The Hermes Agent only says "I will execute the command" but does not actually call the tool, and the common reason is that the tool calling of the current model or custom endpoint is not compatible. Some users in the community have recently encountered this problem with small local models: the model mimics the tool output, but does not initiate real tool calls in the format that Hermes needs.
How to confirm
- First, switch to the official more commonly used provider, such as Nous, OpenRouter, Codex, or Anthropic, and test the same task.
- If the tool can be adjusted after changing the model, the problem lies in the tool call ability of the original model.
- Custom endpoints must be compatible with OpenAI-style chat completions and tool call parameters.
Don't be fooled by "looks like tool output"
Real tool calls appear in Hermes' progress, logs, or front-end tooltips; If it's just the model itself writing a paragraph "I'm going to run ls", that's just text, not a tool.
In a word: just say and don't do, first change to a model that supports tool calling to verify it.
Official open source address: https://github.com/NousResearch/hermes-agent; Official document entry: https://hermes-agent.nousresearch.com/.
