2026-10-31 12:28 UTC
----------------
🛠️ Tool
===================
oh-my-hermes (OMH) is an open-source operating layer for Hermes Agent, the agent framework from NousResearch. It does not patch or replace Hermes. It takes over task routing: picking a skill and a model for each lane, running the work in parallel, and keeping every memory change on an explicit, written record. One install, then a single /dag-ulw command handles goal decomposition and execution.
Model routing by work category
Each lane is tagged with a work category: ultrabrain, deep, visual-engineering, quick, and others. The category determines that lane's model chain and reasoning effort, and every chain is editable. The useful part is failure handling: when a provider rejects a model, the chain moves to the next entry instead of silently downgrading. omh model opens a picker where category, head model, and effort are changed with arrow keys and -/+ adjustments.
/dag-ulw graph runs
A goal becomes a graph of small nodes. Ready nodes run in parallel, and nodes running at the same time never share a write scope. A separate node checks the goal, and failed nodes are repaired before the final report. That is the core scheduling claim of the project: parallelism with write isolation, plus an explicit verification step instead of a self-reported done.
Skills and LSP
The skill router ships more than a hundred omh-* specialists covering frontend, backend, Rust, security review, and verification gates. The user never calls them directly; the router loads the right ones into the run as tool calls. LSP integration is opt-in through the lsp_tools setting: definitions, references, and renames come from the language server rather than text search, so aliases, re-exports, and shadowed names are renamed by meaning. It ships off by default, which reads as a sensible choice until the feature matures.
Memory
Nothing is remembered silently. Each memory candidate is kept, refused, or deferred with a written reason. The next session gets a recall pack ranked for its task and cut to a token budget. OMH never reads or patches Hermes' own memory.
Native TUI
Bare omh opens OMH's own TUI on the Hermes gateway. /theme changes the launch look, and /dag-ulw prints the box graph plus a live lane tree showing each lane's model, effort, and token usage. Agents can be steered or interrupted from the agent list, and ! shell lines are supported.
Setup
curl -fsSL https://raw.githubusercontent.com/rlaope/oh-my-hermes/main/install.sh | sh
omh setup
Windows and other install paths are covered in the project documentation.
Use cases
• Multi-step coding work where subtasks deserve different models and reasoning efforts
• Refactors in codebases where text-search renames break on aliases or shadowed names
• Sessions where silent memory writes are a deal-breaker
Limitations
Everything above comes from the project README and docs. Scheduling behavior, write-scope isolation, and the repair flow are not verified in this writeup; the documentation at omher.app is the better reference. OMH targets Hermes Agent specifically, so it is not general agent middleware.
🔹 tool #AI #AIAgents #automation #opensource
🔗 Source: https://github.com/rlaope/oh-my-hermes
Replies (0)
No replies.