Skip to content
Fast Inference
DOCS / Hermes
Integration

Hermes

Point Hermes Agent at Fast Inference with its supported custom OpenAI-compatible provider configuration.

Hermes Agent supports direct OpenAI-compatible endpoints through its own model configuration. Fast Inference's one-command fast hermes adapter is still in preview, so use the verified manual path below today.

Configure Hermes manually

  1. Create a project API key

    Create a key in API keys, then store it in the Hermes environment file instead of the YAML document.

    bash
    mkdir -p ~/.hermes
printf 'OPENAI_API_KEY=%s\n' "$INFERENCE_API_KEY" >> ~/.hermes/.env

    Make sure INFERENCE_API_KEY is set in the current shell before running the command. Avoid appending the same setting more than once.

  2. Select the custom endpoint

    Edit ~/.hermes/config.yaml and merge this model configuration with your existing settings:

    yaml
    model:
  default: glm-5.2
  provider: custom
  base_url: https://api.inference.net/v1

    Hermes documents config.yaml as the preferred place for custom provider, model, and base URL settings. The direct base_url takes precedence over a named provider.

  3. Restart and verify

    Restart Hermes, open a new session, and send a short prompt. Confirm the model and request under Usage.

Choose a model

Copy a callable ID from the model catalog. The configured model must support the tools and context size required by your Hermes workflow.

For auxiliary tasks such as vision or delegation, Hermes has separate nested configuration blocks. They inherit the main provider only when explicitly set to do so; configure those endpoints separately if you want them routed through Fast Inference.

Revert the manual setup

Before editing, keep a copy of your current files:

bash
cp ~/.hermes/config.yaml ~/.hermes/config.yaml.before-fast-inference
cp ~/.hermes/.env ~/.hermes/.env.before-fast-inference

To disconnect, restore those copies or remove only the model values and OPENAI_API_KEY entry you added. Restart Hermes afterward.

Troubleshooting

  • Validate YAML indentation if Hermes cannot load the configuration.
  • Confirm ~/.hermes/.env is readable only by your user and contains one key.
  • Use https://api.inference.net/v1, including the /v1 suffix.
  • Choose a currently callable model ID when the gateway reports model not found.