Changelog

1.4.2 · 2026-10-10

New Features

• Desktop control: Inspect the screen, click, and type on Windows; macOS uses a native Swift helper for the same capabilities. Supports launching applications by name, and window snapshots now capture overlay popups and modal layers. Fully supported in Entanglement mode. Linux is currently not supported and planned for future release.
• Model monitoring in Box: A new dedicated page that checks configured or authenticated providers only when opened, displaying status history via a 90-day color bar.
• Automatic model selection (experimental): Deconstructs and cross-checks tasks with a dedicated analysis page. Network and conversation pages now provide layer-by-layer visibility into the active model routing, with automatic route diagnostics for stalled requests.
• Direct account sign-in: Added direct sign-in for DeepSeek, MiniMax, and OpenRouter, as well as desktop client sign-in for Gemini.
• Image Studio: Generate images and videos via OpenRouter.
• New models: Added Claude Haiku 5.5 (with optional thinking toggle), Claude Opus 5.5, Mistral Large 4, Nano Banana 2.1, Fledge Alpha, Ling 3.1 Flash, Apodex 1.1 Mini, Aion 3.5 / Mini, Pareto 26.10 Preview, and Glyph Cluster. Ultrafast mode is now available for GPT-6.1 Sol and GPT-6 Astra, including on Codex. Step 5 Preview now supports video input.
• Entanglement mode improvements: Introduces strict verification and adaptive modes. You can now ask questions during active tasks with responses preserved in session state. The causal engine now enforces a "fail-first, pass-after-fix" reproduction gate.
• Conversations: One-click translation for reasoning steps. Web links now display site favicons. File links feature streamlined path hints with built-in image previews. Images can be zoomed with the scroll wheel and panned by dragging. Added the ability to branch from any message turn.
• Agent: View and terminate agent background tasks. Search recent conversations by title. Re-enable disabled Skills directly from the slash menu. Workspaces with multiple root folders now load Skills and rules across every folder.
• Terminal: The status bar now displays real-time command upload and download throughput across Windows, local proxy, and macOS environments.
• Other updates: Copilot displays brand icons for each model. Active plan to-do items can now be renamed or appended. Updated Superpowers skills to v6.4.2. Box usage tracking now records background requests. The model configuration page has been redesigned with a sidebar alongside adjustable list and detail panes.

Fixes

• Fixed an issue where copying messages caused off-screen responses to turn into a long string of "x"s.
• Resolved an issue where failed concurrent edits on the same file repeatedly spawned identical "Edit attempted" status lines.
• Stopping context compression midway now cleanly halts the process without persisting incomplete compressed artifacts.
• Stalled tool calls are now treated as failures and prompt for smaller, subdivided payloads instead of retrying identically.
• Edits are no longer falsely flagged as stale when the model has already read the corresponding range or loosely matched the current text.
• Fixed truncated responses where Google account streaming chunks split across multiple network reads dropped subsequent fragments.
• Pressing Enter repeatedly while a message is actively transmitting no longer queues duplicate sends.
• Prevented inactive Skill files from being injected into the system prompt when all Skills are toggled off.
• Fixed an issue where a hanging tool call in the built-in browser froze the entire conversation session.
• Corrected HTTP proxy requests emitting redundant :443 port suffixes in the Host header; loopback addresses now reliably bypass proxies.
• Resolved an issue where scrolling up in completed conversations unexpectedly snapped the view back to the bottom.
• Fixed layout collisions in the Linux title bar and Image Studio where padding was missing for left-aligned window controls.
• Fixed unconfirmed file edits being silently discarded when a tool executed a full file overwrite.
• Fixed a failure when opening legacy conversations on Windows, and resolved two outdated TypeScript type errors in the settings package.

Improvements

• Long conversations: Reduced lag, latency spikes, and frame stutter during extended sessions.
• Performance diagnostics: Enabling monitoring now surfaces hot-path execution times, in-memory message body counts, and whether disk storage was queried during initialization.
• Runtime: Upgraded Electron to version 43.7.9.
• Pricing & catalog: Model pricing has been synced with upstream provider rates. Hosted Qwen and Grok offerings now strictly adhere to official context window tiers. The Gemini (Google) model catalog is filtered to display only conversation models actually accessible by the account.
• Terminal: Removed redundant command echoes, added detection for themed shell prompts, and now reports the exit code upon command completion. Command cards remain compact until terminal output is generated.
• UI & design: Unified design styling between the model selector and the settings model overlay. The conversation input box and its bottom tray are now consolidated into a single card.

1.4.1 · 2026-10-01

New Features

• Memory pre-analysis (experimental) in Box: Relevant memories are selected before each reply. The memory analysis page shows streaming progress. The model slot is named MemoryPreflight.
• Export memory packs directly from Box.
• Desktop sign-in now confirms in the browser before returning to the editor.
• Performance monitoring toggle in Box: Performance monitoring can be switched on or off. Metrics group windows by remote server, and leftover extension hosts are cleaned up.
• Terminal buffers are cleaned up automatically.
• Quote earlier conversations into the current session, marked with a golden robot icon, and cleared together with the input tray.
• AI browser operations now run in hidden windows by default; page snapshots switch to reading the accessibility tree, configurable in Box feature options.
• Added support for AMD model provider, GPT-6.1 Sol, and Gemini 4 Argon.

Fixes

• Keep All / Reject All only applies to the current conversation. Committed renames and new files settle properly, allowing the action bar to clear.
• Tool steps hold at the loading row until results arrive, then transition to the completed state.
• Streaming code fences now close according to CommonMark specifications. Thinking effort sent to Gemini is restricted to tiers actually supported by Code Assist.
• Memory analysis no longer fails when encountering an empty {}. Search planning is more accurate and consumes fewer tokens.
• Script cards now only display the count of modified files (for example, "1 edited") instead of relisting every filename.
• Clear error details are provided when subscription validation fails. Subprocesses stopped early terminate immediately without lingering.
• Resolved layout jitter and twitching issues on the conversation page.
• Various minor bug fixes and stability improvements.

Improvements

• Diffs in Code Review are now mounted on demand, layout measurements are batched, and memory retrieval is noticeably faster.

1.4.0 · 2026-09-26

New Features

Models & Capabilities

• Preconfigured models: GLM-5.5 Flash, GLM-5.4, Kimi K4, DeepSeek V4.1 Pro, and Muse Spark 1.4.
• Image Studio: Added support for Hy Image 3.5 and Qwen Image 2.1.

Interaction & Design

• Design previews now use the editor's current theme and the project's own design variables instead of applying a fixed color palette. Undefined CSS variables now return guidance instead of failing the request. If the workspace contains DESIGN.md, its rules are included in the workflow prompt.
• Design proposal cards now default to a full-panel, one-at-a-time view. Switch between proposals with tabs or the arrow keys; side-by-side comparison is still available in one click.
• Confirmed proposals are saved as standalone HTML files under .breezell/design/. Tool results now point the specification, plan, and implementation back to the same file so the implementation stays aligned with the selected design.

Notifications

• The notification panel now uses a list on the left and details on the right.
• All important notifications stay pinned at the top instead of showing only the first one.
• The list and detail panes scroll independently.

