5 Commits

Author	SHA1	Message	Date
Ke Jin	1a8d38655d	fix(reasoning): preserve and replay assistant turns (#10045)	2026-08-13 07:52:12 -03:00
stanley	552f2e1563	fix(backend): stop reasoning replay placeholder from self-poisoning (#9573) (#9610) ... Merge-train validated (tip 6ce4effef8). Vitest failures confirmed as base-red (#9679).	2026-08-07 20:54:23 -03:00
mi	94682d2a76	perf(ram): reduce server memory footprint (#2903) ... Integrated into release/v3.8.7	2026-05-29 13:21:36 -03:00
diegosouzapw	f00e9a1272	fix(postinstall): extend native module repair to cover wreq-js for pnpm global installs (#1634) ... - Add fixWreqJsBinary() to postinstall.mjs with 3-strategy repair: 1. Copy platform binary from root node_modules 2. Copy entire rust/ directory (all platform binaries) 3. npm rebuild wreq-js fallback - Fixes macOS arm64 global pnpm installs shipping only Linux binaries - Adds reasoning replay cache feature for DeepSeek V4 (#1628) - Updates CHANGELOG.md with both fixes	2026-04-26 20:15:47 -03:00
diegosouzapw	cabdfb04e7	fix(cache): replay cached reasoning for tool-calling think models ... Prevent upstream 400 failures when clients omit prior reasoning_content in multi-turn tool-calling conversations. Capture reasoning_content from streaming and non-streaming assistant responses, persist it in a memory-plus-SQLite cache keyed by tool_call_id, and re-inject it on later requests when available. Add the reasoning cache migration, service layer, authenticated API endpoints, dashboard tab, and unit coverage to support inspection, cleanup, and crash recovery.	2026-04-26 18:53:54 -03:00