Build Great Things/ Docs/ Get started/ Set up your workspace

Setting up your Build Great Things workspace

Connect your organization's AI tools and get your first task map. Setup is configuration only: there are no agents, proxies, or extensions to install.

Who this is for: the Build Great Things workspace owner, plus your Claude organization owner. Time required: about 30 minutes, then the map builds automatically.

1. Create your workspace

  1. Go to buildgreatthings.ai and sign in with Google, or create an account with email.
  2. Enter your organization's name to create the workspace. You become its owner.

2. Set your privacy threshold

The privacy threshold is the minimum number of people who must share a pattern before it appears anywhere. Nothing below the threshold is ever shown, so no insight is traceable to one person.

  1. Open Settings.
  2. Under Privacy, choose your threshold. The default is 5 people.
Note Prompt text is buffered only until the next map run, then processed into aggregate patterns and the raw prompts are deleted. Only aggregate patterns are kept, never raw prompts, and identity is stored only as a salted hash.

3. Create an ingest token

Your AI tools authenticate to the workspace with a per-organization ingest token. The same token works for every tool you connect.

  1. Open Connections.
  2. Under Set up the OpenTelemetry stream, click Generate ingest token.
  3. Copy the token now. It is shown once; you can revoke it and issue a new one at any time.

4. Connect Claude Code for your whole organization

Your Claude organization owner turns on telemetry once, centrally, with server-managed settings. The settings are delivered to every member's Claude Code automatically; individual users do not configure anything and cannot override them.

  1. Sign in to claude.ai as the organization's Owner or Primary Owner (Claude for Teams or Enterprise).
  2. Go to Admin settings > Claude Code > Managed settings.
  3. Add this env block to the managed settings JSON, pasting your ingest token from step 3, and save:
{
  "env": {
    "CLAUDE_CODE_ENABLE_TELEMETRY": "1",
    "OTEL_METRICS_EXPORTER": "otlp",
    "OTEL_LOGS_EXPORTER": "otlp",
    "OTEL_LOG_USER_PROMPTS": "1",
    "OTEL_EXPORTER_OTLP_PROTOCOL": "http/protobuf",
    "OTEL_EXPORTER_OTLP_METRICS_TEMPORALITY_PREFERENCE": "delta",
    "OTEL_EXPORTER_OTLP_ENDPOINT": "https://buildgreatthings.ai/otlp",
    "OTEL_EXPORTER_OTLP_HEADERS": "Authorization=Bearer <your-ingest-token>"
  }
}

Claude Code fetches managed settings at startup and rechecks hourly. Ask people to restart Claude Code once; on first delivery each person accepts a one-time security prompt confirming the new organization settings.

Note OTEL_LOG_USER_PROMPTS controls whether prompt text is included in the stream. Usage and cost work without it; the task map needs it. Because this is an organization-wide setting, agree on it with your team before rollout.

5. Connect Claude Cowork (desktop app)

If your team works in the Claude desktop app (Cowork) rather than the Claude Code CLI, telemetry is configured in a different place and needs one extra step per computer. Cowork ignores the Claude Code managed settings from step 4.

  1. Sign in to claude.ai as the organization's Owner or Primary Owner.
  2. Go to Admin settings > Data and privacy > Monitoring, under Cowork.
  3. Set OTLP endpoint to https://buildgreatthings.ai/otlp, OTLP protocol to http/protobuf, and OTLP headers to Authorization=Bearer <your-ingest-token> (the token from step 3). Save.

That starts the usage and cost stream. Prompt text needs a per-device opt-in. Since an October 2026 Claude Desktop update, Cowork sends prompt events without their text (the literal value <REDACTED>) unless each computer sets the otlpContentCapture device policy. There is no org-wide switch for it in the admin console. Usage and cost work without this; the task map needs it.

