AI Runtime

Overview

AI Runtime specializes the generic Runner framework for AI workloads.
It introduces context-aware execution, prompt management, and standardized storage for AI interactions.

Core Features

• Agent Abstraction:
  Wraps AiAgentInterface plugins into executable
  TaskInterface units.
• Context Management:
  AgentRunContext carries prompts, conversation history,
  and model configuration across execution steps.
• Pipeline Execution:
  Safely execute agents in sequence or via advanced orchestration logic.
• Persistence:
  Built-in ExecutionStorage for saving prompts, responses,
  and token usage statistics.
• Streaming Support:
  Ready for Server-Sent Events (SSE) via
  StreamingEventCollector.

Dependencies

• Module: runner (Task Execution Engine)
• Module: ai_agents (Agent Plugin System)
• External: Symfony Messenger

Architecture Components

AgentRunContext

Extends the generic RunContext with AI-specific capabilities,
carrying prompts, conversation history, and model configuration
across execution steps.

AgentTask

Adapts an AI Agent implementation into a Runner Task and is responsible for:

• Instantiating the Agent plugin
• Injecting the execution context
• Capturing the LLM response
• Standardizing errors

ExecutionStorage

A specialized key-value store for execution artifacts,
enabling asynchronous result retrieval.

Usage

AI Runtime supports synchronous, asynchronous, and pipeline-based
execution models through the Runner framework.

Extension Points

• Custom Task Managers:
  Implement advanced orchestration patterns such as Map-Reduce or Voting.
• Observers:
  Attach listeners for logging, metrics, or side-effects.
• Context Extensions:
  Extend AgentRunContext to carry domain-specific data.