Run goal-based parallel execution and automatic model assignment for Hermes Agent users
If you want to split complex coding tasks into multiple steps and process them simultaneously but find it burdensome to manually select the right model for each step, this tool coordinates in between. It is an operational layer that sits on top of Hermes Agent, automatically selecting the appropriate model and technology for each task and executing them in parallel while you continue conversing with Hermes. It includes a structure that verifies actual execution evidence to ensure planned tasks are not reported as completed without proof.
Goal-based graph execution and model chains
When a user inputs a goal, it is converted into a graph composed of small nodes, and the prepared nodes are executed in parallel. Each node is assigned a suitable model chain and inference effort based on task categories such as ultrabrain, deep, or quick. If a provider rejects a specific model, it moves to the next model without silently lowering performance. Nodes running simultaneously are restricted to using different file areas to prevent conflicts.
Execution evidence-based status reporting
Before reporting that a task is complete, it verifies whether actual tests, reviews, or CI gates have passed. Status is displayed with both the stage and reliability, such as Plan · not run, Code · reported done, or Test · verified. It does not consider a task verified solely because the executor said it was done, and failed nodes undergo a recovery process before report generation. It does not read or modify the existing memory of Hermes Agent, using only OMH's own memory system.
Installation methods and requirements
On macOS and Linux, you can install by receiving the installation script with the curl command, executing it, and then entering omh setup. Windows uses the irm command in PowerShell 5.1 or higher. It can also be installed via Homebrew, Bun, or npm; when using Bun or npm, Node 18 or higher and Python 3.11 or higher must be in the PATH.
Decision criteria for Hermes Agent users
It is suitable for developers who already use Hermes Agent and need parallel execution and model optimization rather than sequential processing by a single agent. The feature that automatically selects a model suitable for each step is particularly useful for large-scale refactoring, security reviews, or changes spanning multiple files. Enabling LSP (Language Server Protocol) allows you to use definition, reference, and rename features based on code structure rather than text search. It is released under the MIT license and operates independently while maintaining compatibility with Hermes Agent.
