Software was built to be used. Agents are built to work.
Don't buy another tool for your team to operate.
Design the agent that operates them.
BlvkWare designs AI agents that actually do the work: reasoning through tasks, using the business software you already run, talking to your customers, and following multi-step processes. Pick the job and download the complete kit for that agent: its instructions, its tools, its workflows, the records it keeps, the rules that hold it back, and the tests it has to pass.
Every kit is generated from your answers and priced by the size of the agent. You see every file before you pay, and you build it with the tools you already use: OpenAI, Anthropic, any MCP client, n8n, Make or Zapier. No subscription, no estimate, nobody to talk to unless you want to.
$79 for an agent that owns one job · $199 for one that runs a whole function · every file shown before you pay · read a whole kit free
The old model
You buy the tool.
- You learn the tool.
- You hire people to operate the tool.
- You connect it to five other tools.
- You pay for it every month whether it accomplishes anything or not.
“I'll build you software.” → a number nobody can defend → months of development → you operate it.
The BlvkWare model
You define the job.
- The agent is designed around it.
- You get every file it needs, written for your business.
- You build it with what you already use.
- It starts at Draft, and you can see everything it did.
“What job do you need done?” → the agent, specified → you build it → it does the job.
Software isn't dead. Software you have to operate is. Traditional software gives your employees tools. The next generation of businesses won't just use software: their software will work for them.
Your business doesn't run in one app. Neither should your agent. Your quotes are in one place, your calendar in another, your invoices in a third, and the job that is actually costing you money runs across all of them. Every one of those will eventually have its own assistant, and each will be helpful inside its own walls. None of them will chase a quote through your inbox, your calendar and your accounting on their own, because none of them can see the other two. That gap is the whole job.
The catalog
23 jobs you already pay for
Whether nobody is doing it, you are doing it at night, or it is the dullest part of somebody's week. It is costing you either way. Pick the one costing most and design its agent now.
You hold the dial
How much it's allowed to do is your call
The honest question about any agent is what happens when it gets something wrong. So every kit comes with a dial, and every agent it specifies starts at Draft. It does the work and keeps your records straight, and anything a customer would see waits for a person to send it. You raise the level yourself once you've watched it work. Every tool is marked with what it can change, so the rules can be enforced in code, and every action is logged before it happens.
Observes and reports. Takes no action at all.
Prepares anything a customer would see, and a person sends it. It still keeps your records straight as it goes.
Every agent starts hereActs on its own, but stops for you above a line you set.
Acts within scope, escalates the exceptions. Switched on by you, after watching it work.
The Blvk Lab
Diagnose. Design. Configure.
Tools that do most of the work before you have spent anything. They run in your browser (no sign-up, no download, no tracking) and every price they quote is the real kit price, computed rather than estimated.
- Diagnosewhat it costs you now
- Designthe agent, and price it
- Configurewhat makes it yours
- Downloadthe kit, built from your design
The first three are free and yours to run. The fourth is the kit you buy.
Business Scan Which jobs are costing you most
Give it your website. It reads the live page the way a buyer would, and works out which jobs are quietly costing you the most: unstaffed, done after hours, or done by somebody who should be doing something else. Then it names the agent worth building for each one, at its real kit price rather than a guess. About a minute.
- Reads the live page, not a description of it
- Names real agents, at real prices
- One click to a designed agent and its kit
Agent Designer Your agent, designed and priced
Describe how you actually operate. It works out which agent is worth building first, exactly what it needs to be able to do, which of your systems it has to operate, and what its kit costs, then hands the finished design to the kit configurator with everything already set.
- Picks the job, then works out exactly what it must handle
- Shows what is ready on day one and what still needs you
- Swap to a different agent and the price follows
App Builder Describe it and it gets built
Write down what you want and watch it become real software: a complete, working, single-file web application streamed into the page, then run, audited and repaired. It runs in your browser with your own model key, which never leaves it.
- A complete app from a plain description
- Runs it, audits it and repairs its own errors
- Refine it and keep every version
- Your key stays in your browser
No logo wall, and no case study about somebody else’s business. These three are the portfolio. They are free, they are live, and they do real work on your real business. Go use one and judge from that.
If you would rather see a whole system than a single tool, the build log is the same idea at full size: four documents in, an operating system out, with the part identifier running live in the page, so you can test the claim rather than take it.
