0.90.0

Highlights

📁 File Editor Modal Overhaul: File Tree Sidebar, Tab Context Menus & State Persistence

Navigating and editing project files without leaving your active task just got significantly easier. The workspace file editor modal now features an integrated file tree sidebar, allowing you to browse, expand folders, and open files directly inside the editor view. We have also added tab context menus (Close Others, Close to the Right, Close All) for faster tab hygiene, alongside full editor state persistence - your open tabs, active tab, and unsaved editor content stay intact even if you close and reopen the modal or navigate between views.

⚙️ Dedicated Model Overrides for Compaction & Handoff

You can now configure separate model overrides for conversation compaction and agent handoffs in Context Management settings. High-end reasoning models are ideal for complex coding prompts, but using them for compaction summaries can consume unnecessary tokens and risk hitting rate limits. With this update, you can route background summarization and compaction workloads to faster, cost-effective models (such as Claude 3.5 Haiku, Gemini 2.0 Flash, or GPT-4o-mini) while keeping your primary frontier model focused on code generation. Learn more in the Handoff & Compaction documentation.

🎯 Editable Model Modalities & Capability-Aware Attachment Filtering

Different models excel with different inputs, and the Model Library now makes this explicit and configurable. You can view and edit supported input and output modalities (text, image, audio, video, PDF) for any provider or model. AiderDesk uses these capabilities to automatically filter incompatible file attachments before sending requests to models that cannot process them, preventing frustrating API errors. For custom or unknown model identifiers, attachments are preserved as-is so new endpoints and experimental multimodal workflows continue to work smoothly. Check out the Providers documentation for details.

🤖 Enhanced Task Tools: Prepared Tasks, Execution & Autonomy Controls

Agent task orchestration receives major improvements with expanded controls for Task tools:

• Prepared Tasks: Agents can prepare tasks with initial prompts and start or resume them later without restarting conversations.
• Execution Controls: Support for explicit execution modes (create_only, wait_for_finish, run_in_background).
• Autonomy Settings: Fine-grained autonomy levels (manual, guided, autonomous) can now be passed directly to task executions.
• Readable Tracking: Tool messages in the chat interface now display actual task names rather than raw identifiers, making multi-task workflows much easier to follow.

See the Task Tools documentation for more information.

🔌 Extension Ecosystem: Background Generation, Worktree Hooks & Custom Catalogs

This release introduces substantial capabilities for extension developers:

• Background Context Generation: Extensions can now trigger background text generation and streaming (generateText, streamText) directly within task context.
• Worktree Lifecycle Hooks: New onWorktreeCreated and onWorktreeMerged hooks allow extensions to run project-specific automation, custom environment setups, or branch synchronization scripts when switching working modes.
• Custom Catalogs & Independent Favorites: Extension model selector components can now provide custom model catalogs and maintain their own favorite selections independently from global model settings.
• Reliable Recovery: Fixed a bug where onPromptFinished was missed during agent resume and restart recovery.

Explore the updated Extension API Reference to start building with these new APIs.

Other Changes

• Security & Runtimes:
  • Updated Electron, Node.js, and uv runtimes for upstream security vulnerabilities.
  • Added automatic Python certificate bundle updates and startup warnings for unpatched Node.js versions.
  • Prevented potential OS command injection in extension repository clone paths.
• Provider & Model Configuration: Added required streaming settings for providers and individual models, allowing you to enforce streaming on models that mandate it.
• Mobile & Remote Reconnects: Fixed duplicate prompt execution caused by mobile clients replaying requests after reconnecting to the server.
• Chat UI: Improved grouping and spacing of tool-only messages in Compact message view for a cleaner conversation flow.
• Attachments: Preserved detected MIME types in context image attachments instead of falling back to generic image types, preventing request failures on less common formats.

🧩 Extensions

• Sound Notification: Made audio playback non-blocking with format-aware Linux player fallbacks (paplay, aplay, pw-play) and an execution watchdog to prevent notification hangs.