Fixes

Conversation & Agent

• Creating a branch now copies the source conversation body first. Older branches that already lost their body can recover it by line number.
• Reloaded messages release their placeholder height after the body returns. Collapsed tool and reasoning blocks no longer keep the height they had while expanded. The branch button ignores repeated clicks until branching finishes.
• Long conversations now open as one continuous list: bodies that have not yet scrolled into view keep placeholders and load on demand. Removed the “archived messages” notice and the old tail-only window. Line height remains valid after remounting, and streaming syntax highlighting now updates incrementally instead of flashing the whole code block.
• Opening the currently selected conversation again no longer replays the switch transition, preventing the conversation from being shortened or losing content outside the tail. If a local Git snapshot does not cover files from that conversation, the pending diff count in history is preserved.
• Conversation bodies are released only by the view that is currently displaying them. If a conversation is already open in an editor tab, opening it again returns to that tab instead of pulling a second copy into the sidebar, preventing the two views from unmounting each other's on-screen content.
• While a window is minimized, Breezell disables background throttling and prevents app suspension for active runs. Normal behavior resumes when the run ends, the window reloads, or the window closes.
• If the selected model is deleted or hidden, the model chip automatically switches to another visible model. If every model is disabled, Breezell prompts you to enable one first.
• External file read/write permissions were temporarily defaulted off and that false value was persisted, so later default changes could not affect already-saved settings. Upgrading now re-enables both permissions and records a migration marker, while preserving choices users make afterward.
• The commit dialog is now mounted at the workbench layer. Branch operations started from editor tabs are no longer clipped by the sidebar.

Storage

• Deleting a conversation now immediately compacts the database so space that previously only moved to SQLite's freelist is returned to disk. Older databases run one full VACUUM, and logs record the measured amount of storage reclaimed.
• Scheduled full VACUUM operations now wait for the history-write queue and run only when no conversation is streaming or executing tools.
• Deleting a conversation also removes its raw transcript from breezellTranscripts. Multi-workspace cleanup keeps the same grace period used for disk files for conversations that are still active in other windows.

Improvements

• The uncommitted-path set in History is now computed once when changes occur and cached until SCM state, path restoration, or the diff range changes. With about 1,000 dirty files, a redraw that previously took roughly 25 ms is reduced; with about 10,000 files, the previous roughly 265 ms redraw cost is substantially reduced.
• Language functions in Settings are now cached per language instead of being recreated on every render, avoiding whole-page cache invalidation. The Box shell rerenders only when the language changes. Navigation search, collapsing groups, or collapsing the sidebar no longer redraws the content page. OAuth polling reuses the previous result when the state has not changed.

1.3.9 · 2026-09-22

New Features

• Preconfigured models: Claude Opus 5.5, GPT 6 Sol, GPT 6 Terra, GPT 6 Luna, Gemini 4 Pro, Qwen4 Max, Qwen4 Plus, Qwen4 Flash, and Qwen4 27B.
• China Telecom Xingchen: Xing4.0-29B.
• Xiaomi: MiMo-V2.6 Pro, Pro UltraSpeed, and Flash. Also MiMo-V2.5 and V2.5 Pro.
• Tongyi: Qwen3.8-2.4T-A95B, Qwen-Doc-Turbo, and Qwen-Long.
• Sign in to Gemini with a Google account (Antigravity Code Assist).
• Image generation now includes Hy Image 3.5 and Qwen Image 2.1.
• When Super Memory is used up, a notice card appears above the input box.
• Editor notifications can show images and links; documents open on click.
• Conversation export and import now have a progress panel and can be canceled mid-way.
• Grok shows your current subscription tier.
• In Box, broken compatible services and models get a badge and can be retested individually.

Fixes

• Canceling generation no longer wipes the reply that already streamed.
• Streaming output no longer jitters line height; tool cards stay clickable.
• Right-clicking the title bar no longer crashes the window.
• Abnormal values or looping data no longer leave the sidebar blank.
• Chat sidebar failed to load because of a CSS comment syntax error.
• Backslashes in Windows paths are preserved; if a file read fails, Breezell retries with forward slashes.
• Looking up a Git commit no longer walks the entire history.

Improvements

• Long conversations scroll more stably.
• Editor performance is faster.

1.3.8 · 2026-09-19

More models in one editor, cleaner switching mid-conversation, and a tighter Plan loop — same thread, same context.

New Features

• Providers & models: StepFun and AiMass are now first-class providers. Qwen 3.8 lands with the 27B and Omni-Flash variants. Zhipu GLM-5.3-FlashX is in the list.
• Doubao Seed 2.1: Pro 0915 and Evolving are available.
• Official protocol wiring: OpenCode, Zhipu, Kimi, and DeepSeek now follow each vendor’s official plan tiers and protocol formats (Chat / Responses / Anthropic).
• OpenRouter: New slugs appear in the model list. Free models show a FREE ribbon; retired models are stamped Unavailable.
• Model picker in chat: Switching models opens an in-panel picker with Models and Compatible tabs.
• Models page: Jump back to top, or jump to the latest additions.

Fixes

• Paste image: Images are sent to the model as pixels. Breezell no longer looks for image.png in the repo.
• Composer: Line breaks are preserved. Shift+Enter is hinted in the input.
• Plan / Todo: Task status in .plan.md stays in sync as work moves. Resuming an old plan pulls todos with it. update_todo_list no longer re-marks items while a plan is executing.
• Codebase Exploring: Fold groups rebuilt.
• Edit cards: Completed cards collapse by default — controlled from the box feature toggle.
• Citation pills: Styling matches the editor selection helper.
• Account pool: Rotation is per request, not pinned to a single conversation.
• Built-in browser: Tabs are isolated per workbench window.
• Explorer: Project folders named dev or run are no longer hidden.
• PDF preview: Scroll, zoom, and drag work.
• Video preview: Duplicate selectors removed. Send to chat is available.
• Minimize hang: The unresponsive dialog waits 15 seconds before appearing, so fewer false alarms.
• Context windows: Qwen 3.8-27B can switch 256K / 1M. Short-context windows and list prices follow the official tables.
• Thinking intensity: Official tier names aligned across providers, including DeepSeek.
• Compatible slots: Shared config-column width. Capability tags sit under the token line.
• Admin: Model-management table header stays pinned. Fetching the model list is fixed. Reference price and long-cache fields line up.
• Sidebar: The overflow panel on the top activity bar no longer paints outside the sidebar.

Improvements

• Retired GPT-5.3-Codex-Spark removed from the catalog.
• Conversation body color matches tool-card filenames.
• Electron updated to 43.7.3.

1.3.7 · 2026-09-17

Improvements

• Dropped the on-device full-repo vector index. Search now uses grep.
• Background jobs and indexing no longer stall the conversation you are in.
• Usage rankings align with the cards and can switch by period. Sub-Agent usage is billed under the parent conversation.
• Conversation tags no longer interfere with each other. Cloning a chat no longer blows out the layout. The model stays attached to that conversation.

Fixes

• Deleted chat history is no longer restored automatically.
• Terminal cards near the input no longer yank the page down and bounce it back.
• Plan cards and the chat layout no longer push each other apart. Todo cards render correctly.
• Built-in SSH works again. It no longer goes through Microsoft Remote-SSH.
• The selection search assistant no longer covers highlights. Consecutive tool-call counts stay visible.
• Opening More on a conversation page no longer crashes.
• Clicking a branch no longer overflows the stack.
• On the conversation page, OpenAI-compatible and API providers make the current model obvious when you pick one.
• Plan mode is fixed.

1.3.6 · 2026-09-15

New Features

• Atlas local code search: Semantic search and index cards. Excluded directories are remembered. Questions about how code relates prefer Atlas.
• Plan mode blueprints: Click Start on the blueprint card to begin. Execution still uses todos.
• Selection assistant: Search selected text in the built-in browser, an external browser, or Bing. Queued messages can be reordered by dragging.
• Oversized paste or drop: Huge pastes no longer blow up the composer. They land as a file chip; the model reads them as a file.
• PDF attachments: Sakana, Mistral, Gateway, xAI, Kimi, Qwen-Doc, and similar channels can attach PDFs.
• Completion sounds: When a turn finishes, Breezell can ring or speak. Error speech is off by default. Pick a voice in Box.
• Open Browser in the editor. The tab-strip scrollbar is gone; the middle mouse wheel pans left and right.
• Usage: Cache and Token in separate columns. Cache reads and writes bill at official usage rates. Sub-Agents roll up under the parent session. The OpenAI-compatible account pool has its own page.
• Models and providers: Volcengine Ark can pull the model list with AK/SK, including Agent Plan (OpenAI / Anthropic) presets. Added AIHubMix and Sakana Fugu (Max / Ultra v2). Azure Foundry and Vertex Gemini are filled in. DeepSeek V4.1 Flash and Kimi K2.8 Preview join the roster.

