oh-my-hermes: Enhance Hermes Agent with Advanced AI Workflow and Memory

Introduction

oh-my-hermes (OMH) is a powerful, all-in-one plugin built to augment the capabilities of the Hermes Agent. It introduces a sophisticated operational layer, enhancing Hermes with advanced coding intelligence, a long-term memory system, and optimized workflow packages. OMH does not replace Hermes, but rather strengthens existing Hermes workflows by converting typical requests into actionable functions, useful next steps, and an honest state of what has occurred and what has not.

Why Use and Key Features

oh-my-hermes significantly enhances Hermes Agent by introducing a suite of powerful features designed for professional development workflows:

Intelligent Model Routing: OMH implements a mixture-of-models routing system, dynamically assigning tasks to the most suitable AI model and inference strength based on complexity. This optimization leads to substantial savings in cost and time, as demonstrated by its ability to achieve similar results with significantly fewer tokens and shorter durations.
Customizable Category Chains: Users can define and own model+effort chains for various coding executors, such as ultrabrain, deep, and quick. This allows for tailored model selection, ensuring that if a provider rejects a model, the system intelligently falls back to the next in the chain, preventing silent downgrades.
Calibrated Prompting: OMH fine-tunes prompting for 13 different model families, optimizing instructions based on each family's documented characteristics. This ensures precise communication with models like Claude, Gemini, Qwen3-Coder, and DeepSeek, leading to more accurate and efficient outputs.
Parallel and Verified Workflows: The ulw-work workflow enables the division of approved plans into non-overlapping units, executing them in parallel. Each unit returns typed results, and a verification gate ensures that only genuinely confirmed outcomes are reported as complete, distinguishing between "reported done" and "verified."
Comprehensive Workflow Interface (HUD): A native TUI surface provides a live, detailed Heads-Up Display (HUD) for OMH workflows. It shows per-lane metrics including model, effort, turns, tokens, cost, and evidence status, along with phase-structured TODOs, offering full transparency into the agent's progress.
Integrated Expert Skills: OMH includes over 100 omh-* specialized skills covering areas like frontend, backend, Rust, native debugging, security review, and performance budgeting. These skills seamlessly integrate into execution as tool calls, raising the bar for what the agent considers "complete."
Reviewer-Approved Long-Term Memory: Unlike typical memory systems, OMH's file-based memory requires reviewer approval for new entries. Memories are captured from sessions, reviewed, and then stored with source and review deadlines. This ensures that only relevant and verified lessons are recalled for future tasks, without interfering with Hermes's own memory.
Evidence-Based Reporting: OMH adheres to a strict "Claim vs. Evidence" principle, reporting only what it has observed. It clearly differentiates between a task being reported done by an executor and being Test · verified through actual checks, reviews, or CI gates.

Installation

To get started with oh-my-hermes, follow these quick installation steps:

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/rlaope/oh-my-hermes/main/install.sh | sh

Windows (PowerShell 5.1+):

irm https://raw.githubusercontent.com/rlaope/oh-my-hermes/main/install.ps1 | iex

After installation, run the setup command:

omh setup

To update oh-my-hermes:

omh update

Alternatively, you can install via package managers like Homebrew, Bun, or npm:

Homebrew: brew install rlaope/tap/omh
Bun: bun install -g oh-my-hermes
npm: npm install -g oh-my-hermes

Remember to run omh setup after any package manager installation.

Examples

Simply typing omh will launch Hermes with the OMH identity, providing a rich interface for your AI agent workflows:

omh

OMH offers nine ulw- workflows that can be triggered directly in chat. Some examples include:

ulw-context: Aligns reviewed project terminology and incorporates confirmed candidates.
ulw-research: Investigates actual code and the web, providing sources and cross-referencing suspicious claims.
ulw-plan: Creates a reviewed plan with agreed-upon scope, code change ownership, tests, and completion criteria.
ulw-work: Executes an approved plan across parallel lanes that do not touch the same files.
ulw-qa: Attacks with deliberately harsh scenarios and fixes what breaks.