v1.21.2 — Sync fix on Chrome for Android

Shipped 2026-04-20

• Cross-device sync now works on Chrome for Android — a pre-flight check was wrongly blocking it.
• Updated stale Settings copy that still mentioned the retired Electron build.

v1.20.1 — Bug fixes

Shipped 2026-04-16

• Custom API key now works after page reload (#124).
• Context cards (diet, sleep, stress, and the rest) update on screen immediately after saving — no reload needed (#123).
• Closed a same-origin bypass on the dev-server proxy route (#119 follow-up).

v1.20.0 — Custom Knowledge Source

Shipped 2026-04-15

• Connect your own RAG knowledge endpoint to the Interpretive Lens — the AI grounds its analysis in research, clinical guides, or documents you provide.
• Works across chat, multi-persona discussions, and the focus card.
• Bug fixes: GPT-5 / o-series support in Custom API (#114), Custom API settings now persist across backup + sync (#116).

This is the release that introduced the Custom Knowledge Source — the foundation that v1.21.x builds on with multi-library support, WebGPU embeddings, and expanded parsers.

v1.19.2 — Sync Context Fix

Shipped 2026-04-14

• Specialty lab data (Fatty Acids, OAT, etc.) now syncs completely even when excluded from AI chat
• Fixed a bug where sync silently failed to push lab data to the gateway

v1.19.0 — Precise Supplement Dosing

Shipped 2026-04-12

• New "Doses/day" field on supplements — set once as the default multiplier for every ingredient in a combo product
• Per-ingredient ×/day override — for stacks where different ingredients have different schedules (e.g. Mg Bisglycinate 1× AM + Taurate 2×/day)
• Live daily total next to each ingredient as you type — catches entry mistakes before saving
• AI receives explicit daily totals (no more guessing the math from free-text dosing)
• Comma-decimal parsing fixed — "5,4 mg" now parses correctly instead of dropping the decimal
• Impact analysis cache is now per-supplement — editing one supp re-analyzes only that supp, not the whole batch

v1.21.9 — chat.js refactor

Shipped 2026-04-20 · Part 3 of the v1.21.7–9 code hygiene batch

Internal refactor — no user-visible change. Chat behavior and streaming hot-path are untouched.

• Image attachments extracted to chat-images.js (212 lines).
• Thread management extracted to chat-threads.js (478 lines).
• chat.js went 3410 → 2779 lines (-18%).
• Back-references use the existing window.fn() pattern to avoid circular imports.

v1.21.8 — CLAUDE.md trim

Shipped 2026-04-20 · Part 2 of the v1.21.7–9 code hygiene batch

Internal contributor-facing change — no user-visible change.

• CLAUDE.md (the contributor architecture doc) trimmed 18.4 KB → 13.2 KB (-29%).
• Regrouped the 47-module roster by concern (core / data / AI / views / chat / sync / knowledge-base) instead of a flat 42-bullet list.
• Dropped per-function enumerations that duplicated the source — the code is the reference.

v1.21.7 — Dead-code cleanup

Shipped 2026-04-20 · Part 1 of the v1.21.7–9 code hygiene batch

Internal cleanup — no user-visible change.

• Removed 5 unused Routstr/PPQ wallet helpers (-105 lines net).
• Dropped 4 stale handleLocalLens* aliases and 3 debug getters with no callers.
• Pruned 19 unused imports.
• Every deletion grep-verified zero-caller before removal.

v1.21.6 — Docs accuracy sweep

Shipped 2026-04-20

• Custom API added to the provider list (it's been supported since v1.16.1 — the table just missed it).
• Personal-agents and Agent Access setup guides updated to the one-command installer.

v1.21.5 — Security hardening

Shipped 2026-04-20

• Attribute-safe HTML escaping across every user-authored field — closes a class of self-XSS when strings contained a bare double quote.
• Tightened the Vercel AI proxy so it can't be abused to reach private networks or cloud-metadata services.
• New tests covering the markdown renderer (every streamed AI response passes through it).