hermes-together-provider

A Hermes Agent model-provider plugin for Together AI — an OpenAI-compatible inference API with 270+ serverless open-weight models.

Install

hermes plugins install pepijndevos/hermes-together-provider

Or manually: drop __init__.py and plugin.yaml into $HERMES_HOME/plugins/together/ (flat, matching the kind: model-provider discovery rule — see the Model Provider Plugins guide).

Then set your key in ~/.hermes/.env:

TOGETHER_API_KEY=your-key-here

Get a key at https://api.together.ai/settings/api-keys.

What it adds

• together provider (aliases together-ai, togetherai) selectable via hermes model, --provider together, or /model <model-id> --provider together.
• Live model-catalog discovery from Together's /v1/models endpoint — this plugin overrides fetch_models() specifically so model-switch validation routes through the shape-aware catalog fetch instead of a generic probe that assumes the OpenAI {"data": [...]} envelope (Together's /v1/models returns a bare JSON array; see upstream hermes-agent#141782).
• A curated fallback_models list (current-generation, good price/performance picks) used when the live catalog is unreachable.

Disclosure

• Network calls: only to https://api.together.xyz/v1 (inference) and its /models endpoint (catalog discovery), both using the API key you provide.
• No credentials are read from disk directly — the API key is resolved through Hermes's own provider-credential routes (TOGETHER_API_KEY / TOGETHER_BASE_URL env vars), the same mechanism every built-in provider (OpenRouter, Fireworks, GMI, ...) uses.
• No telemetry, no background processes, no self-updating code.
• No core overrides — registers through the public register_provider() / ProviderProfile API only.

Why standalone, not in-tree

hermes-agent's contribution policy treats third-party vendor model-provider integrations as plugins, not core-tree additions (see CONTRIBUTING.md and the prior in-tree PR, closed as not-planned). This repo ships it the way the policy asks for.

License

MIT