Requested by: @BYK
Merge target: (default)
Quick links:
Assign the accepted label to this issue to approve the release.
Targets
Checked targets will be skipped (either already published or user-requested skip). Uncheck to retry a target.
📋 Changelog
New Features ✨
Cch
Add seeds for Claude Code up to 2.1.146 by @github-actions in #436
Add seeds for Claude Code up to 2.1.145 by @github-actions in #412
Add seeds for Claude Code up to 2.1.144 by @github-actions in #379
Ui
Visual cost bars, savings hero, and daily trend chart by @BYK in #418
Chat-bubble conversation view for session page by @BYK in #398
Other
Migrate eval system to vitest-evals by @BYK in #441
Allow multiple recall tool calls per request (multi-turn recall) by @BYK in #404
Improve recall tool description + add cross-session cue eval scenarios by @BYK in #396
Inflate + replay compatibility — skip filler turns in gateway replay by @BYK in #393
Wire --inflate flag into eval CLI for 400K token scenario testing by @BYK in #386
Lightweight cross-session action tag context for curator by @BYK in #385
Scenario inflator for 400K token eval scenarios by @BYK in #384
Add local LLM provider support (vllm, ollama, llama.cpp, etc.) by @BYK in #383
Add temperature parameter to LLMClient interface by @BYK in #382
Lower echo threshold + clustering for broader behavioral pattern detection by @BYK in #377
Action tagging in distillation for implicit behavioral pattern detection by @BYK in #376
Vector similarity-based behavioral pattern detection by @BYK in #375
Add session recording, replay, and scenario filtering to eval harness by @BYK in #374
Add comprehensive eval suite for Lore's five key dimensions by @BYK in #369
Tool-call-aware cache warming + /lore:warm:* commands + UI controls by @BYK in #370
Bug Fixes 🐛
(cache-warmer) Tighten warming heuristics to reduce net negative spend by @BYK in #429
(eval) Build realistic Lore context for QA questions by @BYK in #414
(ui) Daily cost trend chart not rendering bars by @BYK in #426
1h TTL cache write pricing and dynamic savings/overhead wording by @BYK in #445
Resolve Sentry issues — JSONC config, ENOENT guard, Node.js port fallback, transient error filtering by @BYK in #443
Use compaction as primary baseline, fix threshold, update marketing by @BYK in #437
Context health note — signal omitted details instead of false assurance by @BYK in #434
Scripted interceptor for eval replay + distillation prompt tuning by @BYK in #433
Downweight knowledge in recall when session content exists by @BYK in #432
Store scripted assistant content in eval + session-affinity recall boost by @BYK in #431
Add amnesia mode and x-lore-no-store to prevent QA contamination in recall by @BYK in #430
Prevent tool_use/tool_result mismatch at gradient prefix/raw boundary (bug: inflated eval produces tool_use/tool_result mismatch errors through gateway #424 @BYK in #428
Resume sessions across client restarts instead of orphaning them by @BYK in #427
Create dev shims in gateway build so workspace consumers resolve bun exports by @BYK in #425
Drop callCount from dedup tests — still flaky after fix: use promise identity in dedup tests to eliminate cross-test flakiness #419 @BYK in #421
Use promise identity in dedup tests to eliminate cross-test flakiness by @BYK in #419
Build Bun-native ESM bundle for OpenCode plugin by @BYK in #415
Remove distilled=0 filter from temporal BM25 search by @BYK in #413
Add deepseek- dash-prefix routing to UPSTREAM_ROUTES by @rgutzen in #406
Add anti-imitation instruction to recall tool description by @BYK in #411
Inject X-Lore-Project header to eliminate cwd fallback warning by @BYK in #408
Code block alignment in Simplicity & Migration sections by @BYK in #407
Background worker rate-limit resilience for Claude Max by @BYK in #397
QA prompt instructs LLM to use recall tool by @BYK in #400
Pin user assertions in distillation to prevent loss in long sessions by @BYK in #394
Fall back to upstream when compaction summary generation fails by @BYK in #392
Pass local:true when starting gateway in-process to prevent hosted mode by @BYK in #390
Curator uses distilled observations when all messages are distilled by @BYK in #389
Judge prompt should not penalize extra correct information by @BYK in #388
Remove distilled=0 filter from temporal vector search by @BYK in #387
Preserve temporal embeddings on distillation + expose source IDs in recall by @BYK in #380
Improve distillation detail preservation for code-specific artifacts by @BYK in #378
Warmup cooldown timing + rising cost threshold + accurate TTL pricing by @BYK in #373
Improve curator preference detection and evolution handling by @BYK in #372
Documentation 📚
Replace waitlist form with Loops.so integration by @BYK in #447
Hero stat 13/20 → 2.6x total recall by @BYK in #444
Update website copy — sessions lasting days, crystal-clear memory by @BYK in #442
Add releasing instructions to AGENTS.md by @BYK in #438
Update eval results — context retention 3.9→4.6, +77% vs tail-window by @BYK in #435
Update website and README with context retention eval results by @BYK in #420
Update website hero stats with 400K eval results by @BYK in #405
Update README with 400K token eval results by @BYK in #403
Internal Changes 🔧
(deps) Bump pytest from 8.2.2 to 9.0.3 in /packages/core/eval/fixtures/projects/python-api in the pip group across 1 directory by @dependabot in #371
Disable eval workflow on PRs by @BYK in #446
Improve distillation detail retention at 400K+ token sessions by @BYK in #423
Remove deprecated eval infra and update docs with new eval results by @BYK in #422
Other
eval: add 2.3M-token mega-session scenario — Lore 4.0 vs Compaction 2.4 (+70%) by @BYK in #440
Requested by: @BYK
Merge target: (default)
Quick links:
Assign the accepted label to this issue to approve the release.
Targets
Checked targets will be skipped (either already published or user-requested skip). Uncheck to retry a target.
📋 Changelog
New Features ✨
Cch
Ui
Other
Bug Fixes 🐛
Documentation 📚
Internal Changes 🔧
Other