Agents

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

Agent Lifecycle

Configuration

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

Deployment

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

Task Reception

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

Reasoning Loop

The agent enters a recursive reasoning loop:

Analyze the task
Plan an approach
Execute steps
Validate output
Self-correct if needed

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:

{
  "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
    }
  }
}

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:

{
  "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"
  }
}

Getting Started

Core Concepts

Guides

Agent Catalog

What is an Agent?

Agent Lifecycle

Agent Types

Agent Configuration

Communication Protocol

​What is an Agent?

​Agent Lifecycle

​Agent Types

​Agent Configuration

​Communication Protocol

What is an Agent?

Agent Lifecycle

Agent Types

Agent Configuration

Communication Protocol