On each Mac, either install a configuration profile (MDM or double-click) whose com.anthropic.claudefordesktop payload sets otlpContentCapture to ["userPrompts"], or run:

CONSOLE_USER=$(stat -f%Su /dev/console)
sudo defaults write "/Library/Managed Preferences/$CONSOLE_USER/com.anthropic.claudefordesktop" otlpContentCapture -array userPrompts
sudo defaults write "/Library/Managed Preferences/com.anthropic.claudefordesktop" otlpContentCapture -array userPrompts
sudo killall cfprefsd

On each Windows machine, add a REG_SZ value (no admin rights needed unless machine policy already exists under HKLM\SOFTWARE\Policies\Claude):

reg add "HKCU\SOFTWARE\Policies\Claude" /v otlpContentCapture /t REG_SZ /d "[\"userPrompts\"]" /f

Then fully quit and reopen the Claude app. New Cowork sessions include prompt text.

Note On current Claude Desktop versions, capturing user prompts also exports model response text. Build Great Things drops response events at ingest and never stores them. Because this touches what leaves each person's machine, agree on it with your team before rollout, the same conversation as step 4's note. If prompts ever show as redacted on the Conversations page, a machine is missing this key, usually after a new computer or a reinstall.

6. Connect Gemini CLI (optional)

Gemini CLI also exports OpenTelemetry. Add the telemetry block to ~/.gemini/settings.json and set two environment variables:

# ~/.gemini/settings.json
{"telemetry": {"enabled": true, "target": "local",
  "otlpEndpoint": "https://buildgreatthings.ai/otlp", "otlpProtocol": "http", "logPrompts": true}}

# environment (bearer token; delta temporality so usage metrics ingest)
export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer <your-ingest-token>"
export OTEL_EXPORTER_OTLP_METRICS_TEMPORALITY_PREFERENCE=delta
Note Gemini reports tokens but not spend, so Build Great Things computes cost from current model prices. Sign-in with Google is required for usage to group by person; API-key sessions are still counted, but anonymously per machine.

7. Import your team roster (optional)

A roster lets Usage & Cost break spend down by team. Emails are hashed with your workspace key at upload and never stored.

  1. Prepare a CSV with an email column and a team column. manager_email and display_name columns are accepted for later features.
  2. Open Settings, then under Teams upload the file. Re-uploading replaces the whole roster.
email,team
jane@yourcompany.com,Platform
sam@yourcompany.com,Sales
Note Matching happens as people use their tools after the upload, not retroactively, because no email is ever kept to match against. Expect the match rate in Settings to fill in over the first day or two.

8. Drop in a firm survey (optional)

If your firm already surveys itself about work, importance, effort, or time allocation, import it: the task map gains a survey signal (which patterns serve important work) and an adoption-gaps view (important, effortful work with no AI usage yet).

  1. Open Onboarding from Settings (Context, Import or replace) and drop your survey in as CSV or XLSX. Any column layout works: we map it to workstream, importance, and effort, and show you the parsed result before saving.
  2. Confirm the preview. Re-importing replaces the prior survey.
Note No survey yet? Download the CSV template from the onboarding page, circulate it internally, and import the responses when they come back.

9. Drop your drive in (optional)

Your OneDrive or Google Drive folder is already synced to a laptop. Drop that folder on the Connections page and the map learns your firm's real business processes and grounds skills in your actual templates.

  1. Open Connections, find Drop your drive in, and drag your synced drive folder onto the drop zone (or browse for it).
  2. The browser reads the folder locally and uploads only a metadata index: path, name, extension, size, and modified time. For a large team site expect a "reading 91,000 file names" progress state; uncheck any folders you want left out before uploading.
  3. Freshness is manual by design: drop the folder again whenever. The page shows how old the snapshot is, and a stale snapshot never breaks anything.
Content grounding By default we read file names, never your files. An owner or admin can enable content grounding on Connections: after a drop, the top few hundred most relevant documents are uploaded, analyzed for short excerpts, and the raw files are discarded. Disabling the setting purges stored excerpts. See the security page for exactly what is read and kept.

10. Verify your setup

Within a few minutes of someone using a connected tool:

Checkpoint Ask one person to run a couple of prompts in Claude Code, then check Connections. If nothing appears within five minutes, re-check the endpoint URL and that the token was pasted whole.

Next steps