hermes - 💡(How to fix) Fix [Feature Request]: Configurable pricing and billing routes for custom providers [1 comments, 2 participants]

Summary

Please consider adding first-class configuration for pricing and billing metadata for custom providers.

Today, deployments that use custom provider routes or Anthropic/OpenAI-compatible proxies may need to patch source code so usage accounting can:

• classify a custom provider under the correct billing family
• map a runtime provider to a separate billing_provider
• define per-model input/output prices
• define cache read/write pricing behavior
• handle model-family fallback for custom model names

This is difficult to maintain because every upstream upgrade can overwrite local pricing and billing patches.

Proposed capability

Allow custom providers to declare pricing and billing behavior in config, for example:

pricing:
  providers:
    custom:my-anthropic-proxy:
      billing_provider: my-anthropic-billing
      model_family: anthropic
      currency: USD
      models:
        claude-opus-4-7:
          input_per_mtok: 25.0
          output_per_mtok: 125.0
          cache_read_multiplier: 0.10
          cache_write_multiplier: 1.25
      fallback:
        match:
          - claude
          - opus
        input_per_mtok: 25.0
        output_per_mtok: 125.0

The exact shape can differ, but the important part is that pricing/billing metadata can be supplied without editing Hermes source.

Use cases

• Private Anthropic-compatible gateways
• OpenAI-compatible aggregators with different billing rates
• Enterprise providers where the runtime provider name differs from the billing entity
• Local deployments that need accurate cost reporting across multiple providers
• Providers with cache read/cache write billing multipliers

Expected outcome

A custom provider should be able to work end-to-end with accurate usage accounting using only config files. Upgrades should not require reapplying source patches for pricing tables or billing-provider routing.

Why this matters

Hermes already supports flexible provider routing. Usage accounting should have a matching extension point, otherwise custom-provider deployments end up forking core pricing code for what is fundamentally deployment configuration.