Fixes

• Code Review re-diffs against each edit's own pre-edit snapshot, so sessions no longer mix.
• Dragging chat tabs no longer crashes the window. Terminal cards don't interfere with each other.
• The selection assistant no longer covers highlights. Consecutive send count is visible. Quote no longer looks like a full page. Reasoning stays visible on empty replies.
• The Box usage tab is no longer squashed. Rankings fill the width. Provider cards can switch time ranges.
• Official model lists and list prices come through correctly. Ordinary plan lists still render as todos. Blueprint nodes stay consistent in files, cards, and Start bubbles.
• Background Agent work and index writes no longer stall the conversation you are looking at.
• Vue tool errors are reported correctly.

Improvements

• Agent work style is tighter: retry on failure, use draft files, and follow along when state changes.
• Electron upgraded to 43.7.0.
• Box toggles are searchable. The dashboard keeps painting. Cleanup drops unused blueprints and terminal logs.

1.3.5 · 2026-09-09

Fixes

• Opening a new window after 1.3.4 could render a black frame and stay blank. The window now paints on first show.

1.3.4 · 2026-09-09

New Features

• Prompt rewrite: The wand in the dashed input tray rewrites the draft in the composer. Preview first, then replace. It never auto-sends.
• Feature usage accounting: Tokens for send message, follow-up suggestions, and prompt rewrite now land in Box → Feature usage, separate from conversation usage.
• Rollback confirm: Rolling back from a user bubble asks for confirmation in the input tray first, so you don't drop a branch by accident.
• Copy / branch from the turn bar: Copy is plain text. Branching from a turn opens a new chat; the editor conversation stays in the editor.
• Editor branch menu: Sidebar Open in Editor now opens a dedicated chat column instead of crowding the file tab strip.
• Image 2.5 Studio: Image workflows are upgraded; branches can be managed from the editor.
• Database auto-approve: Creating a database connection can go through Auto.
• Memory write tool: update_memory is the only path that writes user memory (workspace / global; add / replace / delete). Other notes on disk are no longer mutated ad hoc.
• Native tokenization: Local token counts follow each model's tokenizer.
• GitHub-style Review diffs: Code Review now uses git-diff-view.
• Chat backup zip: Export as a zip archive. Legacy JSON still imports.
• Command palette jump: Opening a file jumps to that file's latest hunk.
• Compatible API pricing: OpenAI-compatible rows can set a custom $/1M rate.
• Box network route: On the Models page: Direct, System proxy, Global, or Custom.
• Mode menu opens upward: The mode picker above the composer prefers opening up.

Fixes

• Changing the model on one chat no longer changes the global default or other conversations.
• Light-theme Codex usage: removed hard-to-read colored capsules (for example 5d). Usage is now plain text on the tray.
• Stop ends the turn immediately. Canceled edits are not written back to disk. Cards no longer vanish.
• edit_file flicker: clean files refresh by etag. Full-file rewrites shrink to the changed range. An open editor no longer reloads the whole page.
• Packaged MCP servers failed to start (Ajv constructor error). Remote PATH / cwd / env and error reporting are fixed as well.
• WSL: The install lock is released after the service is up, so later windows and Open Folder no longer time out.
• Grok API key: All traffic goes through the Responses API. Streaming no longer arrives in late chunks.
• Error cards now follow the provider that actually failed.
• Invalid URIs on the Review page no longer crash the whole page.
• Home-directory (~/) paths in replies are clickable.
• Vault cards: removed the hover lift that made cards "float."
• Light-theme Quote / Copy icons use the button foreground color and no longer wash out.
• Collapsed-rail avatars stay square.
• The usage filter dropdown anchors to the left of the button and no longer covers the Box sidebar.
• Segmented tabs wrap in narrow panels.
• Title-bar ellipsis regex: the full-width period is now a \u escape.

Improvements

• Composer scrollbar overlays content like the chat list. No extra gutter on the right.
• Search / exec timeouts: stuck rg processes now have a hard timeout so the window doesn't freeze.
• Context overflow: budget, uv hints, and empty landing cards match the current chat shell.
• History-bar long jumps teleport a slice instead of expanding the entire tail.
• Rollback confirm panel is wider; the confirm action shows ⏎.
• Review diffs tokenize visible lines only. Paging is faster.
• Box Models page: the title sticks to the top. Each provider row shows status and route.
• Context-ring bubble uses the same background as the input tray.
• Skills and Memory copy is localized. Delete controls use a neutral color.

1.3.3 · 2026-09-07

New Features

Box workbench

• The title bar now switches between Box and Agent in one click. Box is a standalone management surface that brings together models and providers, account and usage, Super Memory, Code Review, the MCP marketplace, the database browser, data cleanup, and Settings.
• Models: Provider config sits next to the model list. Add or remove OpenAI-compatible endpoints. Choose a network path per provider — including official Breezell models — Direct, System proxy, Global proxy, or a custom proxy. Probe and test connectivity in one click.
• Vault: Encrypts relay endpoints and account credentials, then fills them in where you need them. No more copying keys around.
• Code completion (FIM): Completions can use their own connection and each vendor’s native FIM protocol. Major vendors ship as presets.
• Data cleanup: Inspect local data stats, compact the database, and run integrity checks. Every maintenance pass is logged.
• Database browser: Rebuilt, with multiple connections and clearer grouped navigation.
• MCP marketplace: Organized by category, with a redesigned page.
• Privacy notes now use a card layout and cover more of the details.

Breezell Mind

• Conversations keep a running working notebook. When a long thread hits the budget line, older history is folded down to the notes and the essentials.
• Quote this conversation is lighter: quoting another thread only carries the latest turn plus the files involved. Full history is read on demand.
• Context compression now prepares and refreshes a background handoff document. The raw record stays on disk. You get a warning if compression has run too many times.

Conversation

• After a turn finishes, you get a few “what you can do next” suggestions — click to continue.
• A conversation can open as an editor tab, side by side with the code.
• @ mentions for files and symbols are as fast as Quick Open. / opens the Skills menu.
• Model menu: Breezell Max Plus toggle.
• Agent system prompt rewritten: clearer about what each turn is for. After real work lands, you get a short wrap-up. Comment language follows yours.
• File paths in messages render as clickable file chips and stay consistent across renders.

Models

• Added GPT-6 Astra, Claude Fable 5.1, Gemini 3.8 Flash, Muse Spark 1.3, and OpenCode Go Omen Alpha.
• Models without built-in web search now fall back to the provider’s native search.
• Claude Sonnet 5 pricing updated to the latest official rates.

Editor

• Edit Markdown in place and toggle preview without leaving the file.
• Diff view adds a Diff View submenu for comparison modes. Custom editors can show diffs.
• When opening a file, you can choose among editor types.
• Feature options in Settings redesigned. Git commit-message generation has its own model card.
• Account panel restyled as a membership card: plan badge, balance, expiry, and Super Memory quota at a glance. The account button shows your avatar.

