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

# Agents

> Agents are the autonomous workers of the MINDBLIX platform. Learn how they think, reason, and execute tasks.

## What is an Agent?

An **Agent** in MINDBLIX is an autonomous AI entity specialized for a specific domain or task type. Each agent combines:

* A **foundation model** (e.g., Gemini 1.5 Flash) for reasoning
* **Domain-specific prompts** that define its behavior and expertise
* **Tool access** for interacting with external systems
* **Knowledge graph context** for grounded, factual outputs

```mermaid theme={null}
graph TD
    A[User Input] --> B[Agent]
    B --> C[Foundation Model]
    B --> D[Domain Prompts]
    B --> E[Tools]
    B --> F[Knowledge Graph]
    C --> G[Reasoning]
    D --> G
    E --> G
    F --> G
    G --> H[Output]

    style B fill:#9300FF,color:#fff
    style G fill:#6A00B8,color:#fff
```

***

## Agent Lifecycle

**1. Configuration**

Define the agent's name, template, model, tools, and behavior parameters.

**2. Deployment**

The agent is deployed to the MINDBLIX runtime and becomes available for task execution.

**3. Task Reception**

The agent receives a task from the orchestrator or a direct API call.

**4. Reasoning Loop**

The agent enters a recursive reasoning loop:

1. Analyze the task
2. Plan an approach
3. Execute steps
4. Validate output
5. Self-correct if needed

**5. Output Delivery**

Validated output is returned to the caller or passed to the next agent in the workflow.

***

## Agent Types

| Type | Specialization | Example Use Cases |
| - | - | - |
| **Research** | Information gathering, fact-checking | Market research, competitor analysis |
| **Code** | Code generation, review, debugging | Feature development, code review |
| **Data** | Analysis, visualization, modeling | Business intelligence, reporting |
| **Security** | Vulnerability scanning, compliance | Pen testing, audit reports |
| **Finance** | Financial analysis, forecasting | Portfolio analysis, risk assessment |
| **Health** | Medical data analysis, diagnosis support | Clinical trial analysis, symptom checking |
| **Education** | Content creation, tutoring | Course generation, student assessment |
| **Marketing** | Campaign optimization, content | SEO optimization, A/B testing |

***

## Agent Configuration

Agents are configured via JSON:

```json theme={null}
{
  "name": "Research Agent",
  "template": "research",
  "model": "gemini-1.5-flash",
  "config": {
    "max_iterations": 5,
    "temperature": 0.7,
    "tools": ["web_search", "document_reader", "summarizer"],
    "output_format": "json",
    "validation": {
      "check_knowledge_graph": true,
      "min_confidence": 0.85
    }
  }
}
```

> [!WARNING]
>
> Setting `min_confidence` too low may result in outputs that haven't been sufficiently validated. We recommend keeping it above `0.8` for production use cases.

***

## Communication Protocol

Agents communicate using the **A2A (Agent-to-Agent)** protocol:

```json theme={null}
{
  "type": "a2a_message",
  "from": "research-agent-001",
  "to": "data-agent-003",
  "payload": {
    "task": "analyze",
    "data": { ... },
    "context": { ... }
  },
  "metadata": {
    "correlation_id": "wf-12345",
    "timestamp": "2026-03-04T10:00:00Z"
  }
}
```


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