Organisation deployment guide

Do this first, so there is somewhere for the next step to put things. Create two empty private repositories, named exactly:

AI-marketing-studio-code
AI-marketing-studio-data

On Azure DevOps: open your project, then Repos in the left sidebar, then the repository dropdown at the top, then New repository. Set Repository name, leave Add a README unticked, and click Create. Repeat for the second name.

On GitHub: click + (top right), then New repository. Set Repository name, choose Private, leave Add a README file unticked, and click Create repository. Repeat for the second name.

Now find each repository's clone address. You use them in Step A2 in a minute, and again every time somebody new joins.

Access: everyone who will use the Studio needs write access to AI-marketing-studio-data and read access to AI-marketing-studio-code.

The whole thing is one folder of files on GitHub. You do not need a GitHub account, and nothing has to be installed to download it.

1. Open github.com/abraun64/soundtrak-ai-studio-team.
2. Click the green < > Code button, top right of the file list, then Download ZIP.
3. The ZIP lands in your Downloads folder. Right-click it, choose Extract All, and extract into your own user folder, for example C:\Users\jsmith.
4. Rename the extracted folder to exactly:

   deploy

   so you end up with C:\Users\jsmith\deploy and the paths below match.

Open deploy. You should see exactly two folders, side by side:

deploy\code
deploy\data

Open code and check you can see README.md, index.html, and folders called docs and craft.

python .claude/lib/build_seed.py --out $HOME\deploy --profile team

This is the step that puts everything where your whole team can reach it. You are turning each folder into a repository and sending it up.

The code folder first. Open PowerShell in deploy and paste this whole block. Replace <code clone address> with the first address from Step A1:

cd $HOME\deploy\code
git init
git add -A
git commit -m "Initial version"
git branch -M main
git tag -a v1.0.0 -m "Version 1.0.0"
git remote add origin <code clone address>
git push -u origin main
git push --tags

Then the data folder, the same way, with the second address. Note there is no version label here: labels apply to the Studio, not to your work.

cd $HOME\deploy\data
git init
git add -A
git commit -m "Initial version"
git branch -M main
git remote add origin <data clone address>
git push -u origin main
cd $HOME

You will be asked to sign in to your host the first time. Refresh both repository pages in your browser afterwards: where they were empty, you should now see files.

That is where everything lives from now on. Every person who joins gets their own copy of these two, and their work flows back here. You can now delete the deploy folder.

Campaigns produce images and video. Those are stored using Git LFS, which the data repository is already set up to use from its first upload. You just need it switched on at the host, and installed on each machine.

Azure DevOps: LFS is on by default. Nothing to do.

GitHub: LFS is on by default for the repository. Check your plan's included storage and bandwidth under Settings then Billing and licensing.

On each machine (yours and everyone else's), Git LFS has to be installed once. Run:

git lfs install

If that says the command is not recognised, install it from git-lfs.com and run it again. Part B's health check tests for this, so people will be told if they have missed it.

Colleagues without Claude Code still need to see dashboards and finished work. Everyone who uses the Studio publishes to the same SharePoint library, so the pages stay current whoever happens to be working, rather than going stale because one person is away.

First, create the library if you do not already have one. In SharePoint, go to the team site your marketing people can reach, choose New → Document library, and name it something obvious such as Marketing Studio. If a suitable library already exists, use that instead. Nothing to create, nothing to sync: this step is the one people skip, and then go looking for a folder that was never made.

Then put the library on each person's machine. Open it in a browser and click the one button in the toolbar that puts it on their computer. Depending on how new your tenant is that button says either Sync or Add shortcut to OneDrive — both do the same job here.

SharePoint lives on the web and the Studio writes to folders on a computer; this is what bridges the two. If neither button is there, syncing is switched off by policy and IT has to turn it on.

The library then appears in File Explorer as an ordinary folder, named after itself. It is usually under your organisation's name or your work OneDrive, but it can be anywhere — people often move their OneDrive somewhere shorter like C:\claude, and the folder follows.

Open it and copy the address from the bar at the top — that is the path you need below. Publishing writes a campaigns folder and an index page, so point it at a subfolder such as AI Studio dashboards rather than the top of a library other people use.

Then each person tells the Studio where to publish. In PowerShell, once per machine, replacing the path with the one you just copied:

[Environment]::SetEnvironmentVariable("MAS_PUBLISH_DIR", "C:\Users\jsmith\Contoso\Marketing Studio - Documents", "User")

Close PowerShell and open a new one, or the change will not be visible yet. Then, from the code folder:

python .claude/lib/publish_surfaces.py

It prints how many pages it copied.

The normal OneDrive sync does the upload. No app registration and no admin consent needed. It only ever writes outward, so nothing anyone does in SharePoint can affect the real work.

The Studio keeps everyone's pages up to date automatically using a Claude Code feature called hooks. Some organisations apply a policy that blocks these.

Have one person do Part B on a real, managed company laptop and read their health check. It tests this directly and says so in plain language.