Fixes

• An unclosed < in a stream no longer freezes the rest of the text. Reasoning-pane code renders as plain text; code fences without a language still highlight.
• Empty model replies trigger an automatic retry instead of silently ending the turn. Streams the upstream later marks empty are kept.
• Token counts after context compression now match what is actually sent.
• Changing the conversation model no longer overwrites the models chosen for Apply and Git commit messages.
• Reasoning auto-scroll no longer detaches on its own; streaming code blocks no longer flash white.
• Analyze sub-agents no longer spuriously ask you to switch to Agent mode.
• The built-in browser no longer types keystrokes into the AI input when incognito is turned off.
• Poolside Laguna requests that include a screenshot no longer fail the whole call; they now go through the vision proxy.
• Attachment token estimates before send are accurate. Emergency compression can fold the current turn.
• Error cards name the provider that actually failed.
• Terminal extension commands no longer wait out the full shell-integration timeout before they run.
• Plan-execution bubbles keep their corner radius. User-message shadows no longer cover the text below.
• Model hover settings align to the row they belong to.
• The code-index page in Settings no longer rescans the whole workspace on every open. Build artifacts are excluded from indexing.
• Linux: Restored the 8 KiB Node buffer pool to avoid regressions in some workloads.

Improvements

• repo_map no longer freezes the UI for tens of seconds. Line-level diffs for streaming edits and conversation persistence both leave the UI thread.
• Explore tools have timeouts — a dead network will not pin the Agent for five minutes. Failed explores collapse to one quiet gray “tried” line.
• Sidebar is lighter: MCP marketplace and Super Memory live in Box. History is paginated and grouped. Account badge placement tweaked.
• Runtime upgraded to Electron 43.6 (Chromium 150 / Node 24).
• Installers ship only the binaries for the current platform, so packages are smaller.

1.3.2 · 2026-09-01

Fixes

• Fixed an issue where the entire chat history could disappear after switching workspaces, changing WSL / remote paths, or losing local state.
• Fixed freezes and unresponsive windows in long conversations, as well as UI lockups during large-scale file changes or git checkout.
• Fixed local file links in chat appearing as blocked and becoming unclickable.
• Fixed built-in browser zoom resetting to 100% after a refresh, and model dropdowns or mode popovers being obscured by web content.
• Fixed queue actions such as Edit, Send Now, and Delete being unavailable on touch devices.
• Fixed shared browser deadlocks and new Agent drafts being mixed across conversations.
• Fixed conversation workspace identity changing unexpectedly, which could cause memory to leak across projects.
• Fixed model group collapse states not persisting across restarts.
• Fixed long conversations incorrectly reporting a “Network Error” after around 50 seconds when the actual issue was a full context window.
• Fixed switching the chat model overwriting separately configured models for Apply and Git commit messages.
• Fixed reasoning output getting stuck while scrolling, and white flashes in streaming code blocks.
• Fixed Code Index scanning build artifacts, which caused slower indexing and unnecessary disk usage. Opening Settings no longer repeatedly triggers “Scanning workspace.”
• Fixed Agent incorrectly insisting that code comments must be in English and refusing explicit requests to use Chinese.
• Fixed the web exploration tool idling for around five minutes without making progress.
• Fixed clipped corners on plan execution cards and pinned popover shadows overlapping the first line of replies.
• Fixed remote server version mismatches when connecting to remote development environments from source builds.

Improvements

• Long-conversation processing no longer blocks the interface, allowing replies to appear progressively as they are generated.
• When the context window is close to full, Breezell now explains the situation earlier. Older results are moved to local storage and reloaded only when needed.
• Improved responsiveness for code blocks and Diffs in long responses. Syntax highlighting for large files no longer locks up the window.
• Chat history is now grouped into Today, Yesterday, and Earlier, with Agent and Teams conversations separated.
• Provider headers in the model list now stay pinned while scrolling. The currently focused conversation no longer jitters with the input area.
• Code Index now respects .gitignore and focuses on source files instead of build artifacts.
• Inspecting repository structure no longer causes stalls lasting more than ten seconds.
• Sending messages in extremely long conversations is now more responsive.
• The model settings popover now aligns with the currently selected row.
• Git commit message generation can now use a separately selected model.
• GPT-5.6 reasoning levels, including Pro, are now visible and selectable.
• Added Hunyuan HY4, Kimi K3 / Video, Qwen3.8 Flash, GLM-5.3 Flash, and a range of new video models.
• Image Studio now supports Grok Imagine image generation 2.0 and video generation 1.5.
• After long-conversation compression, Breezell now keeps a handoff summary so the conversation can continue smoothly, while the complete local history remains searchable by keyword.

1.3.1 · 2026-08-26

Fixes

• Fixed an issue where chat history could disappear after the workspace identity changed.
• Fixed an issue in Breezell Settings where incorrectly mounted dropdown components could cause unexpected behavior.
• Fixed incorrect zoom behavior in the built-in browser.
• Fixed chat layout sizing so the conversation area is no longer compressed by the page height.
• Fixed an issue where edit tool rendering could cause a crash.
• Fixed system notifications incorrectly stretching to full width when using the new Breezell UI.
• Corrected the bottom input area's alignment so its left and right spacing now matches the surrounding chat panel.

Improvements

• Improved overall performance and responsiveness.
• Enhanced model engine capabilities and reliability.

1.3.0 · 2026-08-25

New Features

• Added an all-new interface layout, available under Breezell Settings > Appearance.
• Added SiliconFlow, OpenCode Go, and Meta as model providers, along with per-model API format selection for Chat Completions, Responses, and Anthropic Messages.
• OpenRouter now supports selectable API formats. Claude and OpenAI use their native APIs by default.
• Added support for Gemini 3.7 Flash and Dots Studio dots3-note Preview.
• Chat now supports video attachments, with video preview directly inside the workspace.
• Added Teams worktree isolation: teammates can edit code in independent Git worktrees, and idle teammates can be automatically spun back up when needed.
• MCP servers now run inside the workspace extension host. Closing the project window also terminates their child processes.
• Redesigned Advanced Model Settings, including default thinking settings and pass-through support for reasoning effort.
• Added a new input tray, consolidating quick commands and tool popovers into a cleaner interaction flow.
• Unsent drafts are now preserved in chat history, and teammates can now be assigned labels.
• Added two new themes: Breezell Warm and Breezell Paper. Theme palettes have also been expanded, and the launch screen now follows the editor's active color scheme.
• Added a folder picker in the All list as a fallback when folders cannot be dragged from the Explorer.
• Added an About window that displays the installed Breezell version.
• Code completion now takes recent edits and language-specific comments into account for more relevant suggestions.

Fixes

• When using ds4 or other text-only models through an OpenAI-Compatible provider, images can now be sent directly if vision support is configured on the provider. A separate vision proxy no longer needs to be configured in the IDE.
• Official text-only models such as DeepSeek still block image attachments when no vision model is configured, preventing invalid requests.
• Switching modes while a response is still streaming no longer immediately displays a "Switched to ..." notification. The notification now appears only after the current response finishes.
• Edit cards now jump to the correct location on the first click.
• Context usage now reflects the context actually used by the request, fixing inflated or inconsistent usage readings.
• The MCP server list now refreshes after the extension host restarts, and server names are now emphasized in the input tray.
• Fixed synchronization issues across Git, Command Palette, History, and Explorer, including cases where pending diffs disappeared after a commit.
• Fixed chat notifications not appearing while the Breezell window is minimized.
• Fixed WSL / Remote URI handling and REH cache validation.
• Fixed white flashes when the browser webview is recreated and cases where the native webview intercepted splitter dragging.
• GLM [1m] is no longer incorrectly routed through Chat Completions.
• Improved default theme behavior after installation and corrected red/green diff contrast.

Improvements

• Refined the chat interface, Mermaid rendering, theme controls, Settings rows, and terminal approval copy.
• Improved interactions across the input tray and tool popovers.
• Improved Plan mode switching behavior: Breezell will still ask questions when clarification is needed, and relevant cards will no longer be hidden.
• Added adjustable font sizing for Ask Question options.
• Improved memory recall reliability, with clearer messaging while recall is in progress.
• Code completion suggestions now better match the language of the current file.