💡 We'd love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

0.89.0

Highlights

🔄 Automatic Task and Subagent Recovery After Restarts

Agent runs frequently perform multi-step workflows across tools and subagents. In the event of an unexpected application restart or crash, you should never lose completed progress or risk re-executing unsafe actions. AiderDesk now features automatic task and subagent recovery, durably checkpointing streamed assistant responses, tool calls, and finalized outputs as they happen directly into the local application database.

When your project reopens, interrupted runs resume right where they left off. Completed tool calls retain their results instead of repeating, subagents recover under their original identity to avoid duplicate tasks, and unfinished operations are handled according to configurable per-tool recovery modes (Replay safely, Notify interrupted, and Forget unfinished). Non-idempotent actions like shell commands or file edits notify the model to inspect current workspace state before retrying, preventing duplicate side effects. You can fine-tune recovery modes per tool in Settings → Agents → Tools or toggle startup resume in Settings → General → Start Up → Recovery.

📸 Pre-Compaction Context Snapshots Kept for Agents

Context compaction keeps long-running tasks responsive and within model token budgets, but earlier conversation details were previously summarized and replaced. With pre-compaction context snapshots, AiderDesk now saves the entire uncompacted context into sequentially numbered files (context.pre-compaction.001.json) within the task directory before every manual or automatic compaction.

These snapshots are permanently retained and never pruned. Agents running with the built-in system prompt are automatically informed of existing snapshot paths, giving them the ability to inspect earlier messages whenever historical implementation details, past reasoning, or discarded code snippets are needed later in the workflow.

📊 Theme-Aware Mermaid Diagrams with Pele

Visual architecture diagrams and sequence charts are now rendered using the pele library, bringing a substantial visual and usability upgrade to Mermaid rendering in chat. Diagrams adapt to your active theme with coordinated colors, typography, and crisp styling in both light and dark modes.

Diagrams also support smooth streaming updates as your agent writes Mermaid code, keeping the previous valid diagram intact during syntax generation instead of flickering or breaking. Built-in interactive zoom and reset controls make navigating large system architecture diagrams effortless.

🌿 Push Worktree Fixes Directly to Base Branch Upstream

Developing features and bug fixes in git worktrees provides clean isolation, but pushing those fixes back to the main remote previously required extra branch manipulation or switching back to the local repository. The Git push dialog in worktree tasks now includes an option to push directly to the base branch's upstream remote.

With a single checkbox, your worktree commits are pushed straight to the base branch's tracking remote destination, eliminating repetitive Git gymnastics when publishing patches or preparing releases from isolated workspaces.

🛡️ Native Tool Approval for Extension Providers

External agent providers and custom model extensions can now integrate seamlessly with AiderDesk's safety policies via the new ProviderToolApprovalHandler interface. When external agents execute native tools (such as OpenCode native shell or file operations), providers can check tool execution directly against AiderDesk's active approval engine.

This ensures task autonomy settings (Manual, Guided, Autonomous), Power tool approval policies (Always, Ask, Never), bash command deny-lists, and extension approval veto hooks (onToolApproval, onHandleApproval) are fully enforced, giving you uniform governance regardless of whether tools run natively or through AiderDesk.

Other Changes

• UI & Small Screen Responsiveness:
  • Fixed dialog overflow on small screens with responsive widths and auto-wrapping footer action buttons.
  • Fixed notification overflow and centering on compact displays.
  • Improved Model Library layout on small screens with a wrapping toolbar and a horizontally scrollable models table.
  • Matched message map scrollbar styling to the rest of the application interface.
• Agent & Tool Reliability:
  • Fixed duplicated and incorrectly grouped assistant messages when streaming reasoning traces alongside tool calls.
  • Fixed application crashes when displaying truncated tool results while preserving full tool input parameters and inspection details.
  • Gracefully handled retired AWS Bedrock models with debug logging instead of hard errors.
• Internationalization:
  • Added German (de) locale translation. (Thanks @777marvin)

🧩 Extensions

• OpenCode Free: A new experimental agent provider offering free models from OpenCode. Runs an owned, isolated OpenCode server with persistent session storage, native shell/read approvals tied to AiderDesk policies, and loopback MCP bridging for remaining AiderDesk tools.
• Claude Agent SDK: Added multimodal support for attaching and inspecting images and PDF documents.

