Provider Archetypes: Subscription, BYO-Key, Local

+15 Mana ✨
Provider Integrations(DOCUMENTATION)

Introduction

The dozens of model providers Hermes supports look chaotic at first glance, but they fall into three archetypes: subscription, BYO-key, and local. Reasoning at the archetype level, not at the per-provider level, keeps your config simple and your decisions transferable. When a new provider launches next quarter, you will know which archetype it is and how to slot it in.

Key Concepts

  • Subscription archetype: You pay one provider a monthly fee for bundled access. Examples: Nous Portal subscription, vendor-managed packages.
  • BYO-key archetype: You bring your own API key to a provider; you pay per use. Examples: OpenAI, Anthropic, OpenRouter.
  • Local archetype: You run the model yourself, paying with hardware instead of tokens. Examples: Ollama, llama.cpp, vLLM, self-hosted endpoints.

Real World Context

If you understand the archetype, the operational implications follow:

  • Subscription = predictable bill, capped by what is bundled, fewest setup steps.
  • BYO-key = pay-per-use bill, full provider catalog, more setup.
  • Local = no bill, capability ceiling = your hardware.

A team picking a provider does not need to compare twenty endpoints, they need to pick an archetype and then pick the best fit inside it.

Deep Dive

Archetype comparison

text
Archetype     Bill type        Setup       Capability ceiling
────────────  ───────────────  ──────────  ────────────────────────
Subscription  Flat / bundled   Lowest      Provider's catalog
BYO-key       Per-token        Medium      Provider's catalog
Local         Hardware (sunk)  Highest     Your model + hardware

Subscription, when it wins

When you do not want to think about per-call cost. Nous Portal's subscription gets you running in minutes with Hermes models, no key management, predictable billing.

BYO-key, when it wins

When you want flexibility. OpenRouter's "one key, many models" approach is the canonical example: switch models without switching keys. Direct provider keys (Anthropic, OpenAI) win when you need a specific model and want the cleanest billing trail.

Local, when it wins

When data must not leave your hardware, when you have GPU headroom, or when you want zero marginal cost on a high-volume workflow. Ollama is the most common entry point for local because it abstracts model downloads and GGUF formats.

Common Pitfalls

  1. Comparing across archetypes by per-token price, Subscription and local do not have a per-token price in any meaningful sense. Compare archetypes by total cost of ownership for your workload.
  2. Assuming local means free, Hardware, electricity, and the time you spend tuning a runtime are all real costs.

Best Practices

  1. Pick an archetype first, then a provider inside it, This keeps your decision transferable when a specific provider's offer changes.
  2. Mix archetypes deliberately, A subscription for steady-state, BYO-key for spiky needs, local for sensitive prompts is a common mature setup.

Summary

  • Three provider archetypes: Subscription, BYO-key, Local.
  • Each archetype has a different bill type, setup cost, and capability ceiling.
  • Reason at the archetype level to keep decisions transferable.
  • Mature Hermes installs often combine all three.

Code Examples

yaml
# A mature Hermes config can combine all three archetypes
providers:
  - id: nous-portal       # SUBSCRIPTION, predictable bill, zero setup
    plan: pro
  - id: openrouter        # BYO-KEY, flexible model catalog under one key
    api_key_env: OPENROUTER_API_KEY
  - id: ollama-local      # LOCAL, privacy-sensitive prompts stay on disk
    model: hermes-3-llama-8b
✓ Completed