swapnilroy-cell · Nov 6, 2026 · Nov 6, 2026
diff --git a/aigw/integrations/llms/azure-foundry.mdx b/aigw/integrations/llms/azure-foundry.mdx
@@ -604,6 +604,254 @@ The native `/v1/ocr` endpoint requires gateway version **2.18.0** or higher for
 
 ---
 
+## Decisions (Microsoft-Decision-1)
+
+`Microsoft-Decision-1` is a decision model available in Microsoft Foundry. It does not generate text. It reads a piece of text or JSON, answers typed questions about it, and returns probabilities that your application can threshold or rank. Use it for classification, routing, ranking, grading, and yes/no checks.
+
+The AI Gateway routes these calls through `/v1/decisions`, a separate endpoint from chat completions. You still get the gateway's full feature set: authentication, routing, retries, circuit breaking, caching, logging, and cost tracking. See the [Decisions API](/aigw/product/ai-gateway/decisions-api) for the full request and response reference.
+
+<Note>
+  `/v1/decisions` does not support streaming. Only before-request [guardrails](/aigw/product/guardrails/decisions-guardrails) apply, scanning the content to judge only (`state`).
+</Note>
+
+<Info>
+  This section covers how to call Microsoft-Decision-1 through the AI Gateway. For model details, deployment steps in Azure, regional availability, and responsible-use guidance, see Microsoft's official guide: [Deploy and use Microsoft-Decision-1 in Microsoft Foundry](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-microsoft-decision).
+</Info>
+
+### Model at a glance
+
+These values come from [Microsoft's model overview](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-microsoft-decision#microsoft-decision-1-at-a-glance).
+
+| Model name | Model version | Deployment type | API type |
+|:---|:---|:---|:---|
+| `Microsoft-Decision-1` | 1 | `DataZoneStandard` in selected regions, `GlobalStandard` | Decision |
+
+### Request model vs response model
+
+This is the most common source of errors with this model, so read it before you send a request.
+
+| Field | Value | Example |
+|:---|:---|:---|
+| `model` in your **request** | The **deployment name** you chose in Azure AI Foundry | `@azure-foundry/decision-1` |
+| `model` in the **response** | The underlying model | `microsoft-decision-1` |
+
+<Warning>
+  Do not hardcode `microsoft-decision-1` as the request `model`. Azure looks up a **deployment** with that exact name. If your deployment has a different name, the request fails with `The API deployment microsoft-decision-1 does not exist`. Use the deployment name you see in Foundry. Microsoft describes this in the [Call the decision API](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-microsoft-decision) section of its guide.
+</Warning>
+
+For example, if you deploy the model with the deployment name `decision-1`, send `"model": "@azure-foundry/decision-1"`. If you name the deployment `pi-decision-1`, send `"model": "@azure-foundry/pi-decision-1"`. In both cases the response reports `"model": "microsoft-decision-1"`.
+
+The AI Gateway logs and prices the call using the response `model`, so cost tracking works the same no matter what you name your deployment.
+
+### Setup
+
+1. Deploy `Microsoft-Decision-1` in your Foundry project by following [Microsoft's deployment steps](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-microsoft-decision). From the deployment details page, copy the **resource endpoint**, the **deployment name**, and the **API key**.
+2. Create your Azure AI Foundry provider by following [Creating Your Azure AI Foundry Provider](#creating-your-azure-ai-foundry-provider) above. Use the URL as shown in Foundry for your resource, for example `https://your-resource-name.services.ai.azure.com/models`.
+3. If you manage models in **Model Provisioning**, add your deployment name as the model slug (for example `decision-1`), exactly as it appears in Foundry. See [Adding Multiple Models](#adding-multiple-models-to-your-azure-ai-foundry-provider).
+
+<Note>
+  For decisions requests, The AI Gateway sends the call to the resource root at `/providers/microsoft/v1/systemone`. Any path in the saved Foundry URL, including a `/deployments/<name>` suffix, is dropped. No `api-version` is sent. The deployment is selected by the `model` field in the request body.
