Skip to content
AgentStateLabs

AgentStateRouter

AgentStateRouter (AEOE — Agent Execution Optimization Engine) answers the question: what is the best execution strategy for this task?

What it is (and isn’t)

Section titled “What it is (and isn’t)”

AgentStateRouter is not just an LLM router. LLM routers optimize for tokens. AEOE optimizes for workloads — cost-per-unit-of-work, efficiency score, and agent history per task type. A task that costs 10× more tokens but completes correctly the first time may be cheaper than a task that requires three retries.

It’s not a framework. It’s a suggestion engine that plugs into your existing stack without being in the request path.

How it works

Section titled “How it works”

AEOE is built on AgentStateGraph, so every routing decision is a commit: structured intent, reasoning, confidence, and outcome. Plans, tasks, policies, and scoring history live in one graph — queryable, blameable, and auditable.

Three deployment shapes over the same engine:

  1. Suggester (HTTP/JSON-RPC) — pluggable into any product, not in the request path. Ships first.
  2. Proxy facade — OpenAI- and Anthropic-compatible HTTP. Drop-in target for LiteLLM users.
  3. Embedded crate — direct linkage for Rust applications.

What it optimizes

Section titled “What it optimizes”
  • Cost-per-unit-of-work — not just tokens, but successful task completions
  • Efficiency score — tracks agent history per task type and adjusts routing
  • Budget enforcement — circuit breakers, retry limits, and fallback chains
  • Policy gate — per-task authorization before dispatch

Status

Section titled “Status”

Pre-release. Foundation in active development. Built on the same AgentStateGraph substrate as the rest of the portfolio — plans, policies, and task history are already in place.

Source

Section titled “Source”