Linkly AI: baixar versions anteriors

☁️ Cloud embedding is here: Pro users can switch to the cloud💡 Tips: Cloud embedding is a Pro feature — sign in to your Linkly account first and make sure you have an active Pro subscription.🚀 New Features:
- Cloud embedding released: Pro users can switch to the cloud embedding service — faster, better semantic search quality, and lighter on your machine.
- Notes: added timeline navigation.
- Notes: added a tag filter rail.
⚡ Improvements:
- Scanned files are less likely to be OCR’d twice: a scanned PDF that already carries a hidden text layer is now accepted as-is once it passes the check, instead of queuing up for another full recognition pass.
- Magazine-style pages (“full-page image + visible text”) are no longer mistaken for scans.
- Fixed occasional cases where indexing stalled for a long time on lower-spec machines.
- The high-performance indexing mode now makes better use of the machine’s resources.
- Cloud libraries: documents of a million words or more no longer fail to index, and they no longer drag down the rest of their batch.
- Cloud libraries: indexing is now considerably faster.
🐛 Bug Fixes:
- Fixed custom models (OpenAI-compatible, not reporting a context window) being unable to compact a conversation manually.
- Fixed the UI freezing for several seconds when a note carried an out-of-range date (a mistyped year, for example), and the timeline dropping the most recent day in time zones where daylight saving skips midnight (Chile and others).
- Unified how #tags in the note body are parsed, so tags are now recognised identically across the backend, the frontend and every editing path.

📝 You can now take notes in Linkly AI — minimal card notes, made for agents💡 Tips: Notes are plain Markdown files kept in your local library folder. Nothing is uploaded to the cloud, and you can open them in any other editor at any time. Once written, a note is indexed automatically, so it can be searched and cited by AI.🚀 New Features:
- Added local Notes: a new note entry in the Chatbot sidebar (⌘/Ctrl + Shift + N takes you straight there), with a card-based timeline, tag filtering, sorting by time, ⌘F full-text search, one-click copy, and expand to full text.
- AI can read and write your notes: two new MCP tools, note_save and list, let Claude, Cursor, and other AI assistants turn a conversation into a note, append to an existing one, or list notes by tag and pick what to read.
- The Chatbot now supports automatic context compaction: when a long conversation approaches the model’s context limit, history is compacted automatically, so you no longer have to start over and restate the background; you can also type /compact to compact at any time.
⚡ Improvements:
- Image references in browser-extension clips are now recognized correctly: previously, a clipped article whose title contained spaces lost its image mapping for the entire document. Fixed for both local and cloud libraries.
- Markdown preview now renders embedded HTML, so tables recovered by OCR display properly.
- The search index readiness shown in Settings is now accurate — it no longer reports the index as usable before it has been built.
- Pushing to a cloud library now retries automatically when the network hiccups, instead of failing the whole batch and re-uploading from scratch.
- The website has a new home page, plus dedicated browser extension and CLI pages.
🐛 Bug Fixes:
- Fixed temporarily unreadable files (cloud drive not yet synced, external disk not mounted) being retried too frequently and slowing indexing down.
- Fixed the content fingerprint being cleared during a media file refresh, which could lead to a cloud copy being deleted by mistake.
- Fixed libraries with no folders taking part in auto-sync and failing over and over.
- Fixed the embedding progress bar going backwards when documents were added during indexing.
- Fixed file watching not waking the indexing pipeline in some cases, leaving changes unapplied for a long time.
- Fixed duplicate update checks firing within a short window.
- Fixed usage statistics still being sent after being turned off, and data being dropped when a send failed.

🚀 Text inside scanned PDFs and images is now searchable — and readable by AI💡 Tips: Linkly AI has upgraded its text-recognition (OCR) engine. Scanned PDFs and images are now recognized page by page and restored into Markdown. Scanned PDFs already in your library will be re-parsed automatically.🚀 New Features:
- Added text recognition (OCR) for scanned PDFs and images: powered by the new PP-OCRv6 engine, with substantially improved accuracy for Chinese / Japanese / English. Scanned files are restored into structured content with headings and tables, and can be exported as Markdown.
- The built-in Chatbot can now reference cloud libraries: mention a linked cloud library with @ and search it just like a local one.
- Added library titles: a library can have a real title and description, shown consistently across Settings, the Launcher, and the Chatbot.
- The Launcher now supports sorting by time: on top of the existing filters, results can be sorted by time.
- A desktop shortcut is now created automatically once onboarding completes (and removed on uninstall on Windows).
- Added index size compression: after upgrading, the index is compressed automatically to reduce disk usage — shrinking the index by up to 80%.
⚡ Improvements:
- Optimized how heavy compute tasks (OCR, speech transcription, embedding) use resources, avoiding the lag and out-of-memory issues caused by running them in parallel.
- Improved image and PDF extraction for more stable parsing.
- “Rebuild index” is now only allowed while background long-running tasks are idle, to avoid interference.
🐛 Bug Fixes:
- Fixed issues where certain models (such as Kimi) could produce repeated answers, lose the previous turn’s content, or call tools twice in the Chatbot; when a conversation is interrupted, the reason is now shown clearly.
- Fixed several cases where scanned PDFs and images were not recognized, or were recognized incompletely.
- Improved the stability of index upgrades and database migrations, fixing a case where fields required by new features could be missing on certain upgrade paths.