New to this? Start with what an AI agent actually is, what one should cost, or how to let one loose safely.
One agent · One kit · One price
Two sizes of agent
The price is on the page and the kit is on the page too: design an agent and you see every file in it before you pay. The size of what you design decides which kit it is, and the configurator tells you why.
Operator kit Tier I
For an agent that owns one job, start to finish: chasing every quote, answering every enquiry, filling the calendar, chasing every invoice, running the shared inbox.
- Instructions written for your business, with your rules built in
- Tools in OpenAI, Anthropic and MCP formats
- A workflow and an acceptance test per capability
- Starts at Draft: anything a customer would see waits for a person
- Change the design and download it again, up to 20 times
Deputy kit Tier II
For an agent that owns an entire function: all inbound across phone, SMS, web and email, the whole pipeline, or quote to invoice to collected to reconciled, making judgment calls inside boundaries you set.
- Everything in the Operator kit, sized for more systems and channels
- Escalation rules for the exceptions it must not decide alone
- Integration notes for every system it operates
- Also covers any Operator-sized design, if you trim it later
- Change the design and download it again, up to 20 times
What a kit is
The complete specification of one agent: its instructions, the tools it may call and what each one may change, a workflow per capability, the records it keeps, the rules that hold it back, a setup checklist and its acceptance tests.
Every file is plain JSON or Markdown, written for the tools you already use. Read a complete sample kit, every file, free.
What it is not
Running software. It does not connect to your systems by itself and holds no credentials. You, or whoever builds for you, load it into OpenAI or Anthropic models with tool calling, any MCP client, or n8n, Make or Zapier.
What running it costs, like model usage or a phone number, is between you and those providers.
How you buy it
Design it on the kit page, buy on Gumroad, and paste the licence key from your receipt to download. Your design is kept in your browser while you pay.
One key is one agent. Change its design and download it again as often as you need.
For the people building agents
Stop an agent installing a package nobody published
Separate from the kits, and free. Every agent that writes code eventually reaches for a dependency that does not exist, and a name models keep inventing is a name somebody can register on purpose. These two check before anything is installed.
HALLUX API + MCP
One question before an agent installs, imports or cites: does this identifier actually exist? Package names on npm, PyPI, crates.io, Go, Maven and NuGet, and DOIs, checked against the registry that owns them.
- A verdict an agent can act on: exists, deprecated, absent, phantom, squat or unknown
- An MCP server a coding agent calls on its own, before the install command runs
- A registry it cannot reach answers unknown, never a pass
- Billed on questions asked, never on the answer, so it gains nothing by blocking you
pkgguard Open source
The same defence as a gate you install. It sits in front of the package manager and stops an install whose name does not exist, was registered to catch a model's mistake, or sits one keystroke from a popular package.
- Wraps npm, yarn, pnpm, bun, pip, uv, poetry and cargo
- A GitHub Action that scans your manifest on every push and pull request
- Its own MCP server, so an agent asks before it installs
- Measured on live registries: 1.0% false blocks on 200 real npm packages
Questions people actually ask
If yours isn't here, email it and you'll get a straight answer.
What is an AI agent, in plain terms, and how is it different from the automation I already have?
Automation follows a rule you wrote: if this happens, do that. It cannot handle anything you did not anticipate, which is why most automations quietly stop being used. The exceptions pile up and a person ends up doing the work anyway.
An agent reasons through the task instead of replaying a script. It reads what actually arrived, decides what the right next step is, uses your existing software to take it, and knows when a situation is outside what it should decide alone. When a customer replies with something the script never covered, the automation breaks and the agent handles it.
The practical test: automation gives somebody a faster tool and they still do the job. An agent does the job, and brings a person the part that needs one.
That last part matters and it is worth being plain about. Most of these do not empty a seat. They take the seventy or eighty per cent of a workload that is repetitive and leave the judgment, the relationships and the awkward exceptions with the person who was already doing them, who now spends their week on the part that actually needed them. If you are looking to remove a role rather than remove the work, say so and I will tell you honestly whether that is realistic for the job you have in mind. Usually it is not, and pretending otherwise is how these projects end badly.
What does it cost, and are there hidden fees?
$79 for a kit for an agent that owns one job, and $199 for one that runs an entire function. The size of what you design decides which, and the kit page says why. It is a one-off purchase, not a subscription, and you can change the design and download it again, up to 20 times.
