Update enhanced-inference.ts

[HO-git]

remove the unnecessary errors for the providers that aren't wired up in the cache logic in the first place, but without messing up the original cache guard logic

[greptile-apps]

Greptile Summary

This PR prevents PricingNotConfiguredError from crashing in-flight inference calls for openai-compatible provider models. Previously, user-defined (UUID / customEndpoint) openai-compatible models would correctly bypass validatePricingAvailable, but would then throw deep inside getInputPricePerToken / getOutputPricePerToken when no hardcoded pricing entry existed — potentially aborting an already-completed LLM call. The fix gracefully returns 0 (with a warning log) for that provider, treating cost metrics as best-effort.

• getInputPricePerToken (line 703): added early return 0 + console.warn for openai-compatible before the PricingNotConfiguredError throw
• getOutputPricePerToken (line 737): same change mirrored for output pricing
• Missing newline at EOF: fixed
• Potential inconsistency: validatePricingAvailable is not updated to align — non-UUID / non-customEndpoint openai-compatible models will still be rejected at the handler gate even though the pricing functions now handle them gracefully; see inline comment for a suggested fix

Confidence Score: 4/5

• Safe to merge — the change is narrowly scoped and prevents real crashes for openai-compatible user-defined models without affecting billing accuracy for managed providers.
• The fix correctly targets the only scenario where the error was reachable (UUID/customEndpoint models bypassing validatePricingAvailable), returns a semantically correct $0 value, and leaves all other provider error paths untouched. The only concern is the minor inconsistency with validatePricingAvailable not being updated in parallel, which is a style issue rather than a correctness or security problem.
• No files require special attention — the change is self-contained within enhanced-inference.ts.

Important Files Changed

Filename	Overview
deprecated-claude-app/backend/src/services/enhanced-inference.ts	Adds a graceful return 0 (with a console.warn) in getInputPricePerToken and getOutputPricePerToken for openai-compatible providers when no pricing is found, preventing a PricingNotConfiguredError from crashing an in-flight inference call. Also fixes the missing newline at end of file.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Handler receives request] --> B{validatePricingAvailable}
    B -->|UUID / customEndpoint model| C[bypass → valid: true]
    B -->|Admin config found| C
    B -->|OpenRouter pricing found| C
    B -->|Hardcoded table hit| C
    B -->|No pricing source found\nnon-UUID / non-customEndpoint| D[valid: false → send error to client]

    C --> E[EnhancedInferenceService.streamCompletion]
    E --> F[LLM API call]
    F --> G[calculateCostSaved / calculateCostBreakdown]
    G --> H{getInputPricePerToken}

    H -->|Admin config| I[return configPricing.input / 1M]
    H -->|OpenRouter cache| J[return orPricing.input / 1M]
    H -->|Hardcoded table hit| K[return price / 1M]
    H -->|No price found\nprovider = openai-compatible| L["NEW: console.warn + return 0"]
    H -->|No price found\nother provider| M[throw PricingNotConfiguredError]

Loading

_{Last reviewed commit: 9f1ba06}

[greptile-apps]

validatePricingAvailable not updated to match

The upstream validatePricingAvailable function (line 232) does not have a corresponding bypass for openai-compatible providers. This means a non-UUID, non-customEndpoint openai-compatible model without any configured pricing will still be blocked at the handler's validation gate before reaching these functions — so the new return 0 path is currently only reachable by user-defined (isUUID / hasCustomEndpoint) models.

This is consistent with the original intent (user-defined models bypass the pricing gate and accept responsibility for their own costs), but the two behaviours are not obviously co-located. Consider adding a matching bypass or comment in validatePricingAvailable to make it clear that openai-compatible models are expected to gracefully fall back to $0 pricing:

// In validatePricingAvailable, after step 3:
if (model.provider === 'openai-compatible') {
  // openai-compatible models are self-hosted; cost tracking is best-effort
  return { valid: true };
}

This would make the two functions consistent and prevent future confusion if validatePricingAvailable is called independently.