> ## 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.

# Agents Overview

> AI agents are autonomous assistants powered by LLMs. Compare to MCP gateways and understand when to use agents.

Agents are autonomous AI assistants that use language models, tools, memory, and reasoning to accomplish tasks.

## Agents vs Gateways

| Aspect        | Agent                     | Gateway             |
| ------------- | ------------------------- | ------------------- |
| **Purpose**   | Autonomous task execution | Tool exposure       |
| **Reasoning** | ✓ Complex logic, planning | ✗ Pass-through only |
| **Memory**    | ✓ Conversation history    | ✗ Stateless         |
| **Autonomy**  | ✓ Makes decisions         | ✗ User-driven       |
| **Channels**  | ✓ Telegram, Slack, SMS    | ✗ MCP protocol only |
| **Model**     | ✓ Own model selection     | ✓ Configurable      |

**Use Agents when:** You want autonomous, intelligent behavior with memory and decision-making.

**Use Gateways when:** You want to expose tools to MCP clients without autonomous execution.

## Key Features

* **System Prompts** - Define agent personality and behavior
* **Tool Access** - Use capabilities, connectors, plugins
* **Memory** - Remember conversation context
* **Model Routing** - Smart or specific model selection
* **Channels** - Telegram, Slack, SMS integration
* **Automation** - Scheduled and event-triggered runs

## Creating an Agent

1. **Agents** > **Create Agent**
2. Enter name and description
3. Set system prompt (optional starter provided)
4. Choose model routing
5. Attach capabilities
6. Click **Create**

Agent immediately ready to use.

## Using Agents

### Direct Chat

In Console, chat directly with agent. Test and iterate.

### Channels

Deploy to messaging platforms:

* Telegram
* Slack
* SMS

Users interact via their preferred platform.

### API

Programmatic execution:

```bash theme={null}
curl -X POST https://api.noorle.com/v1/agents/{id}/run \
  -H "Authorization: Bearer {token}" \
  -d '{"prompt": "What is 2+2?"}'
```

### Automation

Scheduled or event-triggered execution via Automations.

## Agent Capabilities

<CardGroup cols={2}>
  <Card title="Creating an Agent" href="./creating-an-agent">
    Step-by-step guide to create agents.
  </Card>

  <Card title="System Prompts" href="./system-prompts">
    Write effective prompts to guide agent behavior.
  </Card>

  <Card title="Model Routing" href="./model-routing">
    Choose models (Smart vs Specific).
  </Card>

  <Card title="Attaching Capabilities" href="./attaching-capabilities">
    Add tools to agents.
  </Card>

  <Card title="Sub-Agents" href="./sub-agents">
    Multi-agent delegation and handoffs.
  </Card>

  <Card title="Memory & Context" href="./memory-and-context">
    Conversation memory configuration.
  </Card>

  <Card title="Omni Tool" href="./omni-tool">
    LLM-based tool discovery.
  </Card>
</CardGroup>

## Common Agent Patterns

### Research Agent

* Web Search
* Browser (visual scraping)
* Knowledge Retrieval
* Code Runner (analyze data)

### Customer Support

* Knowledge Retrieval (FAQ, docs)
* HTTP Client (lookup customer data)
* Slack channel
* Warm handoff to human

### Data Analysis

* Files (read/write)
* Code Runner (Python)
* Knowledge Retrieval (context)
* HTTP Client (fetch data)

### Content Creator

* Web Search (research)
* Browser (gather examples)
* Code Runner (process data)
* Slack integration

## Agent State

Agents have lifecycle states:

* **Active** - Enabled and running
* **Inactive** - Disabled, not running
* **Archived** - Removed from view

Toggle **Active** in agent settings to enable/disable.

## Monitoring

Track agent performance:

1. **Agents** > Select agent

2. **Analytics** tab shows:
   * Response count
   * Average latency
   * Token usage
   * Error rate

3. **Activity** tab shows recent interactions

## Best Practices

### System Prompts

* Be specific and clear
* Set expected tone and behavior
* Include guardrails
* Explain constraints

### Tool Selection

* Attach only needed tools
* More tools = higher latency
* Organize by use case
* Document each tool

### Memory

* Configure appropriate working memory
* Set summary thresholds
* Monitor context window
* Archive old conversations

### Channels

* Set appropriate platform
* Configure authorization
* Test before deployment
* Monitor for errors

## Cost

For current pricing details, see [Pricing](https://noorle.com/pricing/).

Monitor in **Account** > **Usage** dashboard.

## Next Steps

* [Create Your First Agent](/docs/run/agents/creating-an-agent)
* [Write System Prompts](/docs/run/agents/system-prompts)
* [Attach Capabilities](/docs/run/agents/attaching-capabilities)
* [Deploy to Slack](/docs/run/channels/slack)