Skip to main content
Hermes Agent is Nous Research’s open-source terminal agent with persistent memory and learned skills. It works with any endpoint that implements /v1/chat/completions, so it can run entirely on local Bonsai.

​
Step 1: Serve Bonsai

Any OpenAI-compatible Bonsai endpoint works. Pick one: 
./scripts/start_llama_server.sh
# Base URL: http://localhost:8080/v1

./scripts/start_mlx_server.sh
# Base URL: http://localhost:8081/v1
Agent work leans hard on tool calling and multi-step instructions; prefer 27B, which is trained for native tool calling, or ternary 8B if memory is tight (see which model for agents).

​
Step 2: Install and configure Hermes

1

Install

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
Windows has a native PowerShell installer; see the Hermes docs.
2

Run the setup wizard

hermes setup
When asked for an inference provider, choose Custom OpenAI-compatible endpoint and enter:
PromptValue
API base URLhttp://localhost:8080/v1 (or your endpoint from step 1)
API keyany non-empty string, e.g. not-needed
Modelauto-detected via GET /v1/models; accept the Bonsai entry
Display namee.g. bonsai-local
Settings land in ~/.hermes/config.yaml (secrets in ~/.hermes/.env). To change just the model later, run hermes setup model, or edit the config directly with hermes config edit. ‹TODO: verify wizard prompts against the current Hermes release›
3

Launch

hermes
Give it a task. Requests appear in your Bonsai server log, confirming the agent is running fully locally.