Hermes Agent Local Setup Tips

running Hermes Agent on a local model, here are five tips: (1) why do replies feel weak on a small local model? check the context window first. Hermes Agent needs at least 64,000 tokens of context for agent work with tools, and when a conversation outgrows the server's window, most servers drop older messages without a warning. (2) which model should I run on my machine? the Hermes Index scores models on tasks run inside Hermes Agent, with task completion and cost per task. tick "Open weights only" to see the ones you can run yourself. (3) do I need Docker to run it safely? Docker is optional. by default Hermes checks every command against a list of dangerous patterns before it runs, and risky ones go through an approval step. an approval prompt left unanswered for 300 seconds is denied, and the file-editing tools are blocked from ~/.ssh and .env files. to run every command inside a container, add this to config.yaml: ``` terminal: backend: docker ``` (4) tools run on a cloud model, how do I get them running locally? tool calls need support from the model and from the server. Ollama has tool calling on by default, so check your model supports it with `ollama show <model>`. llama.cpp needs `--jinja` on its startup command, and LM Studio needs version 0.3.6 or newer with a model that has native tool support. (5) can I run it for free? yes. a local model needs no API key and no subscription. web search works on a fresh install with zero setup too: with no key set, Hermes rotates across the free tiers of several search vendors. to pin a free backend, set `web.backend` to `ddgs` (DuckDuckGo) or `searxng` (self-hosted) in config.yaml.