Changelog

Unified article search semantics

Shared search kernel so chat, discovery, and automation share the same retrieval semantics.

Keyword normalization & unread / retrieval integrity

Refactored keyword normalization and unread/retrieval integrity across sidebar, discovery stats, and automation context.

Flow scroll stability & session cleanup

Pause idle prefetch / optimistic merges while scrolling; sticky columns; abort generation and reset filters after clearing chat.

JA/KO recommended feeds & catalog refactor

Added Japanese and Korean recommended feeds, expanded the catalog, and refactored feed settings categories.

Locale-aware default tag sets

Default tag sets adapt to UI locale with broader coverage.

Broad multi-language i18n coverage

Background database handling via db.worker

Full-text extraction services

AI report dashboard shipped

Clerk auth components and localization

Graph portal and AI report components with SQLite

Admin dashboard and telemetry cleanup

Modular build assets and localization polish

Performance: image loading, parallel queries, first paint

Large-scale subscription and query-layer rewrite

SQLite WASM with OPFS async proxy

Build toolchain migrated to Rspack

AI chat, ticker, and custom prompt management

Automation task management and pricing table

Sidebar redesign and multilingual store descriptions

V3 chat interface rewrite

SmartInput, CommandPalette, command surface, automation orchestration.

SQLite worker + GraphPage

AI report portal and admin surfaces

HTML preview modal with theme, loading, and i18n

Background animation engine, activation service, email-link auth (Clerk)

Six additional locales; AdminEditor XSS hardening

flow.html: voice toggle and mini mode as default

Graph search surface on graph.html

Create scheduled workflows from graph search

Suggested question chips and sidebar tab layout tweaks

Flow view performance and activation UX polish

SQLite + knowledge graph integration

contentType field on extension and desktop paths

Context compression rules: removed 3-bullet cap

Prompt catalog expanded (+6) and existing prompts polished

Command hover cards rebuilt with portal positioning and motion

Mobile reading UX polish; desktop feature parity sync

Version archive (Edge / store sync node)

Long-context compression and truncation improvements

Clerk account registration

reader.html shortcuts and theme switching polish

Desktop client prototype (cross-window IPC, tabs API polyfill)

Fixed reader TOC scroll-spy flicker

Prompts split into Markdown source files

Fully unattended scheduled-workflow loop

Activation permission hardening; report retention default 180 days

Dutch locale; Gemini Web truncation fix

Sidebar/chat performance and faster reader shortcuts

Context compression levels 3/4 for key-point reduction

Sequential Chain and Split-Merge agentic workflows

Fixed date-range filter after feed selection

Global unread count consistency

Fixed discovery index refs and chat overlay positioning

Reader next/prev follows selected feed; faster loads

Scroll-up keyboard shortcut

Starfield theme with glow and sky backgrounds

Direct publish toggle for reports (skip review)

Gemini Web support and boundary bug fixes

Chat filter panel polish; suggestions surface discovery topics

MCP tools: command permissions and 3-turn multi-round chat

Feishu chat endpoint (IM dialogue)

Fixed render stalls from large chains / fast streams and maxToken overflows

Chat mermaid/color polish; automation composite index

Fixed custom command creation bug

Standard MCP bridge for external AI tools to read local reading data

One-click import from SnagFlow / CatchFlow generators

WorkflowService shipped with encoded prompt precompilation

More robust LLM call pipeline

Zustand data-loading rewrite across the sidebar pipeline

Standalone full-text reader (reader.html)

Shortcuts, TOC, light/dark themes, AI summary rail, typewriter effect, source links, adaptive layout.

Sidebar moved to React + Tailwind; mini density mode

Card images with hover zoom; empty states for four views

Flow performance: composite indexes, incremental summaries, date-range queries

LLM stack unified on Vercel AI SDK; AI settings rebuilt

Mermaid fullscreen polish; TypeScript upgrade and locale completion

Fixed per-feed auto AI summary toggle

Workflows use AI summaries as context; manual retry after failure

24h discovery auto-run with browser notifications

Dual AI profiles: default + complex model

Feishu notifications with per-feature toggles

Stable MCP bridge: multi-bot worker, MCP↔Telegram

Local model support; memory-leak and crash fixes

Submitted v0.2.2 deliverables to the extension store

Manifest bump only; no extra feature delta.

Report generation adjusted for store-policy compliance

Telegram push formatting; browser notification toggle

Worker report archive plus style/font iteration

Scheduled workflows running in the background

Chat rewrite: React + Tailwind, new-tab mode, filters, date grouping

Mermaid rendering hardened; smaller bundle

Options page migrated to React + Tailwind

Gap work folded in: discovery, streaming TTS, Telegram, cross-extension push, RSA activation, OPML, flow.html

Code cleanup (dead code, logs, whitespace)

End of the 0.1.x line

Prompt tuning; hide prompt settings entry

Fixed native Gemini API calls

Recommended feeds for more locales

HTML card generation improvements

Memory cleanup fix; HTTP image display errors

Help center added; prompt tweaks

Prompts moved to TXT for broader adaptability

Fixed broken context structure

Style fixes after 0.1.6 effect removal

Removed chat visual effects for performance

Log cleanup and font updates

Fixed missing Portuguese strings and card-generation AI errors

Fixed date-query bugs

Chat mode fully separated and rebuilt

AiUtils, XML-tagged context, multi-turn messages, session history.

Light theme overhaul

Dropped matrix/scanline effects; glass and card styling.

HTML preview via sandboxed offline iframe snapshots with cache

Keyword filtering with tag filters and reset hints

Multiple card styles, image export; commands expanded to 15

Trial length set to 5 days; feed management UX polish

Redeem dialog i18n and Gemini call errors

Fixed scroll jump when marking read in full-text mode

Trial period introduced (7→10→5 day iterations)

Conversational chat mode

Session isolation, context rewrite, streaming fixes, delete chat, chat prompts, model params.

Background auto-summary (~2s concurrency gap)

Command shortcuts (prototype → 11 commands)

DeepSeek and OpenRouter model support

Immersive reading polish; TTS panel and default rate

Dedup fix when 100-item cap broke per-feed uniqueness

Single-summary display polish; font and icon fixes

Error hiding and dev/prod log separation

Faster full-text mode loading

Date-range queries and search field adjustments

Summary modal and immersive reader display fixes

Podcast regenerate trigger logic

Font load, post-summary jitter, and repeat hotkey audio fixes

Audio download; streaming AI output

Unread count fix; TTS panel download strings

Locale switching (Japanese no longer forced by browser locale)

RSSFlow extension foundation

RSS subscribe/read, AI summarization, and early TTS playback.