publish: BYK/loreai@0.13.2

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

Checked targets will be skipped (either already published or user-requested skip). Uncheck to retry a target.

---

📋 Changelog

New Features ✨

Gateway

• Worker model discovery via /v1/models API + models.dev pricing by @BYK in #142
• Transparent recall tool interception by @BYK in #139
• Per-session auth isolation + OAuth Bearer support by @BYK in #135
• Replace content markers with fingerprint-based session tracking by @BYK in #125

Worker Model

• Prefer non-reasoning models and disable thinking on worker calls by @BYK in #122
• Wire up dynamic model selection on session idle by @BYK in #118

Other

• (db) Persist r_compression and c_norm on distillations (schema v12) by @BYK in #116
• (gradient) Range-aware file read deduplication by @BYK in #120
• Emit '[Searching memory...]' indicator for Case 2 (mixed tools) by @BYK in a5a8e215
• Local embeddings, exact keyword boost, and pattern extraction by @BYK in #136
• Batch queue for Anthropic Message Batches API (50% worker cost savings) by @BYK in #134

Bug Fixes 🐛

• (ci) Include @loreai/gateway in pack and publish targets by @BYK in a5c6bc0c
• (db) Use os.homedir() for cross-platform data directory resolution by @BYK in #131
• (gradient) Deterministic timestamps in sanitizeToolParts by @BYK in #124
• (release) Include gateway in bump-version.sh for Craft publish by @BYK in eefdb53f
• Resolve typecheck error in content-based fingerprint and increase hysteresis to 15% with high-water mark budget by @BYK in c8dcf035
• Eliminate gradient cache busts with deterministic IDs, offset-from-end pins, and cost-aware cap by @BYK in 321accc3
• Stabilize gradient cache with budget snapshot, hysteresis, and per-session LTM by @BYK in #144
• Type errors in Case 2 integration tests by @BYK in 4e2282d8
• Deprecated worker models, invalid cache TTL format, and batch error logging by @BYK in #140
• Prevent re-entrant deadlock in event hook (MaxListenersExceeded hang) by @BYK in #138
• Reduce cache-bust cost with idle threshold, sticky layers, bust tracking, and meta-distill gating by @BYK in #132

Internal Changes 🔧

• (gradient) Batch distillation consumption at turn boundaries by @BYK in #123
• Use models.dev unified JSON API for cost data by @BYK in #143
• Case 2 (mixed tools) integration tests for recall interception by @BYK in #141
• Migrate knowledge entries from AGENTS.md to .lore.md by @BYK in 107346f1
• Remove gateway observer hooks and dead recorder exports by @BYK in #130
• Remove read-time temporal enrichment from distillation prefix by @BYK in #129
• Update license to FSL-1.1-Apache-2.0 by @BYK in #126

[github-actions]

Publish failed. View workflow run

[github-actions]

Publish failed. View workflow run