🚀 Your dusty local recordings and videos — now fully usable by AI💡 Tips: Linkly AI now supports local speech transcription. Spoken content in audio and video files can be transcribed to text and indexed, so it becomes searchable and referenceable by the Chatbot just like any document. Everything runs on-device — nothing is uploaded. Enable audio / video parsing under Settings → Indexing to get started.🚀 New Features:
- Added local speech transcription: a built-in lightweight transcription model converts speech in audio/video into indexed text, supporting Mandarin / Cantonese / English / Japanese / Korean.
- Audio formats: mp3, wav, m4a, flac, aac, ogg. Video formats: mp4, mov, mkv, webm.
- Audio parsing and video parsing can be toggled independently, since transcription is resource-intensive.
- Added a standalone Appearance settings page: adjust theme, font size, and macOS Dock icon visibility.
- Added auto-sync for cloud libraries: once enabled, new content is checked and pushed to the cloud automatically every 5 minutes.
- The Linkly AI Space location can now be changed: move the default space directory to a custom location (the former “Linkly AI Folder” is renamed to “Linkly AI Space”).
- Added drag-and-drop folders: drop a folder directly onto the watch list or a library’s detail page to add it.
- Chatbot added download replies: save an entire AI reply, or just a table within it, to your downloads folder.
- Chatbot sidebar now shows an indexing status indicator at the bottom, so you always know indexing progress.
⚡ Improvements:
- Chatbot conversation titles are now generated as soon as a message is received, instead of later.
- Launcher filters now include “Audio” and “Video” document types.
- Onboarding now uses a brand-new sample library.
- Destructive confirmation dialogs now use a consistent, more prominent warning style to reduce accidental actions.
- Website: added Alipay / WeChat Pay for users in mainland China.
- The library explore page now supports one-click starring, with improved popularity ranking.
- Improved website loading speed.
🐛 Bug Fixes:
- Fixed an issue where frequently changing files (e.g. in cloud-sync watched folders) repeatedly interrupted indexing, preventing large libraries from ever finishing.
- Fixed semantic search failing after the embedding model was unloaded due to idle sleep.
- Fixed a possible crash when installing updates on macOS: updates are now installed on next restart, and a downloaded update package is preserved even if installation fails, so it doesn’t need to be re-downloaded.
- Fixed Free users being incorrectly blocked from querying cloud libraries as if it were a Pro-only feature.
- Fixed indexing and cloud push silently failing for some users due to missing database fields.
- Fixed a potential 404 / 308 redirect storm on library share links.

🚀 Linkly AI Pro is here!💡 Tips: After 4 months of iteration, Linkly AI has launched the Pro paid plan, bringing more cloud capabilities and more curated models.🚀 New Features:
- Added official model quota: view more detailed model quota, remaining usage, and more in Chatbot and Web Dashboard.
- Added Free / Pro model access: Free users can continue using base models, while Pro users will gradually unlock curated frontier models such as DeepSeek V4 Pro and Qwen3.7 Plus.
- Cloud libraries now use plan-based quotas: Pro supports more cloud libraries, higher file limits, and more usage calls.
- The desktop receiving flow for browser clips is complete; the browser extension can now save web content directly into the Linkly AI folder for indexing.
- Chatbot sidebar now includes Settings, website, and quota entries, making common actions easier to find.
- Added one-click generation for cloud library introductions.
⚡ Improvements:
- Pricing page and Dashboard now explain Pro benefits in more detail, including models, cloud libraries, MCP Tunnel, and other capabilities.
- Web Dashboard improves API Key creation, library filtering, starring, and connection status display.
- After accepting a cloud library invitation, you can connect it directly to your own MCP service.
- Improved indexing and settings-save stability, reducing repeated retries, state overwrites, and configuration sync issues.
🐛 Bug Fixes:
- Fixed cases where the tray menu could crash or show out-of-sync status.
- Fixed stability issues related to MCP configuration changes, port refresh, and local service state.
- Fixed cases where the indexer could stop unexpectedly or repeatedly retry during delete, retry, and concurrent scan scenarios.
- Fixed several edge cases in browser clip saving, deletion, and scanning, improving file safety and reliability.

