SVO — software verification & operations

Questions that come before the decision.

Where the audit runs

Where the audit is executed, and how I choose

It is decided by the product you hire. In Code, Code Online, Host and MultiHost, the audit is executed inside your environment, with your access, and the report is born and stays there. In Code Prime and Share, it is executed on our infrastructure, and the report arrives in your inbox with the data destruction certificate.

Each product states this in its own description: see the products.

Do I have to give access to my server? In which products?

In the products that run in your environment, the access is yours, and stays yours: the audit runs in there with the access you already use, and we neither ask for nor receive a password or a key.

There are two cases where the access is ours, and it is read-only: in Code Prime with a private address, and in Share — you paste our public key into your repository or your hosting panel. The private key never leaves our area, and you revoke it whenever you want.

In Host and MultiHost we ask whether you have shell access — because that changes how far the audit reaches —, and we never ask for the credential.

Who reads my code

In your environment, the one reading is the Claude on your own account: the audit runs on your machine, with your plan or your credits.

On our infrastructure, the processing is our responsibility, and that is in the contract.

What SVO receives, and what it is for

When the audit runs in your environment, we receive the progress — which stage closed — and the session identifier. It serves so the audit can continue from where it stopped if it needs to restart.

The report does not come to us: it is born with you and stays with you.

How I check this on my own

Use a network monitoring program you already have installed and analyse the traffic while the audit runs in your environment: what goes out, what comes in and where it goes. The check is yours, with your own tools, and it does not depend on taking our word for it.

What the data destruction certificate is, and what it covers

It comes with the report in the products executed on our infrastructure, and states three things: what was destroyed — the working area created for your project and the access key issued for your order —, that the destruction is verified by reading, and what it does not cover: the infrastructure provider's backups, the report already in your hands, and the processing by the language model provider.

It is a contractual commitment and the record of what was carried out.

The products, one by one

What each product covers

There are two families. The code one reads your code line by line — Code, Code Online and Code Prime, and what changes between them is where the code is when the audit reads it. The server one reads the infrastructure around it — Share, Host and MultiHost, and what changes is where you look from and how many domains the server hosts.

Each one is described, with what it asks of you and how the price is set, in Products.

I want both the code and the server: is that two audits?

It is. They are two different readings, with different rulesets, and each one answers its own question — what is written in the code, and the state of the machine it runs on. Anyone who wants both answers hires both.

The targets are independent too: nothing requires the code and the server to belong to the same system. See how the price behaves across more than one audit.

ONE or DEEP: what changes in what I receive

In ONE, one agent searches and that same agent judges each finding against the code. In DEEP, several agents search in parallel and other agents judge, without having taken part in the discovery.

Deep reaches further and costs more, and it pays off where there are boundaries: separate subsystems, client and server, a binary talking to the host, code talking to the database. We measured both cases: with many boundaries Deep found 4.5× more; in a flat folder, 1.23×.

When in doubt, ask for both prices — it is the form's default option, it costs nothing and it commits you to nothing.

Does my product have both levels?

In Share the level is ONE — the target is one domain in a hosting account, and Deep pays off where there are boundaries. In the other five products you choose the level, or ask for both prices and decide afterwards.

Firmware

Firmware is a product of its own, with the C/C++ lens, and it exists only at the DEEP level — because that is where reading memory and the boundary between binary and host pays off. The price is not set per line: it comes from a matrix that weighs criticality, surface and volume.

Write to contact@svo.com.br saying what the target is, and we will put the quote together.

Before hiring

Why hire SVO, if I can just ask my own AI?

The AI is the same for everyone. What changes the result is the method it works under.

SVO developed a method that guides the AI in the search for flaws: what to examine in each kind of system, how to follow the path from outside in to each flaw, and how to document the evidence. And other agents try to knock down every finding before it reaches the report — what reaches you is what survived.

It is the same difference already visible in development: systems built with the same AI come out sounder or weaker depending on who is running the work.

When the audit runs in your environment, it uses your own Claude — and what SVO delivers is the method.

And the method is measured in public: it has already found flaws in software used by other companies, 9 of them fixed and published, 4 with CVE. Anyone can check it by the number.

Do I need a Claude plan in order to hire?

In the products that run in your environment, yes: the audit runs to the end with Claude Max, Team or Enterprise, or with prepaid API credits. In the ones that run on our infrastructure, the processing is in the price: you send the address of the code and receive the report, with any AI plan — or none.

How to install Claude Code and how to pay for the AI.

What if my system is in a language you do not audit yet?

The audit is carried out with our common method, which does not depend on the language: code execution, SQL injection, arbitrary file read and write, authentication bypass, privilege escalation and unguarded endpoints.

