Praison AI

autonomy • AI Agent SDK

Agent Autonomy Module.

autonomy

AI Agent

Agent Autonomy Module.

Provides agent-centric autonomy configuration and helpers. This module integrates escalation, doom loop detection, and observability directly into the Agent class as first-class capabilities.

Unified Architecture: – AutonomyStage is an alias for EscalationStage (single source of truth) – AutonomyTrigger delegates to EscalationTrigger (DRY) – DoomLoopTracker adds recovery actions on top of basic detection

Usage:

python
from praisonaiagents import Agent

# Enable autonomy with defaults
agent = Agent(instructions="...", autonomy=True)

# Enable with custom config
agent = Agent(
    instructions="...",
    autonomy={
        "max_iterations": 30,
        "doom_loop_threshold": 5,
        "auto_escalate": True
    }
)

# Run autonomous task
result = agent.run_autonomous("Refactor the auth module")
print(result.success, result.output)

Import

python
from praisonaiagents.agent import autonomy

Classes

AutonomyConfig

Configuration for Agent autonomy features.

 AutonomySignal

Signals detected from prompts for autonomy decisions.

 AutonomyTrigger

Detects signals from prompts for autonomy decisions.

 AutonomyResult

Result of an autonomous execution.

 DoomLoopTracker

Tracks actions to detect doom loops with graduated recovery.

 AutonomyMixin

Helper-only trait for autonomy utilities.

Constants

NameValue
VALID_AUTONOMY_LEVELS{'suggest', 'auto_edit', 'full_auto'}
_ESCALATION_TO_AUTONOMY_SIGNAL{'simple_question': 'simple_question', 'file_references': 'file_references', 'code_blocks': 'code_blocks', 'edit_intent': 'edit_intent', 'test_intent': 'test_intent', 'refactor_intent': 'refactor_inte...
_AUTONOMY_TO_ESCALATION_SIGNAL{v: k for k, v in _ESCALATION_TO_AUTONOMY_SIGNAL.items()}
VALID_AUTONOMY_MODES{'caller', 'iterative'}
_LEVEL_TO_DEFAULT_MODE{'suggest': 'caller', 'auto_edit': 'caller', 'full_auto': 'iterative'}