Deploy these through Intune, Company Portal, or however you normally distribute software. Every one is in the public winget repository, so no packaging work is needed.

A ready-made configuration file is in the code repository at deploy/studio-prerequisites.winget.yaml. To try it on one machine first:

winget configure .\deploy\studio-prerequisites.winget.yaml

Part A is done. Everything after this is per-person, and you send them the same three things every time. Here is the whole message:

Before you send it, confirm that person has write access to AI-marketing-studio-data and read access to AI-marketing-studio-code. Everything else in Part B is self-service.

Your IT team installs these for you. Open Company Portal from the Start menu and check for them, or just ask IT whether they have been pushed to your machine.

Then open Claude Code once, sign in when it opens your browser, type /model and choose Opus. Use the desktop app, not the browser version: the browser one runs in the cloud and cannot open folders on your computer.

Make an empty folder for the Studio. C:\Studio is a good choice, and it must not be inside OneDrive.

Now open that folder in Claude Code:

1. Open Claude (the desktop app) and click the Code tab at the top.
2. Choose Local, which means "run on my own computer", then click Select folder.
3. Find your Studio folder, click it once to highlight it, then click Open. That picks the folder itself; you do not need to go inside it first.
4. If it asks whether to trust the files in this folder, say yes. It is an empty folder you just made yourself.

Then type this, pasting in the two addresses you were given:

Set up my studio. My code address is <code address>
and my data address is <data address>.

That is the whole of your setup. It is an ordinary sentence, not a command, and you cannot break anything by wording it differently. Claude does the rest and narrates as it goes:

• fetches the Studio and your organisation's work;
• shows you the licence once and asks you to reply I AGREE;
• installs the pieces it needs, including the extras for image and video work;
• records who you are, so your approvals carry your name;
• runs a health check and explains the result.

Claude finishes by running a health check and explaining the result in plain English. You are looking for READY.

If anything is wrong it will say what, and whether it can fix it. Two things it cannot fix on your behalf, because they are not yours to fix:

• a missing app from Step B1, which IT installs;
• access not granted yet on one of the two addresses, which is whoever set up the organisation.

You can re-run it whenever you like by saying "run the health check".

Every message the check can print, blocker or not, is listed in If something looks wrong at the end of this page.

Setup created two folders inside the one you chose: code and data. From now on you always work in code, and this is a one-time switch.

Click + New to start a fresh chat, then Local → Select folder again, and this time choose Studio\code. Say yes if it asks about trusting the folder.

Ask for your home page:

open my studio home

It is one page linking to your campaign dashboard, your system dashboard and the guides. Bookmark it. That single bookmark is your way back into everything, and the pages behind it update themselves as you work, so you only ever refresh.

Then say what you want:

what's awaiting me

Your work saves and reaches your colleagues automatically at the end of every session, with no action from you.

The Operator guide takes it from here: how a campaign actually runs, stage by stage.

All optional. The more you bring, the sharper it starts, but nothing here blocks you and you can add the rest later.

• Your website address: the single most useful source. The Studio reads it while it interviews you.
• Brand style guide, or just your logo, colours and fonts.
• Tone of voice: how you sound, words you use and words you avoid.
• Positioning and segments: what you stand for, and who you sell to.
• Rules you must follow: only if you are regulated (finance, health, government). Disclaimers, and things you cannot claim.
• Where you publish, and your martech: key channels plus tools (CRM, email, analytics). If any needs a password or key, the Studio shows you where to put it safely. Never type one into a file in the shared folder.
• Best-practice examples: a few campaigns or pieces you admire, for it to learn from.

In Claude Code, press + New for a fresh chat, point it at your code folder, and name the chat "Brand setup". You will come back to it.

Then type:

Onboard <your business name>

The Studio interviews you a few questions at a time and builds the foundation. Point it at your website and any public material and it uses them as source while it asks. Anything missing is marked to fill in later rather than blocking you.

If the organisation runs more than one brand, repeat Step C2 for each:

Onboard <sister brand name>

Each brand is independent: its own voice, segments, and compliance rules. A campaign belongs to exactly one of them, so work never bleeds between brands by accident.

The foundation reaches your colleagues the same way everything else does, at the end of the session. From then on, any of them can open Claude Code and say:

start a campaign: <what it is for>

People who join later never repeat Part C. They clone, provision, and the brand is already there.

Same download as Step A2, then from inside your code folder, one line at a time. Replace v1.2.0 with the version number you are publishing:

git add -A
git commit -m "Update to v1.2.0"
git tag -a v1.2.0 -m "v1.2.0"
git push
git push origin v1.2.0

Then tell your team it is available. Nothing notifies them automatically, by design: an upgrade that arrives unannounced is an upgrade that arrives mid-campaign.

In your code folder, say:

check for studio updates

You will be told which version you are on, whether a newer one exists, and what changed in it. Then, when you are ready:

update the studio

If you change your mind, say "roll back the studio" and you are returned to the version you were on before. That is the whole of it.