1.2.9 · 2026-08-24

Fixes

• Fixed cases where some conversations could not be loaded correctly.
• Fixed incorrect output formatting in Entanglement mode.
• Improved stability and fixed cases where Breezell windows could crash or terminate unexpectedly.
• Fixed blank pages when opening websites in the built-in browser.
• Fixed trailing tags being exposed at the end of Analyze mode responses.
• Fixed custom context window settings for Codex models not being applied or displayed correctly.
• Fixed incorrect quota displays when using Grok direct login.
• Fixed automatic compression not being triggered after switching the context window.

1.2.8 · 2026-08-16

New Features

• Reworked Agent, Entanglement, and Teams modes for more reliable workflows.
• Added OpenCode Go and Meta integrations, along with per-model API protocol selection.
• Image Studio now supports Seedance 2.5. DeepSeek models can now hand off images for vision recognition.
• Added GLM-5.3 and Grok 4.6, and improved compatibility with DeepSeek-specific model formats.
• Initial setup now lets users choose their interface language during installation.
• Redesigned the empty conversation page with feature cards and left/right navigation.

Improvements

• Teams now provides more detailed task-progress reporting.
• Long conversations now automatically prune stale or invalid context to reduce context bloat and unnecessary overhead.
• Improved memory-retrieval accuracy, and waiting-state prompts now update dynamically as tasks progress.
• Terminal sessions are now grouped by purpose. Silent commands remain active until their execution budget is exhausted instead of stopping prematurely.
• Improved Git-status refresh and synchronization after review, rollback, and submission actions.
• Mode-switch separators now remain visible, making it easier to review where a mode changed during a conversation.

Fixes

• Fixed Teams conversations freezing or becoming stuck in an idle state, as well as diffs remaining visible after submission.
• Fixed chat notifications not being delivered when the Breezell window is minimized.
• Fixed normal conversations being interrupted after an image-recognition handoff completes, and browser emulation views occasionally becoming stuck.
• Fixed inaccurate terminal output, issues with chained PowerShell commands, and background commands repeatedly looping.
• Fixed the Source Control commit input disappearing or becoming invalid after refresh.
• Fixed queued messages not being sent when pressing Enter, as well as Analyze tasks stopping unexpectedly or repeatedly restarting.
• Fixed issues with Grok subscription-based conversations, hidden reasoning becoming stuck, and invalid paths causing exceptions.
• Fixed rollback operations unintentionally affecting unrelated changes, MCP tools resolving to the wrong path, and Linux packaging issues.
• Fixed various other bugs and stability issues.

1.2.7 · 2026-07-29

New Features

• Improved understanding and navigation efficiency for large codebases.
• Added optional answer-quality insights, presenting this turn’s evaluation across answer quality, requirement understanding, and verification confidence.
• Cache statistics details now support pagination, linked filtering, and multiple languages.

Fixes

Source Control and Agent Runtime

• Fixed the commit-message input becoming empty or invalid after refreshing a source-control repository, switching views, or rebuilding components.
• Fixed Agent reasoning stalls, incorrect terminal-state detection, lost background-command exit codes, and tasks ending prematurely.
• Fixed queue-message editing, tool-call feedback, task-list restoration, and continuous-turn state issues.

Chat and Remote Environments

• Fixed original conversations being truncated prematurely when editing and resending messages. Messages are now replaced atomically after validation completes.
• Fixed remote-path issues in WSL and Remote SSH environments, and improved host startup, password retries, and server caching.

Rendering, Models, and State

• Fixed KaTeX block-formula parsing, code-block themes, Chinese font rendering, and file-icon alignment.
• Fixed some models incorrectly displaying empty responses, exposing error stacks, or issuing abnormal provider network requests.
• Fixed deleted content reappearing in Super Memory and browser-panel states reverting unexpectedly.

Improvements

Agent Search and Execution

• Optimized Agent search, tool invocation, task execution, and termination logic to reduce redundant conclusions, unnecessary waiting, and abnormal interruptions.
• Optimized result limits, execution timeouts, and streaming status for large-scale code searches to reduce resource consumption from oversized result sets.

Chat Experience

• Improved chat completion status, the bottom toolbar, scroll positioning, and long-conversation loading.

Settings and Usage

• Refactored model-account and OAuth sign-in cards, unifying the visual hierarchy and interaction patterns across the Settings page.
• Improved Codex usage, model metadata, and related status indicators.

1.2.6 · 2026-07-27

Added

Session-Level Chat Git

• AI-generated code changes from the same session can now be submitted together.
• Added simultaneous detection and processing for multiple Git repositories.
• Chat tasks and code review status now synchronize automatically after submission.

Providers, Models, and Capabilities

• Added the Poolside Platform provider with access to Poolside Laguna through an OpenAI-compatible configuration.
• Added model configurations for Qwen3.8, Opus 5, and Pangu.
• Enhanced LongCat with inference replay, preserving reasoning context after tool calls.

Remote Workspace Security

• Added a security warning when accessing the Linux root directory / in a remote environment.
• Linux virtual file system paths such as /proc, /sys, and /dev are now excluded automatically.

Improvements

Code Review

• Redesigned the Accept/Reject action bar for individual code blocks.
• Updated Keep All to follow the current theme’s accent color.
• Reduced the height of inline action bars for a more compact code area.
• Review action bars now collapse and reset automatically after a successful Git commit.

Model Selector

• After a key is entered for the first time, only five recommended models are enabled by default to reduce list redundancy.
• OpenAI-compatible provider capabilities now adapt to the selected API format.

Code Submission and SCM Refresh

• Git status in the Activity Bar now refreshes automatically after files are saved.
• Chat Git submissions are now uniformly treated as Keep All Changes, automatically completing the corresponding code review steps.

Interface

• Removed the unread indicator from notification titles.
• Improved hover-menu switching for OpenAI-compatible model groups.
• Fixed the vertical alignment of tool-count text.
• Enabled the Kimi Mono icon style in the light theme.

Fixes

WSL and Remote Git

• Fixed a WSL extension host startup crash.
• Fixed inconsistent WSL server proxy configuration.
• Git status checks and AI-generated commit messages in WSL and SSH environments now run through a remote-aware execution channel.
• Prevented Git commands for remote workspaces from running against local paths.
• Fixed an issue where the Agent automatically closed Git repositories after modifying files.
• Fixed incorrect relative-path display in remote URIs.

Remote SSH

• Fixed Remote SSH password-expiration handling. Saved passwords are now cleared when rejected by the server, and users are prompted to re-enter the password during the same connection attempt.
• Prevented Remote SSH from downloading REH server packages with mismatched commit hashes, avoiding version mismatch errors and remote host startup failures.

Chat Rendering

• Added Trusted Types support for Mermaid charts.
• Improved Mermaid tag sanitization to prevent unsafe or malformed tags from causing rendering failures.

Rules and Memory

• Fixed empty content being saved after an editor lost focus.
• Fixed the resource manager not refreshing promptly after changes to the .breezell configuration.

Remote Tree View

• Stopped meaningless tree-view errors from being emitted repeatedly when SSH Hosts configuration is missing.

1.2.5 · 2026-07-22

Added

• Added provider and model integrations for Tencent Hunyuan HY3 and TokenHub.
• Added Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, both now officially generally available.
• Added project badges to chat histories in multi-root workspaces.
• Expanded capability support for compatible provider channels.
• Added interface translations for Japanese, Spanish, Portuguese, and additional languages.
• Updated the OpenAI protocol guidance with two clearer recommendations:
  • OpenAI-compatible endpoints use the OpenAI protocol by default.
  • The Responses API is recommended for GPT models.

