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

# AI Waste & Optimization

> Find and eliminate token spend that isn't producing useful output.

## What counts as AI waste

OpsBeacon flags AI spend as waste when it identifies:

* **Failed requests** — API calls that errored out but still incurred token costs
* **Duplicate requests** — identical prompts sent multiple times without caching
* **Oversized prompts** — prompts significantly longer than needed for the task
* **Unused completions** — completions generated but never used by the application
* **Over-provisioned models** — using GPT-4 or Claude Opus for tasks a smaller model handles equally well

## How to surface waste

```
@Beacon where is AI spend being wasted?
@Beacon show me waste analysis for the AI projects
@Beacon what would I save if I switched the classification project to a smaller model?
```

## Optimization recommendations

For each waste pattern identified, Beacon provides a specific recommendation and an estimated monthly savings if the fix is applied.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.