You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Oct 11, 2026. It is now read-only.
Repository navigation
This repository was archived by the owner on Oct 11, 2026. It is now read-only.
Support @receptron/laya as an optional local alternative to TypeSafe/Jev for effort selection. Keep the existing request rewrite, fallback, and decision logging behavior.
Map Laya’s typed answers to the router’s supported effort levels without changing the outbound model request contract.
Make backend selection explicit. Preserve TypeSafe/Jev as the default and document Laya’s first-run model download, cache, and memory requirements.
Test effort mapping, initialization failure, timeout, fallback, cancellation, and concurrent requests with a mocked Laya backend. Verify that local mode does not send conversation excerpts to TypeSafe/Jev.
Smoke-test a real Laya model load and one routed request before calling support complete.
Activity
robertn702 commentedon Sep 30, 2026
Implemented in #76: optional in-process Laya classification with explicit backend selection, shared fallback/cancellation behavior, privacy coverage, documentation, and a successful real-model routed smoke.
robertn702 commentedon Oct 11, 2026
Laya is supported as a classifier in reasoning-router, which replaces this project: https://github.com/robertn702/reasoning-router