Skip to content

v2.2.0

Latest

Choose a tag to compare

@trefeon trefeon released this 17 Sep 18:49
· 3 commits to main since this release

Fix: false rejections no longer kill models

  • Upstream sensitive_words_detected 500s are prompt content-filter rejections, not outages: passed through with no retry, no model lock, no breaker penalty (IsContentFilterRejection).
  • Recovery probe carries no API key, so any 4xx (keyless 401) now clears the model lock instead of extending it forever. Previously one bad prompt hid deepseek-v4-flash from /v1/models for 10+ hours.
  • availableModels now counts healthy models, matching /v1/models.

Feat: default safe persona + gpt-6-astra

  • Unset INJECT_SYSTEM_PROMPT injects a benign engineering-framing persona into every request (custom text overrides, empty disables).
  • Added gpt-6-astra (1.05M ctx, 128K out) to the default model list; specs researched via OpenRouter.

Found while mapping: the provider flags the canonical 'You are a helpful assistant' string in user content — reword to 'coding/useful assistant'.