> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ai-stats.phaseo.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Roll out presets and debug routing

> Use presets to standardize prompt and routing defaults, then inspect request details to understand why AI Stats routed a request the way it did.

Use this recipe when you want one reusable preset for multiple services instead of copying the same prompt, model, and routing defaults everywhere.

## Goal

* Standardize prompt and parameter defaults.
* Limit which providers can be used for a rollout or compliance boundary.
* Keep enough request context to explain why a request routed the way it did.

## 1. Create the preset

In **Dashboard -> Settings -> Presets**, create a preset with:

* a clear name and slug
* a system prompt
* approved models
* provider preferences
* default parameter values such as `temperature`

Use the preset slug in your calling code so the same request policy can be reused across environments.

## 2. Keep the request body small

The main advantage of presets is that the caller no longer needs to repeat the same routing and prompt defaults in every request.

<CodeGroup>
  ```bash cURL theme={null}
  curl https://api.phaseo.app/v1/responses \
    -H "Authorization: Bearer YOUR_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "anthropic/claude-sonnet-4-5",
      "input": "Generate a release summary for the last 24 hours.",
      "preset": "release-summary"
    }'
  ```

  ```typescript TypeScript SDK theme={null}
  import AIStats from "@ai-stats/sdk";

  const client = new AIStats({ apiKey: process.env.AI_STATS_API_KEY! });

  const response = await client.generateResponse({
    model: "anthropic/claude-sonnet-4-5",
    input: "Generate a release summary for the last 24 hours.",
    preset: "release-summary",
  });

  console.log(response.output_text);
  ```

  ```python Python SDK theme={null}
  from ai_stats import AIStats

  client = AIStats(api_key="YOUR_API_KEY")

  response = client.generate_response(
      {
          "model": "anthropic/claude-sonnet-4-5",
          "input": "Generate a release summary for the last 24 hours.",
          "preset": "release-summary",
      }
  )

  print(response.get("output_text"))
  ```
</CodeGroup>

## 3. Understand what the preset changes

The preset is applied before provider routing:

* missing request parameters are filled from the preset
* the preset prompt is prepended to the system layer
* provider preferences narrow the list of available providers before ranking
* model restrictions are checked early instead of silently rerouting

## 4. Debug the routing outcome

When a request does not route the way you expected, open the request in **Gateway -> Usage** and inspect:

* providers considered
* ranked candidates
* routing factors attached to each candidate
* workspace-policy or guardrail blocks

This is the fastest way to answer:

* why a provider was skipped
* why a request was blocked
* why a lower-cost or lower-latency target won the ranking

## 5. Operational pattern

Use this rollout pattern when changing routing behavior:

1. Create a new preset slug for the new policy.
2. Move one low-risk caller to the new preset.
3. Inspect request details and activity trends.
4. Expand the rollout only after the routing outcomes match expectations.

## Related guides

* [Presets](../guides/presets.mdx)
* [Routing and Fallbacks](../guides/routing-and-fallbacks.mdx)
* [Error Handling](../developers/error-handling.mdx)