Fixes

• Fixed an issue where newly created todos remained stuck at Waiting for Confirmation.
• Fixed multi-stage plans requiring users to manually click Continue after every batch.
• Fixed the duration, Review, Git, and tool action bars remaining highlighted and clickable after reverting to a checkpoint.
• Fixed entitlement gating in the drawing panel not synchronizing correctly, and improved unclear restriction messages.
• Fixed blank canvases not receiving a default node when opened for the first time.
• Fixed Chat mode sessions leaking or becoming mixed across multiple windows.
• Fixed model selection falling back to an unavailable model after the currently selected model was deleted.
• Fixed single-file staging not working correctly in Chat Git.
• Fixed commit and refresh issues in multi-repository workspaces.
• Fixed chat histories being lost after adding a folder to the workspace.
• Fixed missing Activity Bar icons in WSL and SSH environments.
• Fixed project list flickering in the Explorer.
• Fixed concurrent Git status checks in the repository view overloading the extension host.
• Fixed malformed **** output disrupting OpenAI reasoning streams.
• Fixed the styling of the close button on edit bubbles.

Improvements

• Simplified the startup cards on the Welcome page with dashed tiles and unified Code Review icons.
• Updated secondary Super Memory text to use italic styling.
• Consolidated provider cards on the Settings page.
• Improved the title layout of network configuration dialogs.
• Improved localization for the empty model selector and made it possible to open Settings directly from the selector.
• Ensured the background directory’s supportsFast capability is correctly passed through the request pipeline.
• Corrected pricing information for Qwen 3.7.

1.2.4 · 2026-07-20

New Features

Models and Providers

• Integrated Grok 4.5, including adjustable inference strength and a Fast channel; Claude Sonnet 5 is available with both 200K and 1M context options.
• Added Kimi K3 through both Moonshot API and Kimi Code channels, with 256K and 1M context options, plus full support for Kimi K2.7 Code.
• Added the StreamLake provider with Kuaishou's complete KAT-Coder model family, built-in Coding Plan presets, and pay-as-you-go endpoint presets.
• Added OAuth sign-in for Grok accounts.
• Redesigned Fast channels from a global toggle into independent provider + model configuration, so Copilot, Codex, Claude, and other providers use their official channel semantics.
• OpenAI-compatible configurations can now be pinned so frequently used channels remain at the top.
• Added support for time-limited model trials and discounts, with promotion badges in the model selector.
• Updated the model directory with current flagship models, including Claude Fable 5, Claude Sonnet 5, Nemotron 3 Ultra, DeepSeek V4, GLM-5.2, and MiniMax M3, while removing outdated entries.

Usage, Appearance, and Interface

• Upgraded the Usage Center with multilingual usage and reset-time display for Codex, Kimi, and Copilot in Simplified Chinese, Traditional Chinese, Korean, and Russian. Settings and chat-sidebar data now synchronize in real time, and Copilot usage shows Chat, Completions, and Premium Requests separately.
• Welcome-page titles can now be customized globally or per mode, including font size and font weight. Added the Elegant style for quick-access shortcuts.
• Added Japanese, Spanish, and Brazilian Portuguese interface languages.
• Redesigned the notification center to follow the product language, refreshed the Super Memory card layout, and softened the chat input and message bubble visuals.

Browser, Chat, and Workspace

• The built-in browser now supports muting all tabs or individual tabs, with real-time synchronization between its panel tab list and the browser.
• Chat messages can now include up to 10 images; images are automatically resized and included in context accounting. Added turn-duration display, one-click rollback within messages, and options to open links in the external or built-in browser.
• The chat commit dialog now supports multi-repository workspaces: generate commit information per repository, submit independently, initialize repositories directly from the dialog, and preserve commit/push progress across restarts.
• Added project grouping in the resource manager and more timely collaboration-status updates across multiple windows.
• Added persistent memory, which automatically distills conversation history into reusable experience with multilingual deduplication.

Diagnostics and Productivity

• Added cache visualization with per-turn cache notifications and workspace daily reports comparing today and yesterday, including session/workspace filters, estimated savings, and cache-miss analysis.
• Added interactive design cards that can generate HTML design drafts directly in chat and apply them after confirmation.
• Provider network settings now support one-click parallel speed tests across direct connection, system proxy, global proxy, and custom routes. Recommended routes are labeled automatically, and routing badges update in real time on the Settings page.
• Upgraded the PDF viewer core with a new toolbar. Attachments are metered by page, automatically hidden when context is constrained, and limited by each model's volume and page limits.

Improvements

• Redesigned the backend command experience: status-bar entries stay visible until commands truly finish, while concurrent commands collapse into an expandable summary.
• Expanded advanced-feature authorization and compatibility for mainstream extensions such as Python and Pylance, and improved remote-development compatibility.
• Improved installation and update reliability on domestic networks.
• Made automatic context compression smarter by triggering it dynamically according to available context, reducing unnecessary compression.
• Built-in skills are now enabled on demand and default to lightweight mode.
• Agents now complete multi-stage plans without pausing mid-process for confirmation, with more restrained responses focused on outcomes.
• Deeply optimized the request pipeline to reduce the cost of long conversations.
• Redesigned error prompts with quieter banners, detail drawers, and clearer mixed Chinese-English rendering.
• Restored the title-bar layout toggle, reorganized quick access on empty chat pages, and prevented provider icons from triggering system alerts.

Fixes

• Resolved lag and abnormal memory usage in long conversations, significantly accelerating historical-session opening and improving extended-response smoothness.
• Disabled or offline MCP servers no longer expose tools to models, preventing accidental calls; MCP connections also remain stable after toggling, and configuration read errors no longer clear saved settings.
• Fixed incorrect bolding of selected Settings navigation icons, crashes in certain Linux environments, and several other interface issues.
• Fixed past-session loading issues that could cause interface crashes or unbounded storage growth.
• Fixed false-positive “context compressed” states: real compression is now explicitly visible, and long-conversation understanding drift and cross-session memory interference have been addressed.
• Deleting an active model or channel configuration now automatically selects an available option instead of leaving selection boxes blank.
• Switching between Plan and Analysis tabs now preserves state without reloading content.
• Fixed occasional white screens in the drawing panel, stale entries after background command termination, Settings-page scrollbar issues, and other minor details.
• Improved parameter compatibility across models, including instruction editing, streaming interruption detection, thinking-budget limits, and document alignment.

1.2.1 · 2026-07-08

We are excited to announce the release of Breezell v1.2.1. This update brings a wide range of improvements across source control, background commands, terminal workflows, Skills, model support, image task handling, UI polish, and long-running AI interactions. It also includes many stability fixes for streaming output, editor state synchronization, session persistence, multi-window behavior, and authentication.

New

• The Source Control view now supports expandable commit details and browsing of related files, making change tracking clearer and more intuitive.
• Background command and terminal collaboration has been improved, with support for multi-task status display and clearer feedback when users manually interrupt a task.
• The Skills panel now works together with the Superpowers master toggle, creating a more unified configuration path.
• The conversation home page now includes quick access shortcuts for frequently used capabilities, reducing the time needed to reach common features.
• Brand visuals and interface identifiers have been updated across startup screens and Team Mode.
• Expanded support and parameter adaptation for the latest generation of mainstream AI models.
• Image settings have been upgraded, including status tracking for asynchronous video-style tasks.
• Added the first phase of batch script execution, designed for multi-step automation scenarios.

Fixes

• Fixed issues where AI conversation streaming could become stuck, produce abnormal formatting, or fail to recover smoothly after long-running tasks were interrupted.
• Fixed repeated application of code editing tools, abnormal navigation behavior, and editor state desynchronization after rejecting changes.
• Fixed delayed refresh issues in context usage statistics, progress rings, and hint messages.
• Fixed version update notifications getting stuck and several untranslated text strings.
• Fixed usability and display issues across Super Memory, code review, and related tool flows.
• Fixed session persistence issues and reduced lag in the history list when handling large threads or large pasted content.
• Fixed overlay panels such as MCP and Skills being blocked during conversations.
• Fixed occasional stream interruptions during long research-style outputs.
• Fixed task completion reminders not triggering when the window was out of focus.
• Fixed settings being overwritten across multiple windows and batch operation bars disappearing unexpectedly.
• Fixed authentication sessions requiring users to log in again or showing inconsistent states in certain scenarios.