💡 We'd love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

0.88.0

Highlights

🛡️ Project Trust Confirmation for Safe Workspaces

Opening untrusted projects should never silently run unknown configurations. When opening a repository that defines custom agents, MCP servers, or project-local extensions through the UI, AiderDesk now presents a trust confirmation dialog. You can inspect the configured agents, tools, and extensions before granting permission to load them, giving you complete visibility and control over what runs in your workspace.

📐 LaTeX Formula Rendering in Chat Messages

Chat messages now feature native LaTeX formula rendering, supporting both inline ($...$) and block ($$...$$) mathematical expressions via KaTeX. Whether you are discussing machine learning architectures, reviewing numerical algorithms, or working through mathematical proofs with your agent, equations and matrices render clearly and legibly right in the conversation flow.

📁 File Rename Tracking in Updated Files

Tracking file modifications across renames is now completely seamless. The Updated Files sidebar automatically detects git renames and presents a distinct rename badge showing previous and current filenames. Opening a renamed file provides a rename-aware side-by-side diff displaying edits made alongside the rename, and single-click reverting restores both the filename and its contents. The Extension API has also been upgraded with full support for renamed files.

🌿 Clean Worktrees from Selected Branches

Branching off to test a bugfix or explore a new concept shouldn't mean carrying half-finished local edits with you. The Git branch menu now features an action to create a clean worktree from a selected branch without migrating uncommitted local changes. This lets you spin up a fresh, isolated workspace from any branch while leaving your current task working tree completely untouched.

🧩 Extension API: Message Transformation & Cache Control Hooks

The Extension API continues to expand with low-level message orchestration hooks:

• onTransformMessages - Allows extensions to intercept, inspect, and transform the conversation message array before dispatching requests to LLM providers.
• Cache control hooks - Provides extensions with fine-grained control over provider prompt caching behaviors at the individual message level.
• Message optimization hooks now receive transformed messages, ensuring multi-extension pipelines compose predictably.

⚡ Faster Task Switching & UI Performance

Switching back and forth between long-running tasks is now near-instant. By preserving virtualized message size caches across task switches, the chat view avoids recalculating heights for hundreds of rendered messages, eliminating layout shift and rendering pauses. Paired with lazy item construction in the command palette, faster settings caching, and optimized task list rendering, the UI remains snappy regardless of workspace size.

Other Changes

• Command Palette:
  • Added quick actions for switching task modes directly from the command palette.
  • Improved background performance by constructing palette items only while the palette is open, while preserving active selections during live updates.
• Context Compaction:
  • Fixed task auto-compaction not being disabled when its threshold was set to 0.
  • Fixed image tool results triggering unintended auto-compaction caused by inflated token count estimates.
• Reliability & Bug Fixes:
  • Fixed failed subagent runs incorrectly returning success with empty or partial results.
  • Fixed bash tool calls hanging when command output processing encountered an error.
  • Fixed token counting crashes on text containing special tokens, adding graceful fallback estimation when tokenization fails.
  • Fixed malformed or incomplete compact diffs crashing the chat view, with automatic fallback to plain text.
  • Fixed Aider tool responses losing their grouped visual layout after reloading tasks.
• Extension Settings: Fixed extension settings actions not respecting the currently selected project context.

🧩 Extensions

• Sound Notification: Added custom audio file support, allowing you to select your own audio files for task completion and event notifications.
• Skill Gallery: Added recursive skill scanning for nested directories and added Android source repository support.
• Cursor SDK: Fixed run lifecycle state handling and context message ordering for Cursor agents.

