Changelog

v1.1.6

2026-10-15

Changed

• Deleting a folder now deletes its subfolders too and moves the documents inside to trash. Restored documents return to the deleted folder's parent. The confirmation states how many documents are affected.
• AI organization now removes folders it leaves empty, along with ancestors that become empty, and restores them on undo. Unrelated empty folders are kept.

Fixed

• After applying or undoing AI organization, the dialog shows the result with Done and New run instead of returning to "Generate preview".
• On desktop, deleting documents with changes not yet synced to the cloud first syncs them; if that is not possible (offline or sync failing), it explains what would be lost and asks for confirmation instead of silently dropping new documents or edits.
• Desktop sync no longer gets stuck when a deleted document was edited on another device; the latest version moves to trash.

v1.1.5

2026-10-14

Added

• AI organization opens with a document tree to pick exactly which documents to organize. Folders select everything inside them; unorganized documents and documents in automatic date or activity folders are picked by default, including previously organized ones.

Fixed

• AI organization no longer keeps automatic date or activity folders. They are never offered as destinations: new topics the model nests inside them are created outside instead, and suggestions that only keep a document in such a folder need manual review. Emptied folders are removed on apply and restored on undo.

v1.1.4

2026-10-14

Changed

• AI organization previews now show the cleanup of empty automatic date folders and explain removal on apply and restoration on undo.

Fixed

• AI organization previews and application now remove empty automatic date folders, with the original structure restored on undo. Manual or repurposed topic folders and folders containing excluded or trashed articles are preserved. Existing previews gain cleanup without regeneration.

v1.1.3

2026-10-14

Added

• Desktop release announcements now offer an Upgrade now button that opens the update flow directly. The button only appears for a newer release on supported desktop clients; pending clicks are retained while the updater loads, and manual checks join an ongoing background check.

Fixed

• AI organization now distinguishes automatic date/activity folders from topic classifications, so earlier automatic filing no longer serves as a confirmed example that encourages retaining the old destination.

v1.1.2

2026-10-14

Changed

• AI organization now previews the resulting folder tree with search, collapsible folders, whole-row dragging between folders, sibling reordering and new subfolders. Up/down buttons also support keyboard and touch interaction.
• Expand articles in the tree to edit tags, choose destinations and inspect suggestions. Collapsible tag management supports bulk additions, removals, renaming and merging for checked articles. Changes are saved when the preview is applied.

Fixed

• Sidebar reordering and ordinal renumbering no longer block undoing folders and tags. Position restoration compares relative order and preserves subsequent manual changes.
• Applying a preview checks only the relative order of unchecked anchors and reports a conflict when those anchors change. Order-only changes no longer update article activity timestamps.
• Preview edits use functional state updates, and reordering buttons retain keyboard focus.

v1.1.1

2026-10-14

Changed

• AI organization now allows at most two requests per batch and increases the output allowance after truncation. Timeouts, three consecutive batches without valid suggestions, or a failure rate of at least half after six batches preserve a partial preview; unfinished items stay unchecked, and only valid suggestions are billed.
• AI organization is now an option alongside the existing strategies in the Organize documents menu, restoring the single sidebar entry.

Fixed

• AI organization previews distinguish rejected requests, rate limits, and provider unavailability, with troubleshooting guidance. Only status codes are displayed, never raw provider error contents.
• AI organization preserves other results when a batch is rejected by content or request limits (400/413/422), without charging platform credits for that batch. New retries for empty OpenAI responses are limited to organization, preserving other AI features’ request counts.
• Document organization now accepts a single complete JSON result wrapped in explanatory text, including citations and Markdown links, retries malformed batches and preserves other batches when a retry fails. Affected documents keep their original folders and tags and remain unchecked for review; failed attempts do not consume platform credits. Deletion checks also run before the provider's internal retries.

v1.1.0

2026-10-14

Added

• One-click content organization proposes folders and tags in an editable result preview, skips previously organized documents by default, supports explicit reorganization, and preserves article content and covers. Undo skips later changes and lists affected titles; batches with no valid suggestions cost no platform credits. Desktop clients sync before analysis and cache tags for offline display.

