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

# Guard Overview

> Real-time cost monitoring and budget enforcement for AI agents

# Nolma Guard

Guard gives you complete visibility into what your AI agents cost — and lets you stop runaway spend before it becomes a bill.

## Key features

<CardGroup cols={2}>
  <Card title="Real-time tracking" icon="chart-line">
    Costs update within 500ms of every LLM call. Not daily. Real-time.
  </Card>

  <Card title="Hard enforcement" icon="shield">
    HTTP 429 fires **before** the LLM call. Zero cost incurred when limit is hit.
  </Card>

  <Card title="Multi-provider" icon="globe">
    OpenAI, Anthropic, Gemini, Groq, and Mistral — all in one dashboard.
  </Card>

  <Card title="Per-agent tracking" icon="robot">
    See cost by agent, model, user, environment, and session.
  </Card>
</CardGroup>

## What gets tracked

Every LLM call through the gateway creates a **span** with:

| Field         | Description                                  |
| ------------- | -------------------------------------------- |
| Agent name    | From `NM-Agent` header                       |
| Model         | The LLM model used                           |
| Provider      | openai / anthropic / gemini / groq / mistral |
| Input tokens  | Prompt token count                           |
| Output tokens | Completion token count                       |
| Cost (USD)    | Calculated from token counts                 |
| Latency       | End-to-end response time                     |
| Session ID    | Groups related calls                         |

## Dashboard pages

**Guard → Overview** — Total cost today, burn rate, projected month end, top agents, spend over time chart.

**Guard → Sessions** — Every session with cost, token count, span count, and status. Click any session to see the full span tree.

**Guard → Budgets** — Create and manage budget rules. See real-time utilization from Redis.

**Guard → Alerts** — Alert history with severity, type, and reviewed status.