🚀 Publish cloud libraries publicly, plus filtering and starring🚀 New Features:
- Cloud libraries can now be published with two visibility options, Public and Showcase, making them easier to share and showcase
- “My Libraries” filtering: filter by source (All / Connected / Mine / Shared / Starred) to quickly find the library you need
- Library starring: mark frequently used libraries as starred with one click for quick access
⚡ Improvements:
- After accepting a cloud library invitation, the page now guides you to link in one click right there, with no need to navigate to the library page
- Improved the reliability of the account sign-in flow, fixing an occasional issue where sign-in could not complete

🚀 Document preview, EPUB e-books, and model sleep arrive; a brand-new onboarding flow and account sign-in💡 Tips:
- Upgrading Linkly AI CLI to v0.5.1 is recommended (adds EPUB support)
- Updating linkly-ai-skills to the latest version is recommended
- MCP tool descriptions have been updated (added EPUB and cloud library ownership tags); your connected AI tools may need to reload the MCP server
🚀 New Features:
- Added account sign-in: sign in to your Linkly cloud account in one click via the browser (OAuth), with no more manually entering an API key
- Added EPUB e-book support: .epub files now join full-text and semantic search, and AI can search and read them via MCP/CLI; after upgrading, e-books in already-indexed folders are picked up automatically — no action needed
- Added document preview: preview document content directly in the Launcher and the Chatbot workspace, with a one-click source / rendered toggle for Markdown and HTML
- Brand-new onboarding: the first-launch onboarding screens have been rebuilt with a simpler flow, and a multilingual sample library is prepared automatically
- Added model idle sleep: the embedding model automatically unloads after a period of inactivity to free up memory and wakes on demand; toggleable in Settings
⚡ Improvements:
- Cloud libraries are now tagged with ownership in MCP list_libraries / explore ([yours] / [shared]), so AI can tell library sources apart more clearly
- The embedding model is prefetched when the Launcher is shown, speeding up the first search
- Chatbot error messages are now categorized by type (sign-in / permission / model), with a corrected fallback model selection; the available model list refreshes automatically after sign-in
🐛 Bug Fixes:
- Fixed file watching repeatedly firing delete events and flooding logs because of cloud-sync folders (e.g., temporary files from Synology or Dropbox)
- Fixed the MCP command-line integration sometimes failing to connect because the PATH environment variable was truncated
- Fixed the index data size being shown inaccurately on the Settings page
- Fixed a manual index retry in Settings not actually restarting

📊 PowerPoint support is here; Chatbot gains @ context and a workspace💡 Tips:
- Linkly AI CLI needs to be upgraded to v0.5.0 (older versions reject --type pptx)
- Updating linkly-ai-skills to the latest version is recommended
- MCP tool descriptions have been updated; your connected AI tools may need to reload the MCP server
🚀 New Features:
- Added PowerPoint support: .pptx files now join full-text and semantic search with a purpose-built outline index, and AI can search and read them via MCP/CLI; after upgrading, PPT files in already-indexed folders are picked up automatically — no action needed
- Chatbot gains @ context: type @ to target a library, folder, or file and scope the AI’s answer to exactly that content
- Chatbot gains a workspace: files referenced in a session are gathered automatically, with file preview and browsing of indexed library files
- Added a unified Cloud Services settings page to manage the credentials the client uses for LinklyAI cloud services
- Chatbot is now the default startup window
⚡ Improvements:
- MCP search’s path_glob now uses substring matching: @folders and full paths hit directly, making folder-scoped retrieval more accurate for Agents
- MCP find_paths now returns a ready-to-use path_glob, making folder search easier for Agents
- Long session titles are truncated, and the Launcher shortcut is shown in Chatbot
- The shortcut settings page now includes the newly added shortcuts
- Many Chatbot UI refinements (session list scrolling, source list styling, @ picker appearance, and more)
- App update downloads now retry automatically on failure, with a friendlier error message
- Added folder access loss detection (macOS): after an app update the system may revoke folder access, leaving new files unindexed; Linkly AI now detects this at startup and guides you to re-authorize — one confirmation per folder in a system dialog, no need to remove and re-add folders; inaccessible folders show a ⚠️ badge in Settings → Folders
🐛 Bug Fixes:
- Fixed PDF bookmark outlines possibly being lost after rebuilding the index
- Fixed the cloud library MCP tunnel occasionally reconnecting every two minutes
- Fixed folder permissions on macOS causing cloud library push failures
- Fixed out-of-memory errors when indexing large documents in cloud libraries
- Fixed Chatbot input being lost and errors going unreported when sending a message failed
- Fixed some file previews failing to open and empty files showing no placeholder
- Fixed model selection issues caused by chat data from older versions (auto-repaired on startup)
- Fixed layout issues such as the notification toast squeezing the sidebar and the divider occasionally getting stuck
- Fixed the “Check for Updates” button becoming unresponsive after a failed update until the app was restarted
- Fixed semantic indexing retrying endlessly and wasting resources on documents containing special characters