Short answer
#The whole tree, written out
Printed in full so the recommendation can be checked rather than trusted — including by somebody who never answers a question.
What are you actually trying to do?
Pick the one that is closest. There is a way back at every step.
-
Understand a system I did not build
— Inherited, legacy, or written by people who have gone.
→ next question -
Hand a system over, or take one on
— A change of team, agency or technical leader.
→ next question -
Keep documentation current as the system changes
— It exists and it goes stale.
→ next question -
Find security problems in my system
— Vulnerabilities, exposure, compliance.
→ A security scanner
Will the answer be needed again after today?
Be honest about this one. It is the question that decides most of the outcomes here.
-
No — I need one answer, once
— A specific question about a specific thing, and then it is finished.
→ Use a coding agent on its own -
Yes — other people will need it too
— A new team, a colleague, an investor, a future me.
→ next question
Is the problem that nobody writes it, or that nobody trusts it?
-
Nobody writes it — we have no place to put documentation
— The tooling is the gap.
→ A documentation tool -
We write it and it is wrong six months later
— The maintenance is the gap.
→ next question -
We want diagrams that stay accurate
— Architecture drawings, C4, sequence diagrams.
→ An architecture or diagramming tool
Does it matter whether a claim can be checked?
That is: could somebody who was not there open the file a statement came from and verify it?
-
Yes — people will act on this and need to trust it
— A handover, a due diligence, a new team taking responsibility.
→ next question -
Not really — a good summary is enough
— Orientation rather than evidence.
→ A coding agent, and write down what it says
Can you give a supplier access to your repository?
-
No, or it would be a long conversation
— Policy, client confidentiality, or simply not worth the argument.
→ next question -
Yes, that is not a problem
— Repository access to a vendor is routine here.
→ next question
Do you need to see what changes between one look and the next?
What arrived, what moved, and which earlier statements no longer hold.
-
Yes — the system keeps changing and that is the problem
— Ongoing, not a one-off.
→ 1ADK -
No — a single point-in-time picture is what I need
— One event: a sale, a handover, a diligence.
→ next question
Do you want a person to interpret it and make recommendations?
-
Yes — I want judgement, not just facts
— Somebody to tell me what to do about it.
→ A consultant, or an audit -
No — I want the facts, and I will decide
— The picture, checkable, in a form the company keeps.
→ 1ADK, used once
#The nine endings, in full
Use a coding agent on its own
You do not need a product for this. One answer, once, for you: open Claude Code, Codex or whichever agent you use, point it at the repository and ask. It will do a better job faster than any process involving a subscription, and the answer will be in front of you in minutes.
What you will not get: a record that survives the conversation, a way to tell which statements were read and which were inferred, or anything to compare against next quarter. If none of those matters, none of those matters.
A coding agent, and write down what it says
An agent plus a document is probably enough. You need orientation for other people rather than evidence. Ask an agent for a structured summary, paste it somewhere the team can find it, and put a date on it. That is a real solution and it costs nothing.
It will go out of date silently, and nobody will be able to tell which parts. If that starts to matter — because somebody is acting on it — the answer changes.
A documentation tool
You need somewhere to write, not something to derive. Confluence, Notion, GitBook, a docs folder in the repository — any of them. The gap you described is that there is no home for written knowledge, and that is a tooling problem with well-understood solutions.
1ADK does not solve this and does not try to: nobody types pages into it. If in a year the problem has become "we write it and it is wrong six months later", come back to this page and answer that question differently.
An architecture or diagramming tool
You want diagrams, which is a different thing. Structurizr, IcePanel, Miro, C4 tooling, or diagrams generated from code. If the deliverable is a picture that people look at in meetings, that is what these are for and they are good at it.
A diagram is drawn from somebody's understanding, so it is wrong exactly when their understanding is. If the underlying problem is that nobody is sure what is actually there, the diagram will inherit that.
A security scanner
This is not what 1ADK does. You want SAST, dependency scanning, secret scanning or a penetration test, depending on which part of the problem you have. Snyk, GitHub Advanced Security, Semgrep, or an engagement with a security firm.
1ADK does not test for vulnerabilities and does not claim your system is safe. It can tell you what your system is made of and what it talks to, which is useful input to a security exercise and is not one.
A consultant, or an audit
You want judgement, and that is a person. A one-off technical due diligence or architecture review, delivered by somebody experienced who will read the system and tell you what they think. For a single event — a sale, an investment, a decision about a rewrite — that is often exactly right.
It is a snapshot, it is expensive, and it begins going out of date on the day it is delivered. Several firms will do this well; ask them for the evidence behind their conclusions rather than only the conclusions.
1ADK, used once
The free plan is probably all you need. You want a checkable picture for a specific event, not an ongoing subscription. Free covers one project, the first analysis, one further analysis, and the difference between them — which is enough for a handover or a diligence.
After the second analysis the project stays readable and updates pause. Nothing is deleted and no card is involved at any point.
1ADK
This is what the product is for. A system that keeps changing, an answer that has to outlast the person who asked it, and claims somebody else can check. 1ADK derives the picture from an analysis your own agent runs, records the evidence behind each statement, keeps every version, and reports what changed — including which of its own earlier claims it can no longer stand behind.
It does not judge code quality, does not find vulnerabilities, does not recover why decisions were made, and does not replace written documentation of intent. Start free: one project, two analyses, and the difference between them.
#Why a vendor publishes this
Two reasons, and neither is generosity.
The first is commercial. A customer who arrives needing a security scanner and signs up for a technical memory is a refund, a bad month and somebody telling their network about it. Sorting that out before the sale is cheaper than after.
The second is that a page willing to say "not us" is the only kind whose "us" carries any information. Every vendor claims their product fits; the claim is worth exactly nothing unless there is a visible case where it does not. Six of the nine endings here are somebody else's product, and they are genuinely the right answer for those situations.
#What this does not do
What this does not do
- It does not know your situation. Seven questions cannot capture a company, and the tree is deliberately coarse.
- It does not compare specific products on features or price. Where it names alternatives, those are examples of a category rather than a recommendation of a vendor.
- It does not cover every category — nothing here about observability, testing, or project management, because those solve different problems.
- The answer is often “both”. A wiki for intent and a derived picture for facts is a common correct outcome that a single-ending tree cannot express.
Questions people actually ask
Six of the nine endings are not 1ADK, and the whole tree is printed further down this page so you can read it without answering anything. Check it against what a selector that was rigged would look like.
Because the alternative is worse for everybody. A customer who arrives needing a security scanner and is sold a technical memory is a refund and a bad reputation. And a page willing to say "not us" is the only kind whose "us" means anything.
They often do, and the answer is usually both — a wiki for intent and a derived picture for facts, or a consultant for judgement and a picture for the facts they work from. Run the tree twice with different first answers and see where each ends up.