Architecture Studio
AI governance for architecture firms
Your team already uses AI on project work. Governance is how your firm decides what the assistant follows, what it can reach, and who signs off on what it produces.
Architecture Studio runs inside the Claude or ChatGPT account your firm already has. It gives your firm one place for its access, standards and review, and it never trains on your firm’s data. Your AI provider’s own data terms still apply to your conversations and files.
Add to Claude
Add to ChatGPTGovernance in three parts
Rules the assistant follows. These are your firm’s policies and procedures, written as files the assistant reads before it works.
Permissions that actually lock. Your AI accounts, workspaces, connected services and the access you grant decide what the assistant can reach at all.
People who sign off. Each deliverable has a named preparer, reviewer and issuer. The assistant prepares the work, and your people review it and release it.
A rule is followed; a permission locks. Good AI governance needs both, and it keeps professional responsibility with the people accountable for the project.
What Architecture Studio never sees
Your conversations, the files you attach and the connections you use stay under your provider’s terms and your workspace’s controls. Very little reaches Architecture Studio: the inputs a skill explicitly sends to a hosted tool, and bounded usage events. Your prompts, outputs and project files never do.
Architecture Studio doesn’t resell model access or sit between you and your provider. There is no second data controller and no copy of your work to protect.
- The open-source plugin stores nothing on Architecture Studio servers. Durable records stay in your studio folder.
- The hosted connection stores no project records unless your firm uses a record-holding module such as Materials Library.
- A hosted tool receives only the inputs supplied to it. It doesn’t mount or browse your workspace.
Architecture Studio never trains on your firm’s data
When your firm uses Materials Library, Architecture Studio stores the product records, revisions, provenance and evidence your firm saves. Project files, drawings, workbooks and other documents stay with the firm.
Those records are used only for the operations your firm requests. They are never used for model training, general product improvement, corpus building or cross-firm analysis. Only verified members of your firm can read them. Other firms can’t, and neither can any shared catalog.
Your AI provider’s own data terms still apply. Your conversations and files are handled under your Claude or ChatGPT plan. For professional work, use a firm-approved business, team or enterprise account, and confirm its training, retention, connector and access settings. Architecture Studio doesn’t override those settings.
Control what the assistant can reach
Access is set by the systems that already own it: your AI accounts, workspaces, connected services and the permissions you grant during authorization. Architecture Studio asks only for identity scopes (who you are, your basic profile and your email). It doesn’t request project-read, project-write or administration permissions, and it doesn’t replace your host’s security controls.
Signing in to the Architecture Studio website doesn’t connect an assistant or give anyone access to project files. You review the permissions during authorization, then manage sessions and connected access with your account and host controls.
Try it in your assistant:
Norma, tell me what you can reach in this workspace right now: which connections, which folders, which records, and which OAuth scopes Architecture Studio holds. Then tell me what stays with my provider and never reaches Architecture Studio.Write your firm’s responsible-use rules
Standards are firm policy written as instructions the assistant reads before it works: procedures, requirements, provenance rules and verification obligations. Write one the way you’d brief a new hire. Say when it applies, what must be shown, who owns it, and any named exception. Firm-wide standards live in your studio. Project-specific instructions live with the project.
Architecture Studio ships with rules for transparency, code citations, units and terminology, and your firm adds its own. Here’s the transparency rule:
Transparency. Every output must be verifiable. The user should never have to trust a number.
- Never present a derived number without the inputs.
- Never present a recommendation without the reasoning.
- Every data point from an external source gets a link, not just a name.
The assistant keeps sources visible, respects approval boundaries, and reports what it couldn’t verify instead of quietly relaxing a requirement.
Keep professional judgment with your people
Assurance means the work is checked before it leaves the studio. For each deliverable, your firm sets:
- Named responsibilities: who prepares, who reviews the evidence, and who authorizes issue.
- Evidence thresholds: what counts as a source, what a reviewer checks, and what must be flagged.
- Exceptions: who may approve one, and how missing information and conflicting sources are handled.
- Recorded approvals: each one kept as a dated decision.
The assistant prepares the work and reports against your acceptance criteria. Your named people review it and release it.
What we record, and how to turn it off
The hosted connection records usage data by default: which Architecture Studio tools, skills and references are used, and when. It’s linked to your account by a pseudonymous ID and to your firm by its organization ID, so it isn’t anonymous. It never includes your prompts, conversations, tool inputs, project files, generated outputs or access tokens.
To turn it off for your account, tell your assistant “turn off usage data” while Architecture Studio is connected. A firm admin can say “turn off usage data for my firm”. While it’s off, nothing is recorded. In the open-source plugin, set AS_TELEMETRY_DISABLED=1.
Teach your team to check the work
Governance only holds if people know how to work with the assistant. Learn with AS is a self-paced course of six guided exercises on a fictional architecture project, so a mistake costs nothing real. It runs in Codex or Claude Code. It covers working with the assistant, authorization and project guidelines, then planning work and building a reusable skill.
Open source you can inspect
The Architecture Studio plugin is open source under the MIT license. Your IT lead can read exactly what it does, report issues or fork it.
Setting AI strategy for the whole firm?
Governance is one part of making AI a firm capability. The rest is deciding which work AI should do, what to buy and what to own, and where your firm’s knowledge should live. ALPA, the studio that builds Architecture Studio, offers AI strategy and adoption help for architecture firms, from ALPA.
FAQ
Does Architecture Studio train on our firm’s data?
No. Architecture Studio never trains on your firm’s data. Stored module records are used only for the operations your firm requests, never for model training, product improvement, corpus building or cross-firm analysis. Your AI provider’s own data terms still apply to your conversations, so check your plan’s data controls.
Does Architecture Studio see our prompts or project files?
No. Prompts, outputs and project files stay with your Claude or ChatGPT account. What reaches Architecture Studio is the inputs a skill explicitly sends to a hosted tool, plus usage events that never include prompts, files or outputs.
Who controls what the assistant can access?
Your firm does, through your AI accounts, workspaces, connected services and the permissions you grant. Architecture Studio asks only for identity scopes. Authentication and permissions →
Can we export our records?
Yes. Your firm has a right to a full portable export of its module records, including all revisions, provenance and saved evidence. Email help@architecturestudio.ai to request one. Library CSV exports are also available in the signed-in library. Storage and portability →
Do our AI provider’s terms still apply?
Yes. Architecture Studio doesn’t override your provider’s training, retention, connector or access settings. For professional work, use a firm-approved business, team or enterprise account.
Can we write our own AI rules?
Yes. Add your firm’s standards as files the assistant reads before working, and keep project-specific instructions with each project. How standards work →
We need help with AI strategy, not just governance.
See AI strategy and adoption help for architecture firms, from ALPA.
Start with one task.
Connect Architecture Studio to the Claude or ChatGPT account your firm already uses, then ask what it can reach.