Claude CoWork — Frequently Asked Questions

Claude CoWork is a desktop app that runs Claude AI locally on your Mac or Windows machine. Unlike Claude Chat (which runs in a browser and has no persistent memory), CoWork:

• Reads and writes files directly on your computer — no cloud upload limits
• Remembers your preferences and context across sessions via a memory.md file
• Returns ready-to-use files in your local folders, not just text in a chat window
• Connects to Outlook, Confluence, and Slack via approved connectors

Think of it as Claude that lives on your machine, knows your context, and works in your files.

Step 1: Download and install the Claude desktop app from BRG Company Portal (Mac and Windows supported).

Step 2: Open the app and sign in with your BRG login crediential.

Step 3: Create a local folder in your Documents — name it something like CoWork or AI Workspace.

Step 4: In CoWork, click "Choose a folder" and select the folder you created. This is your workspace.

Step 5: Add a claude.md file to that folder (your personal instructions to Claude) and a memory.md file (where Claude stores notes about your preferences).

In CoWork, a "Project" is a local folder on your machine. CoWork is sandboxed — it can only see files inside the folder you've pointed it to. Each user's workspace is separate.

This means:

• Your CoWork setup doesn't automatically share with anyone else
• There's no live collaboration or co-editing across machines in CoWork
• If you want to share output, export the file manually (or put it on OneDrive/SharePoint)

If real-time sharing is your primary need, CoWork may not be the right tool — raise this with your Workstream Lead and check if ChatGPT Workspace Agents (cloud-native, currently under security review) might be a better fit.

CoworkOS is a setup pattern developed by Vivid Savitri that turns CoWork into a personalised AI operating system with persistent memory and structured workstations. It's built on top of standard CoWork using markdown files.

The pattern uses a single HTML template you copy and customise. Each "workstation" (e.g., Email HQ, Project Hub) has four parts: identity, resources, workflow, editorial role.

Do you need it? Not immediately. Start with basic CoWork first. Once you find yourself repeating the same prompts 2–3 times a week, that's the signal to set up a workstation or skill.

This might happen due to BRG browser/network policy setting, not necessarily CoWork bug.

What to do:

• Check the "What's Allowed at BRG" page — it lists which connectors are approved and any steps needed to enable them
• If a connector you need isn't on the approved list, raise it via the AI tool evaluation form or the AI Slack channel
• Contact your Workstream Lead if you're unsure — they're the first port of call for function-specific connector questions

No, this is by design. The Outlook connector in CoWork is currently read-only. CoWork can read your inbox and calendar but cannot compose or send emails on your behalf.

The workaround most users adopt:

• Ask CoWork to draft the email as a text file or copy it to your clipboard
• Paste manually into Outlook and send from there

CoWork cannot connect to SharePoint directly. The supported workaround is to use OneDrive sync:

• Sync your SharePoint library to your local machine via OneDrive (right-click the SharePoint library in your browser → "Sync")
• The synced folder appears in your Windows Explorer / Finder as a local folder
• Add that folder to your CoWork workspace, or copy files from it into your CoWork folder
• Changes you save back to the OneDrive-synced folder will sync to SharePoint automatically

Claude CoWork is on BRG's approved tool list. Anthropic's enterprise data commitments include:

• Conversations and files are not used to train Anthropic's models
• Data is encrypted in transit and at rest
• Enterprise accounts are isolated from consumer accounts

General guidelines for what's appropriate:

• ✓ Internal documents, drafts, reports, analysis — fine
• ✓ Meeting notes, presentation prep, email drafting — fine
• ~ Commercially sensitive strategy documents — use judgement; check with your manager if unsure
• ✗ Customer personal data (PII) — do not paste raw PII into any AI tool without explicit guidance from Privacy team

AI-generated images that depict real people require release forms from those individuals. This applies to any AI image generation tool, not just CoWork.

Before using AI-generated images of real people in any production content:

• Consult Annabela Lima (Privacy/Security) and Madison Sertic (Privacy)
• Confirm: Is the depicted person real or synthetic? Were they consented? Where was the source image from?

This applies to content generation tools broadly (Nano Banana 2, GPT Image 2, etc.).

No. BRG operates under an enterprise Anthropic agreement. The agreement explicitly covers:

• BRG will not use customer personal data to train external AI models without explicit consent
• Data you share in CoWork is not used to train Anthropic's models

This is consistent with BRG's AI Governance Statement (under Board review since Oct 2025).

Outcome-first prompting means starting with the finished result you want, not a step-by-step instruction set. Instead of "write me a brief about X, then format it, then add a summary," you say: "I need a two-page brief on X saved as brief-X.docx in my Documents folder."

CoWork is designed for this. Because it has access to your local files and persistent memory, it can carry out multi-step tasks and produce files directly — not just give you text to copy out of a chat window.

This is one of CoWork's strongest differentiators. Because it accesses local files directly (no browser upload limit), you can:

• Point CoWork at a folder of PDFs and ask it to synthesise across all of them
• Work with large Excel files without hitting browser upload caps
• Process multiple documents in a single session

Not automatically — CoWork is sandboxed per machine. But you can share the setup files manually:

• Share your claude.md and memory.md templates via Teams/SharePoint

CoWork gives you model options within the app. Vivid's guidance (May 2026):

• Claude Opus — Use for planning and complex reasoning only (it's expensive; ~5× the cost of standard)
• Claude Sonnet — Use for execution, drafting, and day-to-day tasks. Best balance of quality and cost.
• Claude Haiku — Use for fast, repetitive tasks (e.g., daily briefings, formatting) where speed matters more than depth

A Skill is a reusable, multi-step workflow you can trigger with a single click or phrase — essentially a saved prompt sequence. Example skills from the pilot:

• Daily briefing: summarise my Outlook inbox and calendar, prioritise by current week's focus
• Meeting prep: pull the agenda, read relevant Confluence pages, draft talking points
• Weekly report: pull data from a folder, synthesise into a formatted report

When to build a Skill: When you find yourself doing the same task 2–3+ times per week. Don't build Skills prematurely — understand the workflow through iteration first.

Follow this path — in order:

• Anthropic Academy — anthropic.com/learn — free, self-paced CoWork training
• Use-case library — in Teams/SharePoint; built from Finance and Service Delivery pilot learnings. Add your own use cases as you develop them — the library grows through community contributions.
• Community channel — search first; post your own learnings. This is the living knowledge base for CoWork at BRG.
• CoworkOS setup guide — covers setup, workstations, and Skills