MAF V1: Python and .NET

Let a Triage agent route the conversation to Math or History specialists, handle handbacks, and guard against loops in production.

Use a centralized manager to pick speakers each round, compare round-robin and agent-driven strategies, and cap runaway loops.

Use SequentialBuilder or BuildSequential to pass one conversation through Writer, Reviewer, and Finalizer agents without hand-built adapters.

Fan one input out to an expert panel, aggregate the responses, and keep wall-clock time bounded by the slowest branch cleanly.

Wrap a ChatClientAgent as a workflow executor, chain two translators, and hide adapter plumbing behind AgentWorkflowBuilder.

Model deterministic workflows with executors as nodes, edges as routes, and a Pregel-style scheduler in Python and .NET.

Two kinds of workflow events — lifecycle and custom. Subscribe to the stream in Python and .NET, filter by type, and wire a live progress indicator.

Stand up a Python MCP server and consume it from Python and .NET MAF agents, including discovery and AITool wiring from one server.

Wire OpenTelemetry spans for agent runs with GenAI attributes, using console export for dev and OTLP for production tracing.

Use middleware to wrap agent runs, intercept tool calls, and redact PII before the LLM sees it in both Python and .NET safely.

Add per-request context with focused providers, compose them before each LLM call, and plug in TextSearchProvider for simple RAG.

Stream tokens as they arrive and reuse a session across turns so the LLM sees the full conversation. ~60 lines of code per language.