Use Hermes Agent with open-source models hosted in the EU

Hermes Agent is an open-source AI agent from Nous Research. It runs in the terminal, connects to editors and to messaging platforms, schedules jobs, and keeps memory and skills that improve as it works with you. Auxiliary tasks such as context compression, vision and session titles can each use their own model.

Hermes is built to be provider-independent. Beyond its built-in providers, it works with any endpoint that implements /v1/chat/completions, which makes it straightforward to run on open-weight models at inference.eu.

The model settings live in ~/.hermes/config.yaml under model. With provider: custom, Hermes sends requests to base_url, here the inference.eu base URL ending in /v1. key_env names the environment variable that holds your key; Hermes reads it from your shell or from ~/.hermes/.env. default is the model id used for new sessions.

The interactive hermes model command writes the same settings when you pick Custom endpoint. Hermes scopes each key to its own base URL, so your inference.eu key is only ever sent to inference.eu. Auxiliary tasks can be pointed at inference.eu too, with their own base_url and model in the auxiliary section.

Hermes plans, calls tools and keeps long-running memory, so it benefits from models that are strong at agentic work. kimi-k3 and glm-5.3 are good defaults for the main model: they reason well over multi-step tasks and return clean tool calls. For agent sessions that mostly write or review code, qwen3-coder-480b-a35b-instruct is a strong option.

For auxiliary tasks such as compression and titles, a fast model like gpt-oss-120b keeps costs and latency low. Setting those tasks to inference.eu as well means no part of a session falls back to another provider.

A 401 error means the key did not reach Hermes. Check that INFERENCE_EU_API_KEY is set in ~/.hermes/.env or in the shell, and that key_env names exactly that variable. A 404 usually comes from a base_url with an extra path; it should end in /v1.

Note that OPENAI_BASE_URL only affects the openai-api provider, not a custom endpoint, so set the URL in config.yaml. "Model not found" means default differs from the inference.eu id. If auxiliary tasks fail while the main model works, they are still pointed at another provider; give them a base_url and model on inference.eu.

With inference.eu as the custom endpoint, the prompts, files, memory excerpts and tool output Hermes sends for each turn are processed in EU data centres and never used for training. We sign a data processing agreement under Article 28 GDPR, share our subprocessor list and bill in euros.

Hermes can also connect to messaging platforms and outside tools, which are separate services with their own data handling. Point the main model and every auxiliary task at inference.eu, and review which gateways and tools you enable, so that model traffic stays with an EU processor.