Skip to content

0.8.79

Choose a tag to compare

@github-actions github-actions released this 06 Jun 00:21
· 278 commits to main since this release
v0.8.79
b88653f

Added

  • Durable cross-process rate-limit admission. Harn now exposes
    durable_rate_limit_acquire(options) for SQLite-backed sliding-window quota
    reservations across processes, with atomic multi-bucket admission, structured
    timeout results, and mock-clock-friendly tests (#1873).

Changed

  • LLM provider rate limiting now consumes catalog rate_limits metadata,
    including model-specific RPM/TPM and route concurrency, with environment and
    llm_rate_limit overrides for paid/custom quotas.
  • Durable LLM rate-limit admission. Catalog and runtime LLM RPM/TPM
    limits now use shared SQLite admission by default across Harn processes, so
    parallel eval runners and worker fleets respect one provider/model quota
    without relying on per-process sleeps or ad hoc environment-only guardrails
    (#1873).

Install / Upgrade

cargo install harn-cli

Full Changelog: v0.8.78...v0.8.79