Memorix v1.10.3

What's Changed

• feat(pi): support native MCP configuration in Pi 1.x by @noirgif in #367
• docs: close out 1.10.2 and record Pi MCP landing by @AVIDS2 in #370
• feat(llm): add optional Runcrate memory preset by @heydryft in #372
• fix: close codegraph, lazy vectors, and search regressions by @AVIDS2 in #376

New Contributors

• @heydryft made their first contribution in #372

Full Changelog: v1.10.2...v1.10.3

Memorix v1.10.2

What's Changed

• docs: close out the 1.10.1 release by @AVIDS2 in #364
• fix(hermes): register skill file paths by @RaviTharuma in #366
• fix(mcp): return tool output in structuredContent for search and session start by @noirgif in #368
• fix(search): bound Lance index growth (1.10.2) by @AVIDS2 in #369

Full Changelog: v1.10.1...v1.10.2

Memorix 1.10.1

1.10.1 — Retrieval health and provider hardening

• Adds privacy-safe derived-view receipts for FTS, Orama, embedding, and CodeGraph freshness and rebuild diagnostics.
• Adds the opt-in Cheaper Inference memory LLM preset with dedicated credential isolation.
• Records reproducible v1.9.11 to 1.10 benchmark evidence.

Validation: 344 test files, 3232 tests passed, 5 skipped; all 7 release CI checks passed; npm signature audit passed.

Memorix 1.10.0

1.10.0 — Trustworthy context and measured retrieval\n\n- Adds a reproducible 20-case context evaluation across continuity, retrieval, safety, runtime binding, and integration parity.\n- Adds request-level retrieval receipts with candidate/selected/omitted counts, token-budget omissions, and fallback diagnostics across compact, MCP, and CLI JSON surfaces.\n- Records and verifies 40k/100k local capacity gates across SDK, HTTP MCP, hooks, SQLite persistence, and reopen paths.\n- Keeps SQLite canonical, local-first defaults, and existing MCP/agent integration contracts.\n\nValidation: 344 test files, 3225 tests passed, 5 skipped; all 7 release CI checks passed; packed package smoke passed.

v1.9.12

Fixed

• npm publication now requires registry signature verification and fails closed when provenance publication cannot complete.
• The publish workflow no longer retries with an unprovenance package.
• Opt-in background updates disable npm lifecycle scripts during global installation.

See the full CHANGELOG for details.

v1.9.11

Fixed

• CJK full-text search now supports bounded Chinese, Japanese, and Korean character n-grams while preserving Latin tokenization.
• Refreshed runtime dependency resolutions and hardened the production dependency audit baseline.
• Large-store alias migration, alias-aware search visibility, hybrid lexical recovery, and search CLI argument parity are fixed.
• Pi hooks on Node 26 pause and unref stdin without closing file descriptor 0.

Added

• Qwen Code hook installation with official settings preservation and project/global guidance.
• Optional API Route OpenAI-compatible memory processing with isolated credentials and CLI/TUI setup.

See the full CHANGELOG for details.

v1.9.10

What's Changed

• fix(opencode): restore 1.x plugin compatibility by @AVIDS2 in #326

Full Changelog: v1.9.9...v1.9.10

v1.9.9

What's Changed

• feat(llm): add optional Requesty memory preset by @Thibaultjaigu in #314

New Contributors

• @Thibaultjaigu made their first contribution in #314

Full Changelog: v1.9.8...v1.9.9

v1.9.8

What's Changed

• fix: support OpenCode v2 plugin contract by @AVIDS2 in #319

Full Changelog: v1.9.7...v1.9.8

Memorix v1.9.7

Fixed

• Retry transient SQLite FTS5 initialization contention before disabling lexical search.
• Preserve large-store lexical search when embeddings fail at runtime, disappear, or change dimensions.
• Synchronize npm, MCP Registry, and packaged agent metadata for the 1.9.7 patch release.

Contributors: PR #315 and PR #316; follow-up hardening in PR #317.