If more than 50% of your project is in a language we do not audit yet (when in doubt, ask), the recommendation is to request the language from our support and hire once it is available. Write to contact@svo.com.br saying which one it is: a new language is something we build, and knowing there is demand is what puts it in the queue.

What determines the size of the audit

The size is your code: the meter counts the lines of the project leaving out third-party libraries, automatically generated files and duplicated copies, item by item. In a WordPress site or plugin, for example, WordPress's own code stays out of the count — otherwise you would be paying for Automattic's work.

The measurement is free and runs on your machine. How to measure your system.

Can I exclude folders? What happens to what I exclude

You can. What you exclude is not looked at, and no problem that happened to be in there is reported — and that is written in the report, in a field of its own, separate from what was left out for being a third-party library.

The separation exists so that, when you read the limits, you can tell what was your decision and what was the method's.

Can I see a sample

You can: see a sample report. It shows the structure, the fields and how a finding is written.

And, in the products that run in your environment, you can run the sample against your own project before hiring.

What I need to have installed

In the products that run in your environment: Claude Code, on a Max, Team or Enterprise account, or with API credits. The kit arrives by e-mail and you install it on your machine — on a server, it goes in your user folder, without sudo, and the guide includes the command to remove it afterwards.

In the ones that run on our infrastructure, there is nothing to install on your side. See the installation page.

Which languages do you audit?

Today there is a dedicated lens for PHP, JavaScript, TypeScript, Vue, Python, Ruby, Go, Rust, Java, Kotlin, Scala, Clojure, Groovy, C and C++, C#/.NET and VB.NET, Delphi/Pascal, B4X and Shell — besides the WordPress lens.

The list is derived from the meter itself: when a new lens comes in, it shows up in the form on its own. And the common method, which does not depend on the language, runs on any project.

Can I audit only part of the system?

You can — the scope is yours, and the audit does not step outside it.

The practical recommendation is to create a repository or a ZIP with exactly what you want audited: a cut made in the material is more reliable than a cut described in a text field.

My site is WordPress. Does that work?

It does, and there are two paths depending on what you want to look at.

If the target is the live site, it is Share with the WordPress site analysis: every plugin and every theme is compared against the official package, file by file — that is how a tampered plugin on the latest version shows up, the kind that passes any check made by version number.

If the target is a plugin or theme you develop, it is a code product, with the WordPress lens.

My system was written with AI help. Does that change anything?

The pace changes, and that is precisely why the audit has value there. Whoever wrote the code was assembling the behaviour, and the intelligence that helped write it was on the same task: does this work?

The audit arrives afterwards with its attention turned the other way: what else does this path accept? It is the same intelligence, pointed at the opposite question.

Is this a pentest?

It is a read-only audit: it observes and documents. The path from outside in to each flaw is written out in full, with file, line and every step — and confirming it against the live system stays with your team, who can do it safely.

What the audit reaches, and with what depth, is written down before hiring in Reach.

During and after

Do you change my code?

There is no change: the audit is read-only, and what it writes is the report.

What if the audit stops halfway?

You run it again and it continues from where it stopped. Your account's usage limit, a dropped connection, a power cut — the progress is recorded, and that is what it exists for.

A secret is found: what happens to it

A password, token or key the audit finds goes into the report as a finding, with the exposure described and the value masked. What matters for the fix is knowing where it is and who reaches it — not repeating the secret in writing.

What a CVE is

A CVE is the public number of a security flaw, in the worldwide catalogue anyone can consult. The number comes from the catalogue itself, reserved by an organisation authorised by the CVE programme and independent of SVO, which ties it to the flaw. It only becomes public after the people who maintain the software publish the fix — which is why the number and the fix travel together. It is what lets you check, without asking us for anything, the software, the version and the fix behind every item on our scoreboard.

Can I ask for support about a finding?

You can. In the products that run in your environment, the report stays with you and we hold no copy — so when you ask for support about a finding, we tell you what we need you to send in order to analyse that point.

Write to contact@svo.com.br with the order number.

What do I do with the report?

It is read once to fix things, and consulted afterwards to write new code. The fields are always the same precisely for that: it becomes a reference when the next feature touches the same path.

It comes in both formats, always: HTML is for presenting, Markdown is for working.

Can I ask for a new audit after fixing?

You can, and it is the cycle we recommend: audit, fix, verify. Verification is what proves the fix worked.

For the same target the price is not linear — the third audit costs 30% of the price of one. See how the package works.

Did not find your question? Tell us what you want audited and we will answer along with the price.