+</Note>
+
+Decisions requests work with the same authentication methods as the rest of this integration: API key, Azure Entra ID, and Azure Managed Identity. With an API key, The AI Gateway sends it in the `api-key` header, which is what the Microsoft decision endpoint expects.
+
+### Send a decisions request
+
+Microsoft-Decision-1 uses the Jev request shape: `state` for the content to judge, and a `questions` map. Set `model` to `@provider-slug/deployment-name`. The provider holds the Azure credentials, so no provider headers are needed.
+
+```bash cURL
+curl --request POST \
+  --url https://aigw.portkey.ai/v1/decisions \
+  --header 'Content-Type: application/json' \
+  --header "Authorization: Bearer $PORTKEY_API_KEY" \
+  --data '{
+  "model": "@azure-foundry/decision-1",
+  "state": {
+    "current_ask": "hello there, how is your day going?"
+  },
+  "questions": {
+    "greeting": {
+      "type": "choice",
+      "instructions": "Is this a greeting?",
+      "criteria": {
+        "yes": "a greeting",
+        "no": "anything else"
+      }
+    }
+  }
+}'
+```
+
+Response (values are illustrative):
+
+```json
+{
+  "model": "microsoft-decision-1",
+  "answers": {
+    "greeting": {
+      "type": "choice",
+      "choice": "yes",
+      "probabilities": { "no": 0.0067, "yes": 0.9933 },
+      "confidence": 0.9866
+    }
+  },
+  "usage": { "input_tokens": 46, "output_tokens": 1 },
+  "provider": "azure-ai"
+}
+```
+
+The OpenAI request shape (`input` and a `questions` array) also works. See [Request Shapes](/aigw/product/ai-gateway/decisions-api#request-shapes).
+
+### Question types
+
+Each entry in `questions` sets a `type`, which determines the shape of its answer. The `state` can be text or JSON.
+
+| Type | Question it answers | Answer shape |
+|:---|:---|:---|
+| `noul` | Is this true? | `{ noul: <0 to 1 probability> }` |
+| `choice` | Which option is it? | `{ choice, probabilities, confidence }` |
+| `score` | How much? | `{ score, legend, probabilities, confidence }` |
+
+The response `answers` object is keyed by the question names in your request.
+
+#### `noul`: gates and filters
+
+Use for yes/no conditions, such as whether a reply promises a refund or a document contains a prompt injection. `criteria` is optional. Define the true and false conditions to make borderline questions more precise.
+
+```json
+{
+  "model": "@azure-foundry/decision-1",
+  "state": "I refunded your $200. You don't need to contact billing.",
+  "questions": {
+    "promises_refund": {
+      "type": "noul",
+      "instructions": "Does this reply promise the customer a refund?",
+      "criteria": {
+        "true": "Commits to returning money",
+        "false": "Makes no commitment about money"
+      }
+    }
+  }
+}
+```
+
+#### `choice`: classification and routing
+
+Use to pick exactly one label from a fixed set, such as routing tickets or classifying intent. Describe each option clearly. The response includes a probability for every option, so a low top probability is a good signal to escalate to a person or another model.
+
+```json
+{
+  "model": "@azure-foundry/decision-1",
+  "state": "The API returns 500 on every call.",
+  "questions": {
+    "team": {
+      "type": "choice",
+      "instructions": "Which team should handle this ticket?",
+      "criteria": {
+        "billing": "Charges, invoices, and refunds",
+        "engineering": "Bugs, errors, and outages",
+        "support": "How-to and account questions"
+      }
+    }
+  }
+}
+```
+
+#### `score`: ranking and grading
+
+Use to place an item on an ordered scale, such as incident severity. List the levels from lowest to highest. The returned score is the probability-weighted average of the level indexes, so a four-level scale runs from `0` through `3` and the result can fall between levels. Prefer scores for relative ordering and thresholds, not as absolute ratings.
+
+```json
+{
+  "model": "@azure-foundry/decision-1",
+  "state": "Checkout is down for all EU customers since 09:00.",
+  "questions": {
+    "severity": {
+      "type": "score",
+      "instructions": "How severe is this incident?",
+      "criteria": ["Cosmetic", "Minor", "Major", "Critical"]
+    }
+  }
+}
+```
+
+#### Combine questions in one request
