bug: unhandled 'e is not iterable' error on OpenRouter chat endpoint

[whoabuddy]

Problem

The /inference/openrouter/chat endpoint throws an unhandled "e is not iterable" error, likely from destructuring or iterating over a malformed OpenRouter API response.

Evidence from production logs (2026-03-03)

2026-03-03T01:01:46 | /inference/openrouter/chat | Unhandled error | e is not iterable

Root cause (likely)

OpenRouter occasionally returns non-standard response shapes (missing choices array, error objects instead of completions, etc.). The code likely does something like for (const choice of response.choices) without checking that choices exists and is iterable.

Fix

Add defensive checks before iterating over OpenRouter response fields:

if (!response.choices || !Array.isArray(response.choices)) {
  // return structured error instead of throwing
}

Additional: invalid model ID errors

Also from logs (2026-02-26), callers are passing x-ai/grok-4.1-mini which OpenRouter rejects. Consider:

• Validating model IDs against a known list before forwarding to OpenRouter
• Returning a clear 400 error instead of proxying OpenRouter's raw error

Priority

Medium — not currently recurring but will bite again with any OpenRouter API change.

[whoabuddy]

Closing — this was fixed in #67 / #69 (shipped Mar 12–15).

estimateInputTokens now guards against non-iterable input. Production logs confirm zero recurrence since Mar 12 across 10+ days of traffic.