Agents you can actually count on.
NLC is an army of AI agents on a private server we run for you. Every job is bound by a contract. Every plan is attacked before you see it. And no agent can say "done" without proof.
You don't learn NLC. It learns your business.
NLC is built for the visionaries.
The ones who could not settle for a typical life, even if they wanted to. Whether you're just starting out or already making it — your vision is still alive, and you want more.
"Increased productivity" doesn't move you. Your business is your art, and your standards have only grown since Day 1.
So a single agent won't do.
You want an army of autonomous agents across your business — not 'managing' facets of it, but owning them. Responsible for them. Leading each department forward.
You've seen the demos: personal 'Jarvis' builds, AI 'boards of directors'. Impressive — until you notice you still have to drive them, schedule their wake-ups, and scope every job yourself.
They are infantry. You need generals.
NLC gives you that.
Done means done.
Tell an NLC agent what you want.
Vague? It asks questions until it isn't.
Clear? It checks your instructions for conflicts first.
Then it goes silent and works — until it is actually done.
When most agents say "done", they leave you to find the hundred remaining problems.
NLC agents test their own work to hell and back. Done means done.
And there is nothing to learn. No prompt libraries. No slash commands. No babysitting.
Describe the outcome. The system handles the rest.
Agents fail in eight known ways. NLC has a mechanism for every one.
If you're the right fit for NLC, you don't buy hype. When you hear a claim, you want the mechanics of it.
This section is written for you. Eight failure modes, eight mechanisms. Open any of them.
An agent's primary failure mode is scope drift. You give it a task — it does too much, too little, or something else entirely.
Models drift unless a human nudges them constantly. We refused to accept that.
So in NLC, every job creates a contract. It contains:
- A one-sentence, binary Objective Statement
- Success criteria and atomic claims
- Decomposed steps to fulfill them
- Required proof — chosen to be impossible to dispute
- Work items and live status
The contract is reinjected into the agent's context at the start of every turn. After a compaction, a larger and more detailed version goes in. The agent is mechanically forced to update it at every step — and you get a link, so you can watch it happen live.
Twenty compactions into a job, an NLC agent still hasn't drifted.
Every contract lives in a PostgreSQL ledger — who, what, when, how. And the part that matters most: the required proof.
Once the proof standard survives adversarial review, it is locked in the ledger. When the agent tries to claim "done" and end its turn, it must submit that proof. We call this the oracle.
If the proof doesn't satisfy the oracle, a stop-turn hook physically prevents the agent from stopping. It keeps working until the proof lands — or until you hit stop yourself.
In NLC, "done" is not a feeling. It's a checked claim.
Agents don't follow instructions 100% of the time. 90%, maybe. And the missing 10% can do the kind of damage that takes weeks to undo.
So we made the only possible path the right path.
Every dispatch — including research and investigation — passes through the Doorway, where the agent is enriched before any work begins:
- Historical data: recurring issues and how they ended
- Semantic recall of past conversations, with timestamps
- A knowledge graph of what exists, what connects, and what can be inferred
- The relevant code, when there is code
Agents hit the ground running instead of hunting for needles in haystacks.
After enrichment, the agent researches and proposes: objective, success criteria, action plan, and the strongest possible proof. The Red Team attacks the proposal; only the strongest surviving plan ever reaches you.
Your authorization grants a Lease — scoped permissions over exactly which files the agent may touch and which actions it may take. The accepted proof criteria are locked into the ledger at the same moment.
No matching proof, no return — the system holds the agent until the contract is satisfied. Then its findings are deposited back through the Doorway, so every future agent inherits the experience.
Very easy to do the right thing. Very hard to do the wrong thing.
Agents are thorough when they talk and shallow when they execute: bad architecture, side effects nobody predicted, underscoped requirements, missing steps.
The cure is adversarial review — an adaptation of Daniel Miessler's red-teaming work. Every contract goes through it.
