Skip to content

layer2_agents.md

Latest commit

 

History

History
30 lines (24 loc) · 1.26 KB

layer2_agents.md

File metadata and controls

30 lines (24 loc) · 1.26 KB
Raw
description Details Layer 2: Agents, Loops, and Logic.
tags
architecture
agents
layer-2
langgraph
layer 2

Layer 2: Agent Layer (monitor-agents)

The "brain" of the system. It handles narrative intelligence, reasoning, and orchestration using stateless agents.

Core Components

  • Agents: Specialized, stateless workers (e.g., ContextAssembly, Narrator, Resolver, CanonKeeper).
  • Loops: LangGraph StateGraph loops that manage control flow.
  • DSPy: Used for creative reasoning chains (e.g., generating prose, extracting knowledge).
  • Instructor: Enforces strict Pydantic output from LLMs for tool calls.
  • LiteLLM: Provider-agnostic abstraction for LLM calls.

Responsibilities

  • Orchestration: Managing complex, multi-turn interactions.
  • Narrative Logic: Applying rules, rolling dice, generating descriptions.
  • Knowledge Synthesis: Assembling context from Layer 1 tools to feed to prompts.

Strict Rules

  • Rule: Imports from Layer 1.
  • Rule: Never imports from Layer 3.
  • Rule: Agents must be stateless. All persistence is handled via LangGraph Checkpointers (to MongoDB) or MCP tool calls.

See Also