publish: BYK/loreai@0.23.0

[craft-deployer]

Requested by: @BYK

Merge target: (default)

Quick links:

• View changes
• View check runs

Assign the accepted label to this issue to approve the release.

Targets

• npm[@loreai/gateway]
• npm[@loreai/core]
• npm[@loreai/pi]
• npm[@loreai/opencode]
• npm[opencode-lore]
• github

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 2.1.145 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

• 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
• 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) by @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 by @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 📚

• 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
• 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