Skip to main content
Version: 2.0

Micro-Agents

Micro-agents are specialized, lightweight agents that provide platform-level capabilities to enhance primary agents. By separating platform concerns (security, governance, coordination) from business logic, they allow primary agents to focus on domain-specific tasks while handling cross-cutting capabilities transparently and consistently.

With micro-agents, your agents:​

  • Get consistent governance, planning, and coordination without custom implementation
  • Keep business logic focused while platform capabilities remain reusable
  • Benefit from transparent operation that works behind the scenes
  • Are production-ready by default with built-in security and compliance
  • Maintain optimized performance with minimal overhead

Micro-Agents in aiXplain​

Response Generator

Available in All Agents

Formats final outputs (text, markdown, JSON)

Planner

Additional for Team Agents

Breaks down complex tasks into structured plans

Orchestrator

Additional for Team Agents

Coordinates execution across multiple agents

Inspector

Additional for Team Agents

Validates inputs and outputs for safety and compliance

Bodyguard

Additional for Team Agents

Protects access to data sources and tools based on governance rules and RBAC

Debugger

Available in All Agents

Analyzes agent responses for issues and insights

Planner​

Handles advanced reasoning and task decomposition for team agents:

  • Breaks down complex queries into actionable steps
  • Determines optimal agent selection and sequencing
  • Generates structured execution plans before agents act

Value: Improves reasoning quality, reduces inefficient tool calling, and enables sophisticated problem-solving.

Orchestrator​

Coordinates execution across multiple agents and tools:

  • Routes tasks to appropriate agents based on capabilities
  • Manages dependencies between sequential operations
  • Handles parallel execution when possible

Value: Ensures proper sequencing, improves execution efficiency, and maintains consistency across multi-step processes.

Inspector​

Provides runtime guardrails and governance:

  • Validates inputs and outputs against defined policies
  • Checks for policy violations, inappropriate content, and data leakage
  • Detects potential prompt injection attempts
  • Triggers retries or edits when validation fails

Enforcement Actions: Block, warn, rerun, or automatically edit violating content.

Value: Ensures compliance, prevents harmful outputs, protects sensitive data, and provides audit trails—critical for regulated industries.

Response Generator​

Formats the final output based on the specified format (TEXT, MARKDOWN, or JSON) and ensures responses match the expected structure.

Debugger​

Analyzes agent responses to surface issues, explain reasoning failures, and provide actionable insights. Unlike other micro-agents, the Debugger is invoked explicitly rather than operating transparently in the background—either as a standalone call or chained directly from a run response.

Usage:​

  • Standalone: instantiate and run the Debugger independently, passing any agent response you want to analyze
  • Chained: call .debug() directly on any agent response for a more concise workflow

Value: Speeds up development and triage by providing structured feedback on agent reasoning, tool usage, and failure modes—useful when an agent produces unexpected output, skips tools, or hits its iteration limit

Best Practices​

For Production:

  • Always enable Inspector for customer-facing agents
  • Configure Inspector policies specific to your compliance requirements
  • Monitor enforcement actions in execution traces
  • Use the Debugger to triage unexpected agent behavior before escalating issues

For Development:

  • Keep Planner and Inspector enabled to understand reasoning patterns
  • Review execution traces to see how micro-agents coordinate
  • Adjust Inspector thresholds based on validation needs
  • Chain the Debugger during development to get immediate feedback on agent reasoning