Ollama Responses API fallback fires 1,598 Sentry errors — ollama has no /v1/responses

[graycyrus]

Summary

The inference layer attempts Responses API fallback on ollama, but ollama has no /v1/responses endpoint. Every attempt returns 404 and fires a Sentry error event — 1,598 events and escalating.

Problem

What happened: When a chat completions request to ollama fails, the code falls back to the Responses API (/v1/responses). Ollama doesn't have this endpoint, so every fallback returns 404 page not found. Each 404 fires a Sentry error.

Expected: Either:

1. Don't attempt Responses API fallback for ollama (it will never work)
2. Or suppress the 404 from Sentry since it's an expected fallback failure

Impact: 1,598 events in Sentry (TAURI-RUST-59Y), escalating. Tags: provider=ollama, operation=responses_api, model=gemma3:1b-it-qat.

Version / Platform: openhuman@0.56.0, macOS 26.5.0

Solution (optional)

Check the compatible provider's fallback logic in src/openhuman/inference/provider/compatible.rs:

1. Is there a flag to disable Responses API for specific providers?
2. Can ollama be marked as "no Responses API support"?
3. Or add "404 page not found" from the Responses API fallback to the config-rejection/noise suppression list

Acceptance criteria

• Repro gone — Ollama requests don't fire Sentry errors for Responses API 404
• Regression safety — Unit test
• Diff coverage >= 80%

Related

• Sentry: https://sentry.tinyhumans.ai/organizations/tinyhumans/issues/6030/
• Prior art: PR fix(observability): suppress 429/rate-limit Sentry noise — TAURI-RUST-3 #2899 (rate limit suppression), PR fix(observability): demote memory-store PII rejection errors from Sentry to warn #2850 (PII suppression)