HHermes Explained

Operations

Local Models

Models hosted locally through tools such as LM Studio or Ollama-style servers.

Importance 4/54 backlinks1 source refs

Local Models

What it is

Models hosted locally through tools such as LM Studio or Ollama-style servers.

In plain terms

Running a local model is a bit like brewing your own coffee at home instead of walking to a café for one: it’s already in the kitchen, there’s no per-cup charge and no line to wait in, but the setup on your counter can’t quite match a café’s full range of blends. Local models let Hermes think without any request ever leaving your machine.

Why it matters

Served through something like LM Studio or an Ollama-style server, local models let Hermes run without sending prompts to a hosted provider at all, at the cost of the capability gap between a locally-runnable model and a frontier hosted one. This is the choice both cost-control and privacy push toward when the stakes justify it: no per-call billing, no data leaving the machine. Provider switching is what makes mixing local and hosted models per-task practical, rather than an all-or-nothing decision.

How it works

Hermes treats a locally-running model server the same way it treats any hosted API - through a provider alias. LM Studio has a built-in overlay that defaults to http://127.0.0.1:1234/v1 and reads an LM_BASE_URL override if you’ve moved the port (hermes_cli/providers.py). A bare ollama reference maps internally to the generic custom provider, while vllm and llama.cpp map to a local virtual provider - all three resolve through whatever endpoint you’ve set in your own config rather than a hosted catalog entry (hermes_cli/providers.py). The dedicated plugin behind that generic provider describes itself simply as a ‘Custom / Ollama / local OpenAI-compatible endpoint’ (plugins/model-providers/custom/plugin.yaml), since any server speaking the OpenAI chat-completions shape on localhost qualifies.

A concrete example

You install Ollama on your desktop and pull a small open-weight model overnight.

  1. You set the provider to ollama in your Hermes config the next day.
  2. Hermes resolves that alias to the local custom provider pointing at your machine’s own port.
  3. A batch of routine file-renaming requests gets answered entirely by the model on your desktop, with no request ever reaching an outside company’s server, and no per-call bill arriving afterward.

How it connects

This note belongs to the Operations family. Its closest neighboring concepts are: Model Providers, Provider Switching, Agent Runtime, Personal OS.

Source evidence

  • vendor/hermes-agent/run_agent.py:1892-1902#AIAgent._flush_messages_to_session_db — Verified Source Map evidence for AIAgent. AIAgent uses thin forwarding methods to delegate system-prompt construction to agent.system_prompt.build_system_prompt, API kwargs construction to the chat-completion helper, and streaming requests to interruptible_streaming_api_call.

Workflows

  • No first-pass workflow mapping yet; this concept still appears in the vault graph through articles and source evidence.
  • Building a Personal AI System with Hermes, Local Models, Automation, and a Shared Knowledge Base
  • Mastering Hermes Desktop: From Local Chat App to an Always-On Agent Control Center
  • Hermes Agent for Beginners: A Step-by-Step Route from Empty Server to Useful Automation