infra: bump default modelCapacity 50 → 150 (avoid day-one 429s)

[paulyuk]

Problem

Default modelCapacity = 50 (= 50K TPM) is too low for a multi-step agent template. A single daily-briefing run pulls a chunk of inbox, reasons, and renders — easily 10–20K tokens. Two runs within a minute trip the per-deployment rate limit and surface as:

Error calling daily_briefing /chat: HTTP 500
{"error": "...Error code: 429 - {'error': {'code': 'rate_limit_exceeded', 'message': 'Model deployment rate limit exceeded. Too Many Requests...'}}"}

This is a terrible first-deploy UX — operators assume their config, code, or model is broken.

Repro

1. Fresh azd up of this template.
2. uv run python chat.py → option 2 (daily-briefing).
3. Run option 2 again within ~30s. → 429.

Fix

Bump default modelCapacity from 50 → 150 (= 150K TPM).

Why 150 is the right default

• Comfortably handles repeated single-user agent runs — a typical chain is 4–6 LLM calls; 150K TPM gives ~10× headroom per minute.
• Within most subs' regional quota. New subs typically get 1000 units (1M TPM) per model per region for the gpt-5 family on GlobalStandard. 150 is 15% of that → plenty of room.
• Leaves room for multiple template deploys in the same sub before quota-bumping is needed.

Self-check command

az cognitiveservices usage list --location <region> \
  --query "[?contains(name.value, 'gpt-5.4-mini')]" -o table

If limit - currentValue < 150, lower the param at deploy time:

azd env set MODEL_CAPACITY 100   # or whatever fits
azd up

In-place fix for users hit by this on the old default

No need to redeploy from scratch — bump the existing deployment:

az cognitiveservices account deployment create \
  -g <resource-group> -n <ai-services-account> \
  --deployment-name gpt-5.4-mini --model-name gpt-5.4-mini \
  --model-version 2026-03-17 --model-format OpenAI \
  --sku-name GlobalStandard --sku-capacity 150

(deployment create on an existing name is an in-place update — completes in seconds, no app downtime.)

Verified

Tested on a live deploy that was 429-ing — bump took 5s, daily-briefing succeeded on next call.

[manvkaur]

@copilot uv run ruff format --check and create a PR to merge in this branch

[Copilot]

@manvkaur I've opened a new pull request, #19, to work on those changes. Once the pull request is ready, I'll request review from you.

[paulyuk]

Superseded by #19, which included the same modelCapacity 50→150 bump (with the longer description). Closing as redundant after rebase showed an empty diff.