A third-party agent is spawned with the objective and success criteria — and deliberately without the reasoning trail, to prevent bias. It has one job: spawn attackers.
The attackers challenge every claim and every assumption, from every angle. The findings converge into a stronger plan — which gets attacked again, looping until only the strongest version of the argument survives.
The only plan you ever see is the best plan an AI can make for you.
1 TO 32 ATTACKERS, SCALED TO RISK AND COMPLEXITY — SO TOKENS GO WHERE THE STAKES ARE.
Most agents wake up with amnesia. NLC agents wake up with an institution behind them: three memory systems, each solving a different failure.
The Knowledge Graph
9,000+ nodes and 25,000+ edges mapping how code, documents, business logic, and past conversations connect.
Ask "where does session routing happen" and get the three files that own it, the two decisions that shaped them, and the five conversations where it changed.
Rebuilt automatically as the system changes, so it never goes stale.
The Memory Log
A running diary of decisions made, corrections given, failures hit, and lessons learned.
When you say "don't do that again" — this is where it goes, and why the next session doesn't repeat the mistake.
It answers: what did we try, what did you prefer, what already broke.
The Postgres Ledger
Typed, durable records: problems, decisions, workflows, evidence — each carrying provenance, confidence, and verification state.
Agents must consult it before changing anything, and deposit findings back before finishing.
A code change is not proof of a fix. Result evidence is — and it lives here.
How they work together
- Query the knowledge graph for the shape of the system
- Search the memory log for prior attempts and your corrections
- Read the ledger for typed evidence and known failure patterns
- Only then fall back to raw exploration
Every agent starts from accumulated knowledge. Never from zero.
Dump 500 plugins on an agent and three things happen: context starvation, conflict paralysis, and you — stepping in to pick the right tool.
The moment you're choosing tools, you're babysitting an interface, not using an agent. "More is better" is a trap: overflow the context budget and the system starts trimming the instruction manuals; stack five similar skills and the agent hallucinates a hybrid of all their contradictions.
This is the difference between giving a surgeon a room full of 500 unsorted instruments and giving them a kit where the right scalpel is already in hand.
In NLC, capabilities are engineered into a decision tree, not piled into a list. Overlap is resolved before it ever reaches the agent. Say "audit the deployment" — the audit skill, the gateway integration, and the health checks load themselves.
Only relevant tools load, so instructions stay complete and context stays clear. The human bottleneck is removed.
Context starvation, conflict paralysis, and the human bottleneck.
A decision tree, not a flat list. Overlap resolved before it reaches the agent.
COUNTED FROM THE RUNNING SYSTEM · 2026-07-07 · THE SKILL COUNT FALLS AS SMALL SKILLS CONSOLIDATE INTO ROUTED LARGER ONES — THAT IS THE POINT
Agents don't coordinate on their own. They don't look across the table and say "you take the forecast, I'll take the deck." They charge in.
One agent is fine. An army without coordination is an organizational disaster: agents overwriting each other, duplicating work, stacking half-done changes on top of old problems.
So NLC provides the coordination layer — three mechanisms that make parallel work safe:
You assign objectives, watch the ledger, and collect finished work.
FIG. 8Automatic Ticketing
Every authorized job gets a ticket, and from that moment every attempt, step, and finding is tracked. If something goes wrong, you have a permanent record of exactly where.
FIG. 9Isolated Workspaces
Every agent gets a clean room with a copy of what it's working on — nothing can touch your live system. When it believes it's done, it must prove the work inside that room and present the evidence for your approval.
FIG. 10Intelligent Merge Logic
Parallel work creates parallel files; without a merge engine, one gets overwritten — or you get a corrupted hybrid. NLC validates that the pieces fit, integrates them safely, and spots conflicts before the collision, not after.
Parallel work, without the parallel chaos.
Autonomous doesn't mean abandoned.
Sometimes the agents hit a complexity wall. Sometimes you're overwhelmed with possibilities, or need a second opinion on how to structure a massive workflow. When that happens — I've got your back.