Changed

• AI title suggestions now draw on the article body, compare distinct angles with the original and include source-checked quotes plus concrete tradeoffs. High scores no longer suppress alternatives; long articles use bounded, distributed samples and suggestions stay in the existing review panel.

Fixed

• Invalid or duplicate AI title candidates are now discarded individually, preserving valid alternatives and allowing body and structure reviews to complete even when all title candidates are rejected.
• Applying AI optimization suggestions now preserves the article's current cover image, mode, ratio and prompt, including individual suggestions, Apply all and repeated requests, so subsequent edits no longer save a cleared cover.
• Skills/Agent Hub discovery now has five daily execution windows to retry missed or interrupted runs. Actual imports continue past the first 100 candidates toward the daily limit of 20, resume from saved results and retain existing safety and license checks; previews remain limited to 100 candidates.

v1.0.0

2026-10-12

Added

• Skills/Agent Hub adds public Clone, member-only private Fork, Stars and GitHub imports with original attribution and licenses. System repositories use private R2 storage, refresh daily and discover up to 20 new popular Skills repositories per day. Updates check repository identity, licenses and sensitive content; failures preserve the previous snapshot and support retries.
• Development/AI configuration sync now includes DeepSeek Harness under Coding agents, scanning default .dsh settings and profiles; credentials restore with owner-only access on macOS/Linux, while skills and AGENTS.md belong to the separate repository.
• Skills/Agent and configuration sync menus now offer Copy for AI Agent with the MCP endpoint, read-only credentials and local merge instructions; configuration handoff uses a temporary credential scoped to one snapshot revision without sharing the migration password.
• Local restore for Skills/Agent and development configurations maps known configuration locations across Windows, macOS and Linux, previews destinations and rejects path conflicts before writing; existing backups work without re-uploading.
• Skills/Agent repository MCP now includes history browsing, revision restoration and storage queries, supports uploads up to 5 MiB and bounded chunked file reads, and uses a separate connection configuration alongside document MCP.
• My Space now separates Skills/Agent repositories from encrypted development/AI configuration snapshots, with one-click local uploads, file selection and previews, sensitive-data filtering, and confirmation before restoring local files.
• My Space settings now let you allocate extra Agent repository capacity directly from personal storage, apply changes immediately, and reclaim unused capacity. Account storage details show the reserved allocation, while existing files and history remain protected.
• Publishing now supports per-account switches for WeChat, Zhihu, and X, plus custom HTTPS API platforms with a documented JSON contract; the retired Juejin destination is no longer offered.
• Documents can save default, article-image, or AI covers with custom ratios and reuse them when syncing to X, Feishu, and WeChat. Switching documents restores the previous reading position. AI generation also accepts a previewable, replaceable reference image without saving it to the article.
• Added Shift/Ctrl multi-selection, keyboard navigation, and bulk move/delete to the document tree. Added editor tab actions for closing related tabs and creating a new document.