💡 We'd love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

0.87.0

Highlights

🗺️ Message Map: Navigate Long Conversations at a Glance

Long agent runs can produce hundreds of messages, and finding that one moment used to mean endless scrolling. The new message map gives you a compact, color-coded overview of the entire conversation - user messages and assistant responses are visually distinguished, hovering over a segment shows a preview of its content, and clicking jumps straight to that spot in the chat. The map is enabled by default - if you prefer a cleaner look, you can disable or hide it in Settings -> General -> Message Display. Combined with optimized marker rendering, navigating even the longest tasks is now effortless.

⌨️ Segment-by-Segment Path Completion in the Prompt Field

The prompt field has long offered fuzzy path completion - now you can accept paths segment by segment. Press the right arrow key to fill in only part of the selected item's path, which is perfect for autocompleting the folder portion of a file before continuing to type. Less typing, fewer mistakes, faster prompts.

🪝 Skip Git Hooks on Commit

Sometimes you need to commit without triggering pre-commit or commit-msg hooks - for example during quick fix-ups or when hooks are slow. The commit flow now includes a "Skip Git hooks" option, giving you full control over when hooks run.

🧩 Extension API: Custom Mode Icons and Richer Tool Definitions

Extension authors get two upgrades:

• Custom modes can now use icons from icon packs, so extension-provided modes feel right at home in the UI.
• Tools registered through the Extension API now support an optional description and inputSchema, enabling better tool documentation and stronger validation.

Other Changes

• Diff View: The diff view mode switcher has been removed from code block headers - the view mode is now configured globally in Settings. This simplifies the UI and improves code block rendering performance, especially in tasks with many code blocks.
• Stability: A round of crash and reliability fixes:
  • Fixed a React "maximum update depth exceeded" crash when interacting with the diff viewer dropdown while a task is running (and a similar crash from select popover transitions during streaming).
  • Patched @legendapp/list to prevent a crash in the message list caused by excessive container measurements.
  • Fixed the terminal staying open with a dead session when the shell exits (e.g. after Ctrl+D).
  • Fixed task state not re-syncing after a socket reconnect.
  • Fixed the paused chat scroll position being lost when switching between tasks.
  • Fixed delayed expand/collapse of folders in the updated files diff sidebar.
  • Fixed confirmation dialogs being clipped inside messages when deleting memories.
  • Fixed a stale "(Not found)" being shown in the stored memory tool message until refresh.
• Security: Upgraded simple-git to 4.0.2 and @modelcontextprotocol/sdk to 1.32.1, resolving critical git command injection and high MCP OAuth credential leak advisories.
• Platform: Updated Electron to 44.7.0, which also fixes delayed paste (Ctrl+V) on Linux after the app has been idle.
• Telemetry: Provider and model details are now included in telemetry events.

🧩 Extensions

• Programmatic Tool Calls: Added a dedicated PTC mode where the model orchestrates all tools through sandboxed code - saving context tokens and enabling batching, filtering, and complex control flow. Also added tool discovery helpers (searchTools, describeTool) and structured, model-friendly diagnostics for sandbox failures.
• Cursor SDK: Activated skills are now injected into the Cursor agent, so your installed skills work there too.
• Skill Gallery: Added Cursor Pstack as a default source repository.

💡 We'd love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

0.86.0

Highlights

🧩 Extension API: Toolset Hooks, Provider Access, and Compaction Control

The Extension API keeps growing with three significant additions that open up deeper integration than ever:

• onToolsetCreated - a new extension event fired for every agent run, letting extensions inspect and modify the toolset before the run starts.
• getProviders - access your configured provider profiles directly from the extension context.
• smartCompact and getEstimatedTokens - programmatic context compaction and token estimation from the task context.

These APIs already power new extensions shipping in this release - most notably the Compaction Studio extension (see the Extensions section below).