NLC buyers don't get support tickets. They get solutions. I'll help you spot what's safe to automate, refine your proof standards, and scale from one agent to leading an army.
"NLC is my big massive bet for the next 10 years. It's why I'm willing to stay up till 4am working on the app, sleep two hours, and get up to work again."
Bring requests directly to me. I ship fast.
And everything else.
Hermes is a first-class runtime inside NLC — a second engine of capabilities. The first of them: your agents aren't trapped in a dashboard. They live where you do.
You don't have to install NLC.
When you sign up, we provision a private VPS just for you. Your NLC, your work, and all your agents live there — and the files on it belong to you.
We give you the link and the password.
You log in.
And you're in.
Details only when you want them.
NLC lets you zoom in and be the craftsman, or zoom out to be the strategist — whenever you want.
This page was built the way NLC works.
Designed, built, audited, and shipped by an NLC agent, inside NLC, under the same mechanics described above. The design review gate mechanically blocked this page three times for real defects before it was allowed through. The record below is not an illustration — it is the actual gate history of this page and its lineage.
COUNTED FROM THE RUNNING SYSTEM, 2026-07-07: 158 SKILL PACKAGES · 231 SCRIPTS · 25 MCP SERVER IMPLEMENTATIONS · 236 TOOL DEFINITIONS. THESE NUMBERS ARE PROJECTED FROM THE REPO AND MOVE WHEN THE SYSTEM MOVES.
Start with a 30-minute AI Acceleration call.
We'll look at where your business loses time, map the bottleneck, and I'll tell you honestly what NLC can do about it — including "not a fit yet," if that's the truth.
I'm still in "do the unscalable" mode, collecting market feedback — so this call is as valuable to me as it is to you.
Most people I talk to already want agents in their business. They just don't know where to plug them in.
That's where I can help. Book the call and I'll help you figure this out.
A 30-MINUTE ONE-ON-ONE AI ACCELERATION CALL WITH FELIX.
FAQ
Who is the host?+
The Host is Felix. I don't have any IP protection set up on NLC and NLC is closed source. So I need control so people don't steal the whole repo and call it their own. It's also more friction to perform maintenance if NLC is installed in their device over mine.
What happens if you pass away or bankrupt?+
If I die, go bankrupt, or become uncontactable, then you call your bank and dispute the charge. These days banks always favor buyers over sellers so when that happens they will just deduct the funds from my account. You will automatically get unsubscribed and so on. Worst case scenario, call the bank and cancel the card.
Your data is safe. First thing you do when you come into NLC is to tell your agent to set up Autosave. It will save your data on your own Github repo and can be downloaded/inspected there any time. Easy.
Can we transfer the host to ourselves if you pass on?+
No.
What happens if you don't want to work with them anymore or vice versa?+
You collect your data from your GitHub repo (that you Autosaved) and we professionally say goodbye to each other. No, you don't get to keep NLC. That's my work.
If your concern is really just a matter of your personal data, you're safe.
What happens if I stop my monthly paid subscription of my AI?+
You want to use AI without paying for AI? You can't.
There are "free" and open-source models on Hugging Face that you can download. But after you download, you need the hardware to run them. The newer quantized models that can run on your phone are really dumb and I don't think any serious business would count on them to do reliable work.
And powerful AIs chug electricity — a lot. Not to mention the hardware needed to run those models is…not cheap. At all.
OR if you're thinking you want to use those free ChatGPT or Claude subscriptions, I would say…don't. You'll be more frustrated than having any real work done.
Good news is, I can recommend to you cheap subscriptions that REALLY excel in agentic work if you want.
What happens if I stop monthly subscription to NLC?+
I take back my VPS and you take your data from GitHub. You need to, of course, set up Autosave.
You don't get to keep NLC.
What happens if you decide to stop shipping?+
Then you will become an unhappy customer and choose to stop the subscription. I won't try to stop you.
