Engine

The engine is the runtime that evaluates controls, resolves selectors, and runs evaluators. It is used by the server and SDK to apply control logic consistently.

Responsibilities

Evaluator Discovery: Auto-discover evaluators via Python entry points
Selector Evaluation: Extract data from payloads using selector paths
Evaluator Execution: Run evaluators against selected data
Caching: Cache evaluator instances for performance

Evaluator Discovery

The engine provides the public API for evaluator discovery:

from agent_control_engine import discover_evaluators, list_evaluators

# Discover all evaluators (runs once, safe to call multiple times)
discover_evaluators()

# Get all available evaluators
evaluators = list_evaluators()  # Returns dict[str, EvaluatorClass]

# Access a specific evaluator
regex_evaluator = evaluators.get("regex")

Evaluators are discovered via the agent_control.evaluators entry point group. Discovery:

Scans all installed packages for the entry point
Loads each evaluator class
Checks is_available() to verify dependencies
Registers available evaluators

Key Functions

Function	Description
`discover_evaluators()`	Scan entry points and register evaluators
`list_evaluators()`	Get all registered evaluators (triggers discovery)
`ensure_evaluators_discovered()`	Ensure discovery has run
`get_evaluator_instance(config)`	Get cached evaluator instance
`evaluate_control(control, payload)`	Evaluate a single control

Source Code

View the complete engine source code and implementation: Agent Control Engine

Getting Started

Concepts

How-to Guides

Integrations

Third-Party Agents

Examples

Reference

Responsibilities

Evaluator Discovery

Key Functions

Source Code

​Responsibilities

​Evaluator Discovery

​Key Functions

​Source Code

Responsibilities

Evaluator Discovery

Key Functions

Source Code