Improvements

• Upgraded the core runtime and AI dependencies to improve overall stability and compatibility.
• Optimized prompt caching and tool-calling strategies to reduce repeated overhead and improve response consistency.
• Improved streaming rendering and UI update logic, making long conversations and high-load outputs smoother.
• Optimized the LLM request timeout mechanism so long generation tasks are less likely to be interrupted incorrectly while progress is still being made.

1.2.0 · 2026-07-06

Breezell v1.2.0 introduces new AI tools, smarter agent workflows, interface upgrades, stability fixes, and performance improvements across long conversations, tool execution, terminal output, memory isolation, and remote development.

AI Tools and Intelligent Capabilities

• Added the new Agent Research tool, with support for real-time progress tracking and result display.
• Added batch script execution in Code Mode as the first phase of multi-step automation support.
• Added a unified browser tool entry point, consolidating previously separate browser actions into a simpler workflow.

Conversation and Interface

• Added a Message Rail to the chat area, making it easier to browse and jump between conversation messages.
• Added quick access cards for Super Memory and Code Review on the chat home page.
• Added a quick toggle panel for MCP servers in the input area, allowing users to enable or disable MCP without opening Settings.
• Introduced a new brand identity, with unified visual updates across the launch page, welcome page, and Teams Mode.
• Redesigned the provider management interface in Settings, now supporting list-based browsing and detailed drill-down views.
• Added custom theme accent colors, available in both light and dark mode.
• Upgraded image settings to v2, with support for polling and tracking asynchronous video tasks.
• Added health check reports for OpenAI-compatible providers, making it easier to diagnose connection and configuration issues.

Stability

• Fixed lag when loading long conversation histories and slow responses when pasting large blocks of content.
• Fixed occasional UI freezes during AI streaming output, especially when the terminal produced large amounts of output.
• Fixed cases where the Keep All / Undo All action bar could disappear unexpectedly.
• Fixed settings being overwritten when multiple windows were open at the same time.
• Fixed issues where Super Memory could not be used properly in certain modes.
• Fixed incorrect timeout interruptions during AI generation, so slow but healthy generations are no longer stopped too early.
• Fixed refresh token failures that could cause users to log in repeatedly.
• Fixed tool execution results losing their correct pairing when users interrupted with a new message.
• Fixed silent compression of tool results causing the AI to rerun the same task repeatedly.
• Fixed inaccurate terminal exit code display and cases where terminal output could not be fully restored.
• Fixed Remote SSH connections losing hostname casing and port information.

Performance and Experience Improvements

• Improved overall response speed by streamlining the AI tool surface and reducing system overhead for each conversation turn.
• Optimized search and reading strategies with a “locate first, then read in windows” approach, reducing unnecessary full-document reads.
• Added workspace-level memory isolation, preventing memory recall from different projects from interfering with each other.

1.1.9 · 2026-06-26

New Models

• Added support for multiple new large language models, as well as AI image and video generation models, giving you more model options.

New Features

• Agents now support Deep Research, making complex questions easier to explore in greater depth.
• The chat sidebar now supports one-click jump to the latest message and one-click message copying.
• Service provider management now supports a list layout, and the MCP Market interface has been refreshed.

Experience Improvements

• Refined light/dark themes, terminal UI, and multiple interface details for a cleaner visual experience.

Performance & Stability

• Upgraded the underlying runtime engine for smoother startup and operation.
• Agents are now smarter and more stable, reducing repetitive and ineffective actions.
• Fixed issues related to login, terminal, remote connections, and multiple interface areas.

1.1.8 · Bugfix · 2026-06-09

• Added new validation for intensity adjustment
• Optimized repeated dialogues and automatic intelligent context
• Fixed scrolling issues on the conversation page
• Fixed unstable compression

1.1.7 · 2026-06-07

• Fixed the optimization model being overly verbose
• Fixed an abnormal issue with file tagging in the optimization review
• Fixed several bugs

1.1.6 · 2026-06-06

Added

• Code Review Panel: Standalone review view, line-number diff, large-file Myers diff, rollback/revert, select reference/copy, send to chat
• Appearance Settings: Activity Bar width/location, Side Bar position, editor tab style
• Breezell Browser: Redesigned panel UI, address/search bar, incognito mode, auto-close browser when the last tab is closed
• Database Panel: Workspace SQLite discovery; full CRUD for MongoDB/Oracle; connection renaming
• Chat Enhancements: Reference recent 3 rounds of history, chat attachments, rich-text quote/copy, manual context compression, condensing progress
• New Providers: GitHub Copilot, Kimi Code (device OAuth); Codex usage/icon optimization
• Super Memory: Time-aware UI (FreshBadge, recall cards)
• Models/Providers: MiniMax M3, Opus 4.8, glm-5.1; OpenAI-Compatible config page redesign
• Notification Center: Left sidebar tabs + larger split layout
• Error Display: Structured error cards, expandable raw error details, permanent errors skip retry
• Others: Russian translation, context usage ring/tray

Fixes

• Agent/Tools: Always fresh file reading, file tools synchronized with disk
• Code Review/Rendition: GFM tables, Explore spacing, streaming markdown/mermaid long-session OOM
• Remote/Terminal: Full toolchain for SSH, Windows terminal reflow, terminal ID confusion, tab close deadlock
• UI Tweaks: Activity Bar width adjustment reduces panel jitter, notification panel height, Super Memory retention days wording, etc.

Optimizations

• Performance: Significantly improved cache hit rate, startup/splash LCP, long-conversation memory management
• Dependencies/Base: Electron 39.8 → 42.3.3

1.1.5 · 2026-05-24

New Features

• Chat quoting and copying: After selecting text in a chat, quickly convert it into a Markdown quote or copy it directly, while preserving lists, code blocks, and file name tags
• Automatic notification for background commands: Notifications are automatically sent when long-running terminal tasks complete, eliminating the need to repeatedly check progress
• Mathematical formula rendering: LaTeX math rendering is now properly supported in chats, including $x^2$ and $...$
• Error display redesign: Error messages are shown in expandable cards, with one-click copy of the full error content
• Full SSH remote development support: All tool invocations on remote SSH hosts are now fully supported
• Streaming stutter diagnosis: When a streaming response stalls, detailed diagnostics indicate whether it's upstream latency, network buffering, etc.
• Model context usage bar: In the model selector, clearly see the token usage for the current session, displayed by category
• "Default Model" label: The model selector now clearly identifies the default model
• Long-term memory panel upgrade: New time tags such as "Just Now," "Today," and "This Week," frosted-glass style, plus memory recall cards
• Display plan title during streaming: When the model generates a plan, the plan title appears during streaming output
• Permanent errors automatically skip retries: For permanent errors such as authentication failures, no more meaningless automatic retries
• New model support: Added support for models such as Gemini 3.5 and Qwen 3.7
• Russian language interface support
• Chat tool cards can be expanded: Tool outputs such as search results and file readings support collapsing/expanding

Fixes

• Streaming response timeout increased from 120s to 180s, with an intermediate UI prompt for slower-inference models
• Fixed dropdown menu layering issue on the settings page (third-party provider dropdowns no longer obscured)
• Fixed the "[object Object]" error display issue
• Fixed the non-working error drawer copy button
• Fixed the file path prompt box freezing (no longer remains after scrolling away)
• Optimized chat text selection color scheme, fixing previously broken style variables
• Fixed model selector wheel penetration (scrolling the model menu no longer scrolls the chat area below)
• Schema compatibility fix for thinking mode (improved inference parameter compatibility under the Claude protocol)
• Optimized multi-file streaming preview (real-time "N changes" indicators no longer lag)
• Numerous UI detail fixes (scrolling, spacing, alignment, animations, tooltips)

