Skip to content

v1.3.2

Choose a tag to compare

View all tags
@adham90 adham90 released this 28 Jan 18:32
· 270 commits to main since this release
v1.3.2
18ea4b0

What's New

Per-Model Error Tracking

Reliability middleware now tracks individual attempts per model using AttemptTracker. Each execution record includes detailed per-model attempt data (model_id, timing, tokens, error details, short-circuit status) in the attempts JSON column — fully visible on the dashboard.

Non-Fallback Error Handling

Programming errors (ArgumentError, TypeError, NameError, NoMethodError, NotImplementedError) now fail immediately without trying fallback models. Configurable via non_fallback_errors DSL.

Smart Retry Strategy

When fallback models are configured, transient errors skip retries and move directly to the next model. Retries only occur when no fallbacks are available.

Response Persistence

Instrumentation middleware can now persist LLM response content with automatic redaction of sensitive data.

Bug Fix: Gemini Quota Errors

Added "quota" to default retryable rate limiting patterns so Gemini quota exceeded errors properly trigger model fallback.

Assets 2
Loading