A couple of weeks ago I wrote about Arkyv: regulations, project documents, and company knowledge brought inside your design tool. A hosted platform. You connect your files to it and it handles the rest.
Architecture Studio, built by Federico Negro's studio ALPA, is the same problem from the other direction. Instead of a platform you connect to, it is a set of skills, agents, and rules you install into Claude and make your own.
What It Is
The repo is called Architecture Studio. Seven agents, 37 skills, seven rules, and three hooks across nine plugins, built by ALPA and published under an MIT licence. It runs inside Claude Desktop (the chat app, no terminal required) or Claude Code (for those comfortable in the terminal). Type /studio followed by what you need — a site analysis, a space programme, a CSI spec — and it routes to the right agent or skill.
The scope is wide: site planning, zoning analysis, workplace programming, materials and FF&E research, sustainability (EPD parsing and comparison), specifications, and presentations. Several of the zoning and due diligence skills are NYC-specific, which is an obvious limitation outside New York. The pattern they demonstrate is not.
Federico Negro's background is worth noting. He co-founded CASE, a design-innovation consultancy that grew to more than 70 staff doing work for Disney, Apple, Grimshaw, SOM, and Snohetta, and was acquired by WeWork in 2015. He went on to lead WeWork's global design and construction group to over 1,000 architects and engineers across 30 countries, then founded Canoa, a platform tying design and procurement. ALPA is the current vehicle. This is not a GitHub hobby project.
The IP Question
The practical difference between Architecture Studio and a SaaS platform is who touches your project data. With a hosted product, your documents go to the platform's servers. With Architecture Studio, they do not. Your files stay on your machine. The inference runs through your existing Claude subscription or API key — your project data does not pass through ALPA's infrastructure.
That is the meaningful protection: your documents go only to Anthropic, the provider you are already in a direct relationship with, not to any third party. What Anthropic does with those inputs depends on your subscription type and settings — Claude Code and Team or Enterprise subscriptions carry stronger data protections than consumer tiers. For firms with confidentiality requirements, it is worth checking which plan you are on before deciding how sensitive the work can be.
The Learning Angle
Architecture Studio is useful as a working toolset. It is also an education in how structured AI workflows are actually built.
Every skill is a plain text SKILL.md file inside a named folder. You can read them. You can see exactly what the prompt engineering is doing, how the rules layer implements professional guardrails (code citations, CSI formatting, professional disclaimers, output transparency), and how agents orchestrate multiple skills in sequence. The whole stack is visible and legible.
A technologist or computational designer wanting to understand how to build AI workflows for a practice could learn more from reading through this repo than from most tutorials, because it is working production code built by someone with a long track record.
If You Have Done It Three Times
The most useful line in ALPA's documentation: if you've done it three times, make it a skill.
A skill is a SKILL.md file — plain text instructions that tell Claude what to do when you type the command. Read how /spec-writer works. Look at how /environmental-analysis is structured. Then ask what your office does repeatedly on every project and write the equivalent file for it. Meeting minutes in your firm's format. Submittal registers from a spec section. Fee proposals from a scope and a rate table. Pre-issue room schedule checks. The repo includes examples of each structural element you need.
The skills are open source. Contributions are open. An architect who builds a /zoning-analysis-london skill and publishes it creates something the whole community can use, inspect, and adapt. That is a different model from a platform adding a feature to its roadmap.
The Choice
Arkyv and Architecture Studio are not competing for the same user. One is a hosted product for teams who want regulations, project documents, and model queries in one place with minimal setup. The other is an open, configurable system for practitioners who want to build workflows that fit their practice rather than a platform's template.
The choice between them is also a choice about how your practice relates to its tools. One approach says: here is the tool, configure it to your needs. The other says: here is the structure, build what you need.
Both are legitimate. The question is which problem you actually have.