Optimizations

• Underlying technology stack upgrade: Electron upgraded to 42.2.0, including Chromium 148, Node.js 24, V8 14.8
• Memory management optimization: Large chat-history data is automatically written to disk and inference records organized as needed, keeping long conversations lightweight
• Enhanced Mermaid chart rendering: Improved cache management with proactive SVG cleanup on unloading, preventing memory accumulation during long chats

1.1.4 · 2026-04-26

New Features

• DeepSeek V4 series models: Added DeepSeek V4 items in the model list/configuration, selectable directly
• OpenAI-Compatible "Slot" card interface: Redesigned the card-style UI for OpenAI-Compatible configuration areas

Fixes

• Sidebars and settings no longer "hang": After model switching and rule/memory operations, they refresh promptly without needing a restart
• Fixed file links in chat being unopenable or incorrectly parsed
• More reliable terminal tools: Commands now run through the system terminal service
• Fixed chat-area jumping and erratic height when "Thinking / Streaming Content / Terminal Tool Cards" appeared or switched states, and fixed a crash caused by a variable reference error
• Updated PDF preview and related tool usage/display behaviors
• Corrected persistence and handling of large-volume tool results to reduce anomalies or state loss
• Updated LLM error reporting and chat-area interface resources

Optimizations

• Selection of model panels in the input box
• Increased stability of terminal tools

1.1.3-v2 · 2026-04-23

New Features

• Support for GPT-5.5 model, added to all provider lists
• Added Kimi K2.6 model, available on all major providers (OpenRouter, Together AI, Fireworks, DeepInfra, etc.)
• Math formulas can now be displayed in chat (KaTeX rendering)
• Flowcharts can now be displayed in chat (chart rendering)
• Added web search tool with control over search depth and topic parameters
• Terminal tool upgraded to PTY mode for a more realistic experience
• Added code symbol index with jump-to-definition support
• Added dependency analysis tool (cycle detection, dependency tracking: which files are affected by a change, the dependency chain between file A and B, whether circular dependencies exist)
• Skills now support dragging and compressing files/folders
• Added an IDE update status indicator to the title bar
• Major revamp of the multi-Slot panel: search, sorting, health checks, and batch cleaning of invalid APIs

Fixes

• Long chats no longer freeze: Fixed the interface freezing with too many messages
• Scrolling no longer snaps back to the bottom when viewing chat history
• Fixed terminal subprocess leakage (no more memory spikes or crashes from residual processes)
• Fixed file path display (links no longer corrupted into false paths by ellipsis)
• Fixed super memory blanking (AI thinking tags no longer leak, causing blank memory display)
• Fixed terminal command folding (failed commands no longer swallowed by subsequent successful ones)
• Fixed the style of the plan to-do floating reminder (double-layer background issue)
• Fixed chat jumping when expanding the file list
• AI no longer repeatedly asks "continue?" and instead executes autonomously (self-testing required)
• Aligned the scroll bar in the Skills content box
• Fixed cases where command execution recognized exit 0 but the actual build failed

Performance Optimization

• Significantly optimized chat rendering
• Added memo to Markdown rendering to avoid re-parsing the same content
• Memory optimization: tightened the persistence threshold for tool results to prevent spikes in long conversations
• Token counter upgrade
• Added an LRU upper limit to message cache
• New splash screen and onboarding flow design

1.1.1 · 2026-04-11

New Features

• The chat history panel now has a "New Conversation" button for more convenient operation
• Database functionality significantly enhanced: support for Redis and MariaDB, new cell editing confirmation, table-count badges, and the ability to open the database panel directly from chat
• AI conversation supports switching between Plan/Agent modes for a smoother workflow

Fixes and Optimizations

• Fixed repeated code completion (Tab completion)
• Fixed page jumping when focusing on the file browser window
• Fixed false disconnection reports when minimizing the window
• Fixed multiple bugs such as image indexing, IPC memory leakage, and LLM timeouts in chat
• Fixed the white screen caused by React packaging, and optimized package size
• Fixed several issues such as empty panels and undo in the CSS checker
• Upgraded to TypeScript 6.0 and Electron 41.2.0 for more stable overall performance
• Comprehensive UI polishing: unified tool card backgrounds and borders, aligned editor tab icons, fixed diff code block scrolling, etc.

1.1.0 · 2026-04-09

New Features

• Free code completion: All users can now use AI code completion for free without additional configuration
• Significantly upgraded built-in browser: Supports device simulation (mobile/tablet preview), dynamic tab titles and icons, zoom control, optimized CSS inspector, and real-time recording of animation/transition events
• Smarter conversation memory: New automatic de-duplication and merging mechanism so the AI recalls previous content more accurately and avoids duplicate storage
• Sidebar toggle button: A quick sidebar toggle button added to the title bar
• Task list optimization: Supports task merging, priority sorting, and new transition animations
• Redesigned team dashboard: Brand-new dark theme design and faster page loading
• Lazy loading of conversation messages: Long histories are no longer loaded all at once, resulting in faster opening
• API cache mechanism: Added Responses API cache to reduce duplicate requests and save tokens

Improvements

• Editor backend upgrade: Electron upgraded to 41.2.0, TypeScript to 6.0, for more stable and smooth operation
• Redesigned AI tool invocation interface: unified card styles for file operations, MCP tools, etc.
• Code diff optimization: fixed scrolling in collapsed state, horizontal scrollbars, and background color display
• System notification optimization: avoid duplicate pushes, no disconnection reports when minimizing
• Terminal command detection optimization: more accurately identify whether commands have run
• Improved startup performance: optimized startup flow and conversation list rendering
• Prevent background memory overflow: automatically clean up accumulated memory during long-term operation

Fixes

• Fixed queued messages possibly being sent to the wrong thread when switching conversation threads
• Fixed deleting an image attachment possibly deleting other file attachments
• Fixed the background process continuing to run after an AI response timeout
• Fixed the TrustedHTML compatibility issue causing blank pages
• Fixed clicking links in the built-in browser failing to open new pages correctly
• Fixed option buttons occasionally not appearing when asking AI questions
• Fixed misaligned editor tab icons

1.0.9 · 2026-04-03

New Features

• Built-in browser automation: Integrated CDP browser tools (screenshot, click, input, navigation, DOM snapshot, console log); the AI Agent can directly operate browser pages
• Device viewport simulation: The built-in browser supports simulating different device resolutions (mobile/tablet/desktop), with a scrollable device list
• Browser tab management: Dynamic tab title/icon, zoom control, multi-tab switching, quick open of the input-box browser panel
• Multi-tool parallel invocation: Supports the AI returning multiple tool invocations for parallel execution at once
• Title bar sidebar toggle button
• Entanglement panel switching: Supports independent panel switching with automatic alignment when adjusting the sidebar

Fixes

• Terminal command completion detection: Enabled OSC 633;D protocol, fixed ANSI cleaning, accelerated heuristic judgment
• Teammate system notification suppression: Sub-threads no longer send irrelevant notifications to the OS
• Entanglement mode override: User-selected entanglement mode is no longer silently overridden by schedule mode
• Intent leakage: Intent labels in entanglement causal mode no longer leak to displayed text

Optimizations

• Startup performance: Chat O(n) optimization, diff statistics persistence
• Terminal tool UI: Title bar/bottom bar inner-margin optimization, "Open in Browser" button corner-radius correction
• Browser panel positioning: Use floating-ui to eliminate jitter/offset when opening