All CLI requests fail with GatewayTimeout

Description

All WorkIQ CLI requests to the M365 Copilot API consistently fail with GatewayTimeout (upstream request timeout) errors. Every request is retried 4 times and all attempts time out.

This seems to be account specific (?); a colleague on the same MSIT tenant can use WorkIQ CLI successfully. The M365 Copilot desktop app also works fine for my account, so the issue appears to be specific to the WorkIQ API path.

Steps to Reproduce

1. workiq logout
2. workiq ask -q "hello" -v
3. Sign in when prompted
4. All requests time out with GatewayTimeout

Error Output

`
request-id: 271b097c-311d-40cd-93ec-042836da704b
x-ms-ags-diagnostic: {"ServerInfo":{"DataCenter":"West US 2","Slice":"E","Ring":"4","ScaleUnit":"001","RoleInstance":"MW2PEPF00013F5C"}}

Error: Unexpected error: Too many retries performed. More than 3 retries encountered while sending the request.
(HTTP request failed with status code: GatewayTimeout.
{"error":{"code":"UnknownError","message":"upstream request timeout",
"innerError":{"date":"2026-05-04T23:08:43","request-id":"271b097c-311d-40cd-93ec-042836da704b"}}})
`

Diagnostic Info

• Conversation IDs: 0d316522-d553-4b65-a5ad-0521e8e44e9c, 51a7f43e-a4de-4daf-91a4-7b9ed1a95543, 3059dcf0-6923-4539-94ca-4a7dae482ca9
• Sample Request IDs: 271b097c-311d-40cd-93ec-042836da704b, 07e794ea-6471-4450-953f-2a2df6e75b37, 6a190c7f-2da7-4ec8-9935-f4479829e13d
• All requests routed to West US 2 (ScaleUnits 001, 002, 003, 004) and occasionally West US ScaleUnit 004

What I've Tried

• Fresh login (workiq logout then re-authenticate): same result
• Verified M365 Copilot works via the desktop app (m365.cloud.microsoft/chat)
• EULA is accepted, config looks correct

Environment

• WorkIQ version: 0.4.1.19742
• OS: Windows NT 10.0.28561.0
• Node.js: v16.14.0 (npm copy was removed; built-in binary at .agency\WorkIQ.Cli.win-x64)
• Tenant: MSIT (isMSITTenant=true)
• Date: 2026-05-04, ~3:00 PM – 4:15 PM PDT

Start by reproducing the failure with workiq logout followed by workiq ask -q "hello" -v, then review the verbose output using the listed request and conversation IDs. Compare the CLI request path with the working desktop-app path and verify behavior across the reported West US 2 scale units; done means the CLI request succeeds or the account-specific cause is clearly documented.