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

# Usage and Limits

> Monitor token usage, track costs, understand rate limits and billing.

Track your platform usage in real-time and understand costs associated with your agents, gateways, and capabilities.

## Usage Dashboard

### Accessing Usage

1. Go to **Account** > **Usage**
2. View metrics for:
   * Tokens consumed (input + output)
   * Cost per model
   * Request counts
   * Error rates
   * Time period selection (Today, Week, Month, Custom)

### Key Metrics

| Metric       | Description                                   |
| ------------ | --------------------------------------------- |
| **Tokens**   | Total input + output tokens across all models |
| **Cost**     | Calculated based on model pricing and region  |
| **Requests** | Number of API calls and tool executions       |
| **Errors**   | Failed requests and their error types         |
| **Latency**  | Average response time (p50, p95, p99)         |

## Token Accounting

Tokens are the primary unit of usage and cost calculation.

### Token Types

**Input Tokens:**

* Prompt text you send to the model
* System prompt
* Context windows and memory
* Tool descriptions
* Prior conversation history

**Output Tokens:**

* Model response text
* Tool execution results
* Streaming overhead

### Token Pricing

Pricing varies by model. For current rates, see [Pricing](https://noorle.com/pricing/).

## Cost Optimization

### Use Smart Routing

Let Noorle optimize model selection automatically. Smart Routing analyzes complexity and picks the right model:

```json theme={null}
{
  "routing_strategy": {
    "strategy": "smart"
  }
}
```

### Pair Primary + Budget Model

Cheap model for simple tasks, expensive model for complex ones:

```json theme={null}
{
  "model_name": "gpt-4o",
  "budget_model": "gpt-4o-mini"
}
```

Budget model typically saves 50-80% on simple requests.

### Optimize System Prompts

Shorter system prompts = fewer input tokens:

**Verbose (high cost):**

```
You are an AI assistant. You should be helpful, harmless, and honest.
You should always provide detailed explanations...
[continues for 2KB]
```

**Concise (low cost):**

```
Helpful AI assistant. Provide clear, brief responses.
```

### Configure Memory Settings

Reduce context window size for cheaper operations:

1. Agent Settings > Memory Configuration
2. Set `working_memory_size` to minimum needed
3. Reduce `summary_token_threshold`
4. Set `summary_message_threshold` to trigger summaries earlier

### Batch Requests

Process multiple items in one request instead of individual calls:

**Expensive (separate calls):**

```
3 API calls × 1000 tokens each = 3000 tokens
```

**Efficient (batch):**

```
1 API call × 2500 tokens = 2500 tokens saved
```

### Use Caching and Memory

Leverage conversation memory to avoid re-processing:

* Enable agent memory (3-tier system)
* Reuse context across requests
* Let platform manage summarization

## Platform Limits

| Limit               | Value       | Notes                   |
| ------------------- | ----------- | ----------------------- |
| Upload Size         | 20MB        | Maximum request payload |
| MCP Request Timeout | 240 seconds | Per gateway request     |

For plan-specific limits and quotas, see [Pricing](https://noorle.com/pricing/).

<Tip>
  Contact sales for custom enterprise limits.
</Tip>

### Hitting Limits

If you hit a rate limit:

```
HTTP 429 Too Many Requests
```

Response includes:

```json theme={null}
{
  "error": "rate_limit_exceeded",
  "retry_after_seconds": 60
}
```

**Handling:**

* Implement exponential backoff
* Batch requests
* Request custom limits

## Usage Breakdown by Resource

### By Agent

1. Go to **Agents** > Select Agent
2. Click **Usage** tab
3. See tokens and cost for this agent only

### By Gateway

1. Go to **Gateways** > Select Gateway
2. Click **Usage** tab
3. See tokens and cost for this gateway only

### By Model

1. Go to **Account** > **Usage**
2. View "By Model" table
3. See which models are most expensive

### By Capability

1. Go to **Account** > **Usage**
2. View "By Capability" table
3. See cost of each capability (Web Search, Code Runner, etc.)

## Capability Limits Reference

### Web Search

| Limit       | Value      | Notes       |
| ----------- | ---------- | ----------- |
| Timeout     | 30 seconds | Per request |
| Max Results | 10         | Per query   |
| Max Retries | 3          | Automatic   |

**Optimization Tips:** Use specific queries to reduce result processing. Cache recurring searches in Files capability.

### Browser

| Limit            | Value      | Notes                         |
| ---------------- | ---------- | ----------------------------- |
| Timeout          | 30 seconds | Configurable                  |
| Default Viewport | 1280 × 720 | Configurable                  |
| JavaScript       | Enabled    | Fully renders dynamic content |

**Optimization Tips:** Use text extraction instead of screenshots when you only need content. Set appropriate timeouts for slow pages.

### Code Runner

| Limit       | Value      | Notes                                    |
| ----------- | ---------- | ---------------------------------------- |
| Timeout     | 30 seconds | Maximum execution time                   |
| Memory      | 128MB      | Per execution (configurable up to 512MB) |
| CPU Time    | \~100ms    | Maximum CPU time                         |
| Code Size   | 10KB       | Maximum code length                      |
| Output Size | 1MB        | Console output                           |
| Network     | Disabled   | No HTTP requests                         |
| File I/O    | Disabled   | No file read/write                       |

**Optimization Tips:** Keep executions under 100ms CPU time. Process large datasets in chunks. Use Sandbox for anything needing external packages or network access.

