Modular provider for Hermes Agent

Adds Modular Cloud to Hermes Agent as the modular provider: open models on Modular's OpenAI-compatible API (https://api.modular.com/v1).

hermes plugins install modular
export MODULAR_API_KEY=...
hermes model            # pick Modular, then a model

Or for one run: hermes -z "hello" --provider modular -m deepseek/deepseek-v4.1-flash.

Models

The model list is fetched from GET /v1/models with your key and narrowed to the models this plugin describes. Each has a context window, vision and tool support, and reasoning controls measured against the API. Prices are in Modular's public model catalog: https://console.modular.com/api/v1/docs/models.

Reasoning levels map onto each model's own controls:

• A level the model doesn't offer is lowered to the nearest one it does, never raised.
• "Off" uses reasoning.enabled = false where that stops thinking. Where nothing stops it, the plugin sends the lightest effort.
• With no level set, the request carries no reasoning field, so the model's own default applies.

What this plugin does

• Network: it sends requests only to api.modular.com, for chat completions and the model list.
• Credentials: it reads only MODULAR_API_KEY, through Hermes.
• Nothing else: no tools, hooks, background processes or files, and no telemetry. Requests carry Hermes' standard attribution headers (HTTP-Referer, X-Title).

Updates

The plugin is generated by Modular's registry tool from the public catalog and live API measurements, and every model is run end to end in Hermes before a release. A new or changed model ships as a new commit here, followed by a pin update in Hermes' plugin catalog.