+
+Questions that share the same `state` can go in one request. This reduces request count and keeps the context consistent.
+
+```json
+{
+  "model": "@azure-foundry/decision-1",
+  "state": {
+    "message": "Checkout is down again. This is my third request.",
+    "customer_contact_count": 3
+  },
+  "questions": {
+    "team": {
+      "type": "choice",
+      "instructions": "Which team should handle this request?",
+      "criteria": {
+        "billing": "Charges, invoices, and refunds",
+        "engineering": "Bugs, errors, and outages",
+        "support": "How-to and account questions"
+      }
+    },
+    "severity": {
+      "type": "score",
+      "instructions": "How severe is this issue?",
+      "criteria": ["Cosmetic", "Minor", "Major", "Critical"]
+    },
+    "repeat_contact": {
+      "type": "noul",
+      "instructions": "Has the customer contacted support before?"
+    }
+  }
+}
+```
+
+### Pricing and usage
+
+Usage is reported in the response as `usage.input_tokens` and `usage.output_tokens`. The pricing entry for `microsoft-decision-1` charges for input tokens only. Output tokens, which hold the typed judgment itself, are priced at zero.
+
+### Decisions troubleshooting
+
+| Status | Likely cause | What to do |
+|:---|:---|:---|
+| `400` | Invalid question type or shape | Check `state`, `questions`, `type`, and `criteria` |
+| `401` | Missing, invalid, or expired credential | Check the API key or Entra credentials on the provider |
+| `403` | The identity has no access to the deployment | Check role assignments and deployment access in Azure |
+| `404` | Wrong endpoint, or the deployment name does not exist | Confirm the `model` value matches your deployment name exactly. See [Request model vs response model](#request-model-vs-response-model) |
+| `422` | Unsupported deployment name or question definition | Check the `model` value and each question definition |
+| `429` | The deployment rate limit was exceeded | Retry with backoff, or request more quota in Azure. You can also add retries or fallbacks in a [gateway config](/aigw/product/ai-gateway/configs) |
+
+Upstream errors use the standard error shape, with the message prefixed by `azure-ai`:
+
+```json
+{
+  "error": {
+    "message": "azure-ai error: The API deployment microsoft-decision-1 does not exist. If you created the deployment within the last 5 minutes, please wait a moment and try again.",
+    "type": null,
+    "param": null,
+    "code": null
+  },
+  "provider": "azure-ai"
+}
+```
+
+### Limitations and responsible use
+
+From Microsoft's guidance for this model:
+
+- **No explanations.** The model returns probabilities, not a written rationale.
+- **Wording sensitivity.** Scores can change with how you phrase or order questions and options.
+- **Bias.** The model might reflect biases from its base model and training data. Do not use its scores as the sole basis for decisions about individuals.
+- **Safety filtering.** If you use it as a safety filter, it can miss subtle harmful content or flag benign content.
+- **Validate on your data.** Test thresholds against labeled examples from your own workload before you automate a gate.
+
+See [Known limitations and best practices](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-microsoft-decision) in Microsoft's guide for the full list.
+
+---
+
 ## Next Steps
 
 <CardGroup cols={2}>

diff --git a/aigw/product/ai-gateway/decisions-api.mdx b/aigw/product/ai-gateway/decisions-api.mdx
@@ -174,6 +174,7 @@ Either shape works with any provider. See [Switching Shapes](#switching-shapes).
 | [OpenRouter](/aigw/integrations/llms/openrouter) | `openrouter` | Jev | Routes to Jev models |
 | [Workers AI](/aigw/integrations/llms/workers-ai) | `workers-ai` | Jev | Cloudflare model ID, for example `@your-cf-integration-slug/model` |
 | [OpenAI](/aigw/integrations/llms/openai) | `openai` | OpenAI | `gpt-6-luna` |
+| [Microsoft Decision 1](/aigw/integrations/llms/azure-foundry#decisions-microsoft-decision-1) | `azure-ai` | Jev | Azure AI Foundry. `model` is your deployment name |
 
 Any provider accepts either request shape. The AI Gateway converts the request to the provider's native shape and converts the answers back. See [Switching Shapes](#switching-shapes).