Skip to content

v2.7.31

Choose a tag to compare

@berges99 berges99 released this 01 Oct 10:24
78a706d

What's Changed

  • fix: scope fallback provider params per model entry (0d66a94)
  • Preserve recorded context usage for legacy conversations (ba501a8)
  • Describe compaction in terms of the context measurement (5ad61ad)
  • Pair each context measurement with the messages its request covered (87f507d)
  • Salvage only the latest LLM span so a failed call does not duplicate the previous output (3e46d41)
  • Cut the unsent estimate at the last model output and count the tail in every branch (5c753ea)
  • Drop comments that restate the code (7773fbb)
  • Document the compaction signal: last call, unsent messages, run model window (6d74bbd)
  • Take the last LLM span directly and make the run-model test fail on a wrong model (b559952)
  • Count the messages sent after the last reported usage toward utilization (efab932)
  • Measure compaction utilization against the last LLM call and the run's model (7cf8251)

Full Changelog: v2.7.30...v2.7.31