. Daily backups via Hetzner snapshots.
Nichts manages external servers too. An office server connected via Tailscale VPN. A separate Coolify node for managed deployments. SSH or API — Nichts reaches across machines, runs commands, deploys code, and monitors health. Add a new node in minutes.
The entire Nichts network runs behind Tailscale. No ports exposed to the public internet. Dashboard, agent API, node-to-node communication — all encrypted and private. Attackers can't reach what they can't see.
Every service is its own git repo, tracked as a submodule. Nichts commits with clear messages, pushes to GitHub, and the parent repo tracks which commit each service is on. Full rollback to any point in time. Your infra has the same version control as your code.
Not just a server tool. Nichts lives in your browser too. Select anything, anywhere — and Nichts acts on it.
Error messages, code snippets, URLs, documentation — select it, hit the shortcut, Nichts gets it with full context.
See a broken component? Shift+right-click on it. Nichts sees the element, its data, the code behind it — and opens a chat about exactly that thing.
Nichts doesn't just get the text you selected. It gets the page URL, surrounding context, your memory tree, and knows which service you're probably talking about.
One payment. Full source code. Set it up on your VPS and let it run.
Your server will be better tomorrow than it is today.
This entire platform was built with AI coding tools — Claude Code, Cursor, and similar agents. There are no hand-written unit tests. The architecture emerged organically through hundreds of AI sessions, not from a whiteboard. It works great in production, but it's not a polished enterprise product with 90% test coverage.
What this means for you: Setting up Nichts requires comfort with AI coding tools. You'll use Claude Code or similar to configure, customize, and extend it — the same way it was built. We include a FOR_THE_AGENT.md file that explains the entire platform to your AI agent, so it can understand the codebase and help you set everything up.
If you're the type who vibe-codes your infrastructure — this is your dream setup. If you expect npm install && npm start with zero AI involvement — this isn't for you.
One price, everything included. You bring your own API keys (Claude, Codex) — you control that cost.
This isn't VC-funded bloatware. It's the exact tool I run my own infrastructure with. You get the same version, same capabilities.
Get every new feature, service, and improvement for a full year. After that, you keep the code forever — it just stops auto-updating.
If Nichts catches one crash, writes one fix, saves you one emergency — it's already paid for itself. Everything after that is free.