Module pipeline 

Agent Pipeline - Unified LLM interaction pipeline

This module provides a single entry point for all LLM interactions, regardless of input source (text, voice, delegated tasks). It integrates:

• Context analysis and reduction
• Tool filtering based on request
• Agentic loop for tool execution
• Optional ERL-inspired reflection, retry, and memory consolidation
• Streaming and non-streaming responses
• Token estimation and truncation
• Fallback to secondary providers
• Workspace sandboxing for tool execution

Internal organization notes:

• agent_loop owns the runtime control flow and delegates shared iteration/finalization helpers, tracked-task bookkeeping, narration / status emission, and continuation/closeout logic to sidecar modules.
• tool_dispatch owns tool execution; its test suite lives in a dedicated sidecar module so the runtime file stays focused on behavior.

Re-exports

pub use types::*;

Modules

types

Pipeline types for unified LLM interaction.

Structs

AgentPipeline

The main agent pipeline for processing requests

RoutingResult

Result of a pre-flight tool routing operation.

Traits

ToolRouter

Async trait for tool selection strategies.

Functions

build_tool_router

Build a ToolRouter from a ToolRoutingStrategy and app config.

process_ring_stream

Process a stream of gestures from the ring backend and route each one through the agentic pipeline.