Evidence in the shape assessors expect
Each command drafts one governance document from a template built around the standard it answers to, using what's already in your project: principles, stakeholders, requirements, risks and decisions. You review it, fix it and sign it off.
/arckit:mod-secure./arckit:atrs)./arckit:dos).That's eight of 75 commands. The rest cover requirements, architecture decisions, diagrams, Wardley maps, traceability and delivery. See all commands.
AI you can defend in a review
The fair question about AI-written governance documents is whether anyone can check them. ArcKit is built so that someone can.
- Claims cite their sources. When a document draws on your policies, a vendor's submission or a web page, the claim carries a marker that points to the source.
- Every document records how it was made. A block at the end records the command, the AI model and the settings that produced it.
- Documents start as drafts. Approval is a human act, and ArcKit doesn't pretend otherwise.
- It's honest about its limits. One page lists which rules are enforced in code whatever the AI does, which depend on the AI following instructions, and which are yours to own. Read the enforcement page.
The questions your security and commercial colleagues will ask
Does ArcKit send our documents anywhere?
ArcKit itself sends nothing and collects no usage data. It's a set of commands, templates and checks that run inside your AI assistant, and every document it writes stays in your repository.
Your AI assistant does send your prompts and the files it reads to its model provider, under whatever agreement your organisation has with that provider. That's the same with or without ArcKit, so the question to settle is which assistant you're approved to use. Research commands also search the web and public sources such as AWS and Microsoft documentation, UK government code and published tenders. Don't run them against material you can't search on.
Can we use it on OFFICIAL-SENSITIVE work?
That's your organisation's call, and it depends on which AI assistant and model provider you're approved to use at that classification. ArcKit labels every document with a classification in its control table, taken from the standard the document answers to or from a default you set. In Claude Code, automatic checks also stop keys, passwords and tokens from being sent in prompts or written into documents, and stop the AI writing to credential and key files.
What ArcKit doesn't do is control where a classified file goes afterwards. Repository visibility, publishing and how long your assistant keeps conversation history are yours to set.
If your rules are unclear, run the pilot on one of the public example projects first. Most teams find it answers the question for them.
Do we need to procure it?
No. ArcKit is MIT licensed and free, with no supplier, subscription or contract. You install it as a plugin or extension in an AI coding assistant. If your team already has one approved, there's nothing new to buy. If not, that assistant is the thing to get approved, not ArcKit.
We have Microsoft 365 Copilot or ChatGPT. Does that work?
Not directly. ArcKit runs in AI coding assistants, which read and write files in a repository. Microsoft 365 Copilot and ChatGPT are chat assistants, and they don't. What you'd get from ArcKit over a plain prompt is the template, the citations, the traceability between documents, and a record of how each one was made. That's the difference between plausible prose and evidence that survives a review.
We only have GitHub Copilot. Does that work?
Yes. ArcKit ships as prompt files and custom agents for GitHub Copilot. Claude Code gets the most features, including the automatic checks and research agents, but the templates and commands are the same everywhere. Getting started covers each assistant.
We already have Bizzdesign, Ardoq or LeanIX. Why would we need this?
Keep it. An EA repository holds your inventory: what applications you have and how they connect. ArcKit produces the documents an assessment, design authority or procurement asks for. Put an export from your EA tool in the project and ArcKit uses it as a cited source, so the two work together rather than competing.
Who maintains it, and will it still be here next year?
ArcKit is built in the open, with more than 100 releases since October 2025, and practitioners in several countries contribute overlays and fixes. Because everything it produces is plain Markdown in your repository, your documents don't depend on ArcKit continuing to exist.
Try it on a real project, with help
This quarter we're running free two-hour pilot workshops. We work with your team on one of your projects (or a public example project) and finish with a first set of governance documents for the assessment you have coming up. There are also monthly office hours, open to anyone.
If you already use ArcKit, please tell us. You can be named, listed anonymously by sector, or not listed at all.
Tell us you use ArcKit Get startedCan't post publicly? Send the same details on Discord or through the LinkedIn group. See who's using it on the adopters page.