Because if I stop shipment, it can only be because I've stopped working on the project. I don't think that's a likely outcome because NLC is my big massive bet for the next 10 years.
It's why I'm so willing to do the unscalable right now and do shit like staying up till 4am to work on the app, go to sleep for 2 hours and wake up again to work.
But the question is valid and I'm happy you asked.
How is NLC secured if everything runs on a VPS?+
Each customer gets a dedicated VPS with its own public IP. NLC is not deployed as a shared multi-tenant app where multiple customers live on the same machine.
On a fresh install, NLC does a few concrete things automatically:
- Turns on the firewall
- Opens only SSH (22) and the NLC web app port (3080) by default
- Installs fail2ban, which temporarily bans repeated failed SSH login attempts
- Generates fresh secrets for that machine instead of reusing one shared secret across customers
- Protects the web UI behind login with password hashing and temporary lockout after repeated failed attempts
- Keeps some internal control routes reachable only from the VPS itself
- Keeps some local monitoring bound to 127.0.0.1, which means it is not publicly exposed
NLC is deployed behind Nginx with HTTPS on a custom *.nexlevelclaw.tech domain, so customers access it through a proper URL such as felix-tay.nexlevelclaw.tech, not a raw server IP and port.
That does not mean "perfect security forever." It means the server is set up with real protections on day one instead of being left raw on the public internet.
There is also the second half of the problem which most people don't talk about… careless agent behavior.
Most AI data leaks happen because people give the agent too much freedom and then hope it behaves. NLC reduces that risk through scoped work, approvals, and proof requirements rather than giving one raw agent blanket access.
What stack is NLC actually built on?+
The core stack is mainly Node.js / JavaScript. More specifically:
- web app: Express + WebSockets
- process supervision: PM2
- workflow and memory state: PostgreSQL + SQLite
- browser layer: primarily agent-browser and Browser Control
- Playwright: present too, but mainly as a fallback and testing path rather than the main browser lane
- integrations: separated behind gateway / MCP / relay layers
- some heavier subsystems use Docker and optional local model infrastructure when enabled
So this is not one giant black box. The installer, web app, browser layer, workflow/memory layer, and integrations are separated into distinct parts.
How easy is it to request future modifications?+
Just bring that request directly to me. I ship fast.
NLC is not a self-serve product where customers are expected to rewire the core system themselves, but it is a system I can extend and improve over time based on real customer needs.
The reason why I keep control of the core product is so that NLC stays stable and updateable, rather than turning into a patchwork of client-side custom versions.
How scalable is NLC if several people are using it at the same time?+
NLC is purpose-built for many sessions and multiple agents all running concurrently, but concurrency still follows physics.
The first real bottleneck is usually VPS size and workload mix, not the idea of "multiple people" by itself. If several humans, browser work, local models, indexing jobs, and background agents are all sharing a small machine, performance will drop.
So the right question is not just "Can NLC handle six people?" in the abstract. The right question is: what size machine is this team workload actually running on?
If a customer expects heavier team usage, the VPS should be sized for that reality, and we provide options for those upgrades.
Why should I care that NLC allows for many parallel agents working at once?+
Multiple agents are there for:
- division of labor
- parallel work
- review
- adversarial checking
- edge-case inspection
- proof and verification
One agent can prepare the work. Another can attack the plan. Another can check edge cases. Another can verify that the result actually holds up.
That improves thoroughness. It does not replace proper security, proper scoping, or proper server sizing.
Do we get direct access to the data on the VPS?+
Of course. The VPS belongs to me. The files on it belong to you.
That includes project files, generated assets, uploads, logs, transcripts, and working data stored on that machine.
There is even a function in NLC called AutoSave you can set up that backs up all your data to GitHub automatically. Just talk to your NLC agent and you'll get that set up in minutes.
Can we easily download assets generated by the bot and inspect hosting issues ourselves?+
Yes. Generated assets can be retrieved directly — just talk to your NLC agent and ask them for a download link.