hermes-typesafe-jev

# Hermes + Jev: System One Decision Layer for AI Agents

> Typed, probabilistic decisions at 100ms latency — 40-400x cheaper than frontier LLMs for classification, routing, and scoring.

This is the documentation hub for integrating **[Jev](https://typesafe.ai)** (TypeSafe AI's System One Model) with **[Hermes Agent](https://hermes-agent.nousresearch.com)** (Nous Research's autonomous AI agent framework).

## What is Jev?

Jev is a **System One Model** — a class of AI model built for **fast, structured decisions** instead of text generation. While traditional LLMs generate tokens sequentially (great for chat, terrible for `if` statements), Jev takes application state + typed questions and returns **probabilistic decisions with calibrated confidence** in 70-500ms.

| | Frontier LLM (GPT-5.6, Opus 5) | Jev (System One) |
|---|---|---|
| **Latency** | 3-329 seconds | 70-500 milliseconds |
| **Input cost** | $0.20-$10 / MTok | $0.042 / MTok |
| **Output cost** | ~5x input | **Free** |
| **Confidence** | Overconfident, inconsistent | Calibrated probability |
| **Error modes** | Hallucination, refusal, parse failures | Cannot violate schema |

## How it works with Hermes Agent

Hermes has **native Jev integration** built into its core — no plugin required for basic usage. The `hermes-jev` community plugin extends this with a full context engine, nervous system, and adaptive routing.

### Built-in Jev tools (Hermes core)

Hermes exposes 8 `jev_*` tools directly in every session:

| Tool | Purpose |
|------|---------|
| `jev_assess` | Ask Jev questions about the current agent state (choice/score/noul) |
| `jev_decide` | Make one bounded typed decision with TypeSafe Jev |
| `jev_rank` | Rank a bounded set of candidate labels using Jev probabilities |
| `jev_verify` | Verify an e