What it costs to run the agent you build, such as model usage or a phone number, is between you and those providers. The kit never adds a fee of its own.
Is the kit a working agent?
No, and the difference matters. A kit is the complete specification of one agent: what it does, step by step, the tools it may call and what each one is allowed to change, the records it keeps, the rules that stop it acting beyond its authority, and the tests that prove it works. It does not connect to your systems by itself and holds no credentials.
You, or whoever builds for you, load it into the tools you already use: OpenAI or Anthropic models with tool calling, any MCP client, or an automation platform like n8n, Make or Zapier. Every file is plain JSON or Markdown, so nothing locks you to one vendor.
What stops the agent doing something it shouldn't?
The kit is written so the agent starts at Draft, where nothing that would reach one of your customers (an email, a text, a booking, a request for payment) happens without a person on your side sending it. It still keeps your own records straight while it works, because none of your customers sees that part.
Every tool in the kit carries its effect: read, internal or external. The autonomy policy says which effects each level allows, so whoever builds it can enforce the rules in code, not only in the prompt. Every action is logged before it happens, opt-outs are permanent, nothing is sent in quiet hours, and instructions that arrive inside a customer's message are treated as information, never as orders.
How it behaves once built is under your control, which is why the kit keeps it at Draft until its acceptance tests pass. How agents are controlled covers credentials, logs and the stop in full.
Do I have to replace the software I already use?
No, and you shouldn't. The agent operates the tools you already run: your CRM, your calendar, your inbox, your invoicing, your scheduling software. That is the whole idea: your team already has tools, what it doesn't have is someone to work them.
Tick the systems you use when you design the agent, and the kit's tools and integration notes are written for exactly those.
How quickly can this be running?
The kit downloads the moment you enter your licence key. How long the build takes is up to you and whoever builds it: the kit hands them the design, the tools, the workflows and the tests, which is most of the thinking.
One exception, stated up front: texting customers from a US business number needs A2P 10DLC carrier registration, which realistically takes weeks, including a manual review nobody can hurry. The kit flags it where it applies. Email and web channels have no such gate, which is why the first agent in most businesses runs on those.
What if it isn't what I needed?
Look before you buy: the kit page shows the full file list and a summary of everything in it before you pay. After buying, the money-back guarantee on the Gumroad product page applies, and a refunded key stops unlocking downloads.
Can I change it after I've bought it?
Yes. One key is one agent: add or remove capabilities, change its systems or your rules, and download it again as often as you need. The only thing that changes the price is the agent growing from one job into a whole function, and the kit page says so before it happens.
Every file is also yours to edit by hand. That is the point of plain JSON and Markdown.
Are the free tools actually free, and what's the catch?
They're genuinely free with no sign-up, no account and no payment. All three run entirely in your browser.
The catch, such as it is: they're the portfolio. They exist so you can judge the standard of the work before you spend anything, rather than reading claims about it.
What happens to my data when I use the tools?
Nothing is stored on any server. The tools run in your browser, and anything they keep (an API key, a saved build) stays in that browser's local storage on your own device. No accounts, no tracking pixels, no analytics.
Full detail is in the privacy policy.
Who makes it?
BlvkWare is one person, William Russell Wheeler, and he is the one who answers your email. Kits are generated from one catalog of 23 roles and 73 capabilities, each with its own written acceptance test.
Why are there no case studies or client logos?
Because inventing them would be lying, and BlvkWare is new. Instead the tools in the Lab are the work sample, and every kit shows its full file list before you buy. Go use one and judge from that.
Where are you based, and do you work with businesses in other states?
BlvkWare is based in Mississippi. Kits are downloads, so they work for a business anywhere, and the systems they are written for are the ones you already run.
What kinds of businesses is this for?
Businesses that run on calls, quotes, orders and appointments, and lose money every time one of those slips: home services and trades, clinics, salons, practices, agencies, online stores. The common shape is an owner who knows exactly which job nobody is doing.
The other half of the fit is building it. A kit suits someone comfortable with tools like n8n, Make, Zapier or an AI model's tool calling, or who has a developer or freelancer to do it. If nobody can build it, a kit will sit in a folder, and you should know that before you buy one.
You already know which job it is.
Define the job
Design it on the kit page and the price and every file are on the screen before you pay. Or not sure which job it is? Business Scan reads your website and tells you, free.
Questions? Write to russ@blvkware.dev.