### Sandbox

| Limit                  | Value                     | Notes                  |
| ---------------------- | ------------------------- | ---------------------- |
| Sizes                  | XS to XL                  | 5 predefined tiers     |
| vCPU Range             | 0.25 – 4.0                | Based on size          |
| Memory Range           | 1 GB – 12 GB              | Based on size          |
| Disk Range             | 4 GB – 20 GB              | Based on size          |
| Command Timeout        | 5 min default, 30 min max | Configurable           |
| Max Lifetime           | 60 minutes                | Auto-terminates        |
| Idle Timeout           | 20 min                    | Auto-terminates        |
| Containers per Session | 1                         | One active at a time   |
| Network                | Outbound only             | No inbound connections |

**Optimization Tips:** Start with the smallest size (XS) and scale up as needed. Use Code Runner for simple operations that don't need packages. Sandboxes auto-terminate after 20 min idle to save costs.

### HTTP Client

| Limit         | Value       | Notes             |
| ------------- | ----------- | ----------------- |
| Request Body  | 2MB         | Maximum size      |
| Response Body | 2MB         | Maximum size      |
| Timeout       | 30s default | Min 1s, max 300s  |
| Rate Limit    | 1,000/min   | Per user          |
| Max Redirects | 10          | Per request chain |

**Optimization Tips:** Implement retry logic with exponential backoff. Compress large payloads. Use connection pooling for repeated calls to the same host.

### Files

| Limit              | Value | Notes                    |
| ------------------ | ----- | ------------------------ |
| Per-File Size      | 10MB  | All scopes               |
| Files per Scope    | 100   | Per scope                |
| Agent Home Storage | 50MB  | Persistent agent storage |

**Optimization Tips:** Use session-scoped files for temporary processing. Only persist results to agent home when needed. Clean up unneeded files to stay within limits.

### Knowledge Retrieval

| Limit               | Value     | Notes              |
| ------------------- | --------- | ------------------ |
| Query Length        | 500 chars | Search query max   |
| Results per Search  | 100       | Maximum            |
| Document Size       | 10MB      | Per document       |
| Index Size          | 10GB      | Per knowledge base |
| Concurrent Searches | 100/sec   | Per knowledge base |

**Optimization Tips:** Pre-filter with metadata before searching. Use appropriate chunk sizes for your documents. Cache frequent queries.

### Computer

| Limit                | Value          | Notes                        |
| -------------------- | -------------- | ---------------------------- |
| Sizes                | x2 to x32      | 5 predefined tiers           |
| vCPU Range           | 2 – 16         | Based on size                |
| RAM Range            | 2 GB – 32 GB   | Based on size                |
| Disk Range           | 40 GB – 360 GB | Based on size                |
| Default Size         | x4             | 3 vCPU, 4 GB RAM, 80 GB disk |
| SSH Timeout          | 30 seconds     | Per command                  |
| Browser Max Tabs     | 5              | Per session                  |
| Browser Max Download | 50 MB          | Per file                     |
| Snapshot Elements    | 500            | Truncated if exceeded        |

**Optimization Tips:** Start with x4 (default) and scale up for memory-intensive or multi-process workloads. Enable browser subsystem only if needed. Agent-only — not available on MCP gateways.

### Plugins (WebAssembly)

| Limit       | Value                    | Notes                 |
| ----------- | ------------------------ | --------------------- |
| Memory      | 128MB default, 512MB max | Per invocation        |
| CPU Time    | 100ms max                | Per request           |
| Timeout     | 30s default, 120s max    | Configurable          |
| Module Size | 10MB                     | Compressed npack file |

**Optimization Tips:** Configure memory limits based on your workload (up to 512MB max). Set appropriate timeouts (up to 120s max). Keep module size small for faster deployment.

## Pricing and Billing

For current pricing details, plan comparisons, and billing information, see [Pricing](https://noorle.com/pricing/).

## Alerts and Budgets

### Budget Alerts

Set spending threshold to receive notifications:

1. Go to **Account** > **Billing** > **Budget Alerts**
2. Click **Create Alert**
3. Set threshold amount
4. Choose notification method (email, webhook)
5. Click **Save**

You'll be notified when spending reaches the threshold.

### Usage Alerts

Get notified when you approach limits:

1. Go to **Account** > **Usage** > **Alerts**
2. Toggle the limits you want to monitor:
   * Daily API calls
   * Storage usage
   * Token consumption
3. Set notification threshold
4. Click **Save**

## Enterprise Limits

For enterprise plans with custom limits, contact [sales@noorle.com](mailto:sales@noorle.com) or see [Pricing](https://noorle.com/pricing/).

## Troubleshooting

### High Unexpected Cost

**Check:**

1. Go to **Usage** dashboard
2. Filter by time period (daily, hourly)
3. Identify which resource is expensive
4. Check token breakdown by model

**Common causes:**

* Large system prompt
* Expensive model selected
* Budget model not triggering
* Streaming overhead
* Knowledge base queries with large documents

### Hitting Rate Limit

**Solutions:**

1. Implement exponential backoff
2. Batch requests
3. Reduce concurrent requests
4. Upgrade plan
5. Contact support for custom limits

### Usage Not Updating

**Give it time:** Usage updates every 5-15 minutes. Check again in a few minutes.

**Force refresh:** Go to different tab and return.

## API Access

Query usage programmatically:

```bash theme={null}
# Get usage statistics
curl https://api.noorle.com/v1/usage/summary \
  -H "X-API-Key: ak-{your_key}"

# Get detailed usage
curl https://api.noorle.com/v1/usage/detailed?period=month \
  -H "X-API-Key: ak-{your_key}"

# Get resource-specific usage
curl https://api.noorle.com/v1/agents/{agent_id}/usage \
  -H "X-API-Key: ak-{your_key}"
```

## Next Steps

* [Configure Model Routing](/docs/run/platform/model-routing)
* [Optimize Agent Memory](/docs/run/agents/memory-and-context)