Changed

• AI optimization improves the existing title assessment, paragraph edits and structure analysis with reader fit, reasons to click, opening continuity and promise delivery. Whole-article findings guide concrete edits for weak relevance, missing logical connections and empty phrasing. Paragraph editing shares chunk calls with proofreading, preserves author voice and labels title scores as editorial judgments.
• Unlocked configuration revisions can be synced again on the same page without re-entering the password; refreshing or changing the cloud revision requires unlocking again. The public repository directory is now named Skills/Agent Hub. Skills/Agent Hub adds topic navigation, system/community filters, full-directory search and 12 repositories per page. Compact cards emphasize purpose and authors, with horizontal topic navigation on mobile and a README-first detail page. Markdown previews use GitHub-style headings, tables, task lists and highlighted code, safely render embedded HTML, and resolve snapshot-relative links and images. A source view remains available. Hub filters survive detail-page navigation and reloads; withdrawing and republishing keeps the selected repository category. Hub search handles malformed URL parameters safely, and failed refreshes reload the current list instead of fetching the next page. My Space places the Skills/Agent Hub button beside Settings; the New repository menu offers Create directly and Import from GitHub, including for an empty repository list.
• Enabling Skills/Agent repositories no longer requires a token. Repository tokens live in My Space settings, separately from document MCP tokens; Copy for AI Agent guides you there only when a usable read/write token is missing. Skills/Agent file contents, retained history, public snapshots and GitHub imports now use private Cloudflare R2 storage; existing content migrates after upload and integrity verification. PostgreSQL retains metadata and permissions, and Forks retain independent references. Move curated GitHub Skills repositories into an independently owned system catalog, excluding them from administrators’ personal quotas and slots while preserving original attribution and public URLs. Admins can list, withdraw and republish system entries.
• Desktop sync checks cloud updates at startup, on reconnection and window focus, and every 30 seconds while visible. Only local changes trigger uploads; unchanged or failed background checks stay quiet, and hidden windows do not poll.
• Updating an older development/AI configuration snapshot removes Agent files such as CLAUDE.md and .claude/skills/* from that snapshot. Export those files or upload them to an Agent repository before updating; local files remain unchanged.
• WeChat draft sync reuses uploaded article images by account and image content, reducing repeat-sync and retry delays without changing image quality or covers. Existing clients need no update.

Fixed

• Fixed local configuration restore for HOME aliases, Windows path casing and external configuration roots, preserved valid Unix filenames and prevented case-folded remapping; all configuration writes start private and paths are checked before creating directories.
• AI sync no longer copies existing long-lived repository tokens; configuration grants can be revoked and follow account/desktop session invalidation, unavailable clipboards offer manual copying, and large envelopes require whole downloads.
• Fixed DeepSeek Harness sync to exclude its actual storages runtime directory and record Unix 0600 credential permissions in downloaded and encrypted ZIPs, so extraction preserves the permissions Harness requires.
• Hardened local sync: detect multiline cookie credentials, reject Unicode-equivalent destination paths, retry clipboard writes without issuing new credentials, refresh metadata after unlocking, and bound repository transfer batches while preserving special ZIP entry names.
• Fixed shifting buttons and unclear upload status during Skills/Agent sync, and false sensitive-content warnings for cookie-clearing examples in Cloudflare skills.
• Browser history restore now asks for explicit confirmation before restoring sensitive content and displays failures, including revision conflicts. MCP file updates also retain repository and revision audit details. OpenClaw setup now saves environment references before probing the connection.
• Repository restores now reject sensitive historical content for Agent tokens, complete MCP replacements require explicit replaceAll, oversized requests return 413, and Claude Code/OpenClaw configurations preserve environment references instead of storing tokens.
• When token creation reaches the limit, explain that document MCP and Agent repository tokens share 20 active slots and link to the other token settings page.
• Desktop sync now preserves document moves made during a cloud refresh, retries failed startup syncs on reconnection, and honors manual sync requests received while an automatic sync is running. Desktop sign-in now recovers interrupted refresh attempts, prevents concurrent profile updates and delayed requests from replacing credentials, and offers retry guidance when the system credential store is unavailable. Signed-out users can still access login and local mode while offline or during server outages. Old access tokens now expire within 60 seconds of rotation; account mismatches offer reauthorization while preserving local documents.
• Custom platform sync now omits the optional cover field when no cover is set, avoiding empty-URL validation failures. API sync sends the title separately without adding it to Markdown or HTML bodies, preventing duplicate titles downstream.
• Covers selected in WeChat draft settings now persist with the article. Failed saves reuse uploaded images on retry, and desktop cleanup preserves images still referenced by covers, article bodies, or conflict snapshots after sync.
• Fixed WeChat draft timeouts for image-heavy articles with bounded parallel uploads, gateway heartbeats, and relay connections that expire only when idle. Existing clients remain compatible.
• Fixed missing titles after cover load failures, stale default-cover titles, duplicate X cover images in article bodies, stale WeChat draft covers, and Feishu cover failures blocking body sync. Cover metadata now has size and storage quota limits.
• Batched tree mutations now share document revision checks and preserve image reclamation during permanent cleanup.