1
Summarization → Claude Haiku
Route
- when
- request.tags contains 'summarization'
- then
- Route to Claude Haiku 4 (Anthropic, us-east)
Routing & Policies
Rules are evaluated top-down; the first match wins. Drag a card or use the arrows to reprioritize.
2
Foundry fallback
Fallback
- when
- upstream 'Microsoft Foundry' unavailable or 5xx
- then
- Fallback to GPT-4o-mini (Azure OpenAI, eastus2)
3
Large context guard
Guardrail
- when
- tokens_in > 8000 and app not in approved_large_context
- then
- Block with 413 + policy notice
4
Batch traffic to Llama
Route
- when
- header 'x-gw-priority' = 'batch'
- then
- Route to Llama 3.1 70B (westeurope)
5
Budget hard stop
Guardrail
disabled
- when
- team.month_spend >= team.budget_cap and cap.mode = 'hard'
- then
- Reject with 429 and notify FinOps channel
6
EU data residency pin
Route
- when
- app.region = 'EU'
- then
- Pin to swedencentral / westeurope deployments only
Model catalog
Upstream deployments the gateway can route to
| Model | Provider | Region | Cost / 1k tokens | Status |
|---|---|---|---|---|
| GPT-4o | Azure OpenAI | eastus2 | $0.01200 | connected |
| GPT-4o-mini | Microsoft Foundry | swedencentral | $0.00160 | connected |
| Claude Sonnet 4 | Anthropic | us-east | $0.00900 | connected |
| Llama 3.1 70B | Microsoft Foundry | westeurope | $0.00090 | degraded |
| text-embedding-3-large | Azure OpenAI | eastus2 | $0.00013 | connected |
| Claude Haiku 4 | Anthropic | us-east | $0.00110 | error |