NoGuard.AI connects browser chat with a Windows desktop workstation for reverse engineering. Both use one account and a prepaid credit balance. The terminal client is coming soon; its CLI guide previews that workflow.
Start with the interface that can reach the evidence for your task. A question about a code snippet can begin in the browser. A binary that needs local tools, a debugger, and a record of tool activity belongs in the desktop workstation. This guide walks through that choice, the credit model, and a way to make an AI-assisted analysis easier to verify.
Start a conversation in the browser
Create an account, verify your email, and check the free-credit offer shown at signup. The web chat saves conversations to your account so you can continue later. Some model choices require a credit-pack purchase; the model selector and account state determine what you can use.
For the first request, choose a small question with a clear finish. Include the material you want examined and describe the output you need. For example:
“Explain this function from a program I maintain. Identify its inputs, outputs, and error paths. Point to the relevant lines, separate observations from guesses, and list two tests that would confirm your interpretation.”
This is an example prompt, not a sample result or a promise about accuracy. It gives you something to check: the answer should refer to the supplied function, distinguish uncertain claims, and propose tests that actually exercise it.
If the answer depends on material you have not supplied, ask what is missing before expanding the task. A precise follow-up with the relevant excerpt is easier to assess than an explanation built around an assumed implementation.
Choose the desktop app for local reverse engineering
The Windows desktop app places AI agents beside hex, disassembly, graph, and debugger views. Its event journal records work in the project, and it can integrate with tools such as IDA Pro and Ghidra. A credit-pack purchase unlocks the desktop download; see the app page for current requirements and access.
An initial reverse-engineering question should identify a specific target and outcome. For a program you own or are authorized to analyze, a useful starting task is to explain one file header, one parsing function, or one difference between two versions. Avoid treating a broad request to “explain this binary” as a finished specification.
For example, ask:
“Inspect how this program reads its file header. Record the byte offsets and field widths you can establish, identify the branches that reject invalid input, and separate direct evidence from inferred field names. Propose a small test that distinguishes the remaining interpretations.”
Use a bounded review loop:
- Name the target. Record the file or version and the question you want answered so later observations refer to the same material.
- Gather evidence. Ask for the offsets, code locations, strings, or tool observations supporting each claim. Check them in the relevant view.
- Review uncertainty. Keep inferred types, guessed names, and untested behavior distinct from observations.
- Check a prediction. Use a test input or another independent observation to see whether the proposed explanation holds.
- Keep the outcome. Summarize the supported findings, unresolved questions, and next action with the project’s record of work.
The app’s tool views and journal support that process; they do not make an untested model statement correct. Review proposed tool actions and changes through the available approval controls, and inspect the resulting work before using it.
A small header example you can check
Here is an explicitly synthetic example, unrelated to a customer file or a claimed NoGuard test result. Define a toy format with four ASCII magic bytes, a two-byte little-endian version, a two-byte little-endian payload length, and then the payload. For this exercise, version 1 is the only supported version and the file must end immediately after the payload.
The complete file is these 12 bytes: 4e 47 52 44 01 00 04 00 78 56 34 12.
- Offsets 0–3 contain the magic text
NGRD. - Offsets 4–5 hold version 1.
- Offsets 6–7 declare a four-byte payload.
- Offsets 8–11 supply exactly four payload bytes. Their meaning is unspecified.
An analysis should distinguish what the exercise’s format definition supplies from what the bytes show. Those bytes alone do not prove that the payload is a 32-bit number or that a real program interprets it that way.
Ask the AI for tests, then check them against the definition: removing the final byte must fail the length check; changing the version to 2 must fail the version check; appending a byte must fail the exact-length rule. The original sample should pass all three rules. The useful result is an explanation whose predictions you can verify, rather than plausible names for unknown fields.
Understand credits before expanding the task
NoGuard uses prepaid credit packs rather than a recurring subscription. Purchased credits do not expire. Web chat and the desktop app share the same balance, with the upcoming CLI intended to join them. Current pack amounts, bonuses, and payment options are listed on the pricing page.
Replies are metered by tokens through the selected model’s pricing. A longer input, more conversation context, a longer answer, or additional model iterations can change the cost. A credit pack is therefore not a fixed number of messages, and an example task does not establish a general price for reverse engineering.
Begin with one question, inspect its usage, and expand when you understand the workflow and the amount of context it needs. Use the account’s usage and billing views to review recorded spending. The FAQ’s credits section explains the unit and the shared balance.
Check what happens to your data
The desktop project and local tools operate on your computer, while AI requests go through the NoGuard gateway to the configured providers. Local project storage does not mean every model request stays on the device. Web conversations are saved to your account.
Read the privacy policy for processing, storage, staff access, and account controls before choosing what to send. The terms and refund policy describe the service and purchase conditions.
Decide the next step
Use the browser for a focused question you can evaluate from supplied text. Move to the desktop app when your task needs local target evidence and tool activity in one workspace. For a comparison of payment models and ways of working, read NoGuard.AI vs ChatGPT.
You can start with NoGuard chat, or review the desktop workflow and download requirements first. Keep the first task small enough that you can tell whether the result is useful.