🧠 Memories Isolated per Agent Profile

Memories were previously shared across all agent profiles, which could blur the line between a profile tuned for quick code edits and one set up for deep architectural work. Each agent profile can now keep its own isolated memory store, so every profile builds up knowledge that stays relevant to the way it works. Enable isolation in the agent profile settings if you want your profiles to remember things independently.

🔀 Agent-Based Conflict Resolution for Project Roots

When a rebase runs into merge conflicts, AiderDesk can hand them to the agent to resolve. This previously worked only in worktrees - now it also works at the project root level, so you get the same smooth, agent-assisted conflict resolution no matter which working mode you prefer.

🚀 Choose Your Path: Agent Mode or Code Mode

New to AiderDesk? The onboarding now lets you explicitly choose how you want to work - the autonomous agent mode or the classic code mode. To match, there's a new setting for the default mode used when creating new tasks and new projects, so AiderDesk always starts the way you like it.

🧰 Extension Tools in the Agent Selector

Tools provided by extensions now show up in a dedicated Extension Tools section of the agent selector, and the list refreshes automatically whenever extensions are installed, updated, or removed. No more guessing which extension tools are available - and no more restarting to see changes.

Other Changes

• Prompt Input: Pressing ArrowUp in an empty prompt field now enters edit mode for your last user message - a familiar shell-style shortcut for quickly fixing a previous instruction.
• Reconnect Handling: Tasks and the readonly view now automatically resynchronize their state after a socket reconnect, keeping everything consistent after connection hiccups.

Bug Fixes

• Fixed a crash during streaming (React maximum update depth exceeded) in the message list.
• Fixed the updated files section not refreshing for tasks sharing the same worktree or project root.
• Fixed task.json and todo.json being left half-written when the app crashes during save (writes are now atomic).
• Fixed commits failing for already-staged file deletions.
• Fixed the todo list not auto-expanding again when a completed list is replaced with new items.
• Fixed subagent token usage overwriting the main run's context size info.

🧩 Extensions

• Compaction Studio: A new extension for manual context compaction with full control. Pick exactly what to compact (errored tool calls, file reads, searches, bash outputs, reasoning, and more), protect a safe window of recent messages, and choose from three compaction levels - all with per-category token savings and before/after estimates, powered by the new smartCompact API.
• Jev: A new extension bringing Jev - a fast, typed decision model (~300 ms, a fraction of a cent per call) - to your agents via ask-jev and ask-jev-files tools. Agents ask yes/no, choice, or score questions about files, command output, or state, and receive only the typed answers, never the underlying content.
• DBX: A new extension embedding the DBX database client into AiderDesk, giving you a full UI for 100+ databases (MySQL, PostgreSQL, SQLite, Redis, MongoDB, and more) plus message queues like Kafka and RabbitMQ - without leaving the app.
• Binary Files: Added a custom message renderer for binary file extension tool results.
• OpenAI Codex: Added GPT-6 series models.
• Git Log: Replaced custom action dialogs with consistent ConfirmDialog components.

💡 We'd love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

0.85.0

Highlights

🛠️ Take Control of Your Skills

Skills just got a serious upgrade in flexibility. You can now disable skills per project, so each task only loads the skills that actually matter for the codebase it's working in. Skill authors get new frontmatter options too: disable-model-invocation prevents the agent from triggering a skill on its own, while user-invocable skills let you decide exactly when a skill should run. To make skills more discoverable, built-in skill names now appear in the / command autocompletion.

👁️ Live File Watching During Tasks

A new file watcher now monitors the task directory while a task is running, keeping task files, worktree status, and autocompletion fresh in real time. When your agent creates, modifies, or deletes files mid-run, the UI reflects those changes immediately - no more stale file lists or outdated worktree information while the work is still in progress.

🔍 Find in File Editor

The file editor now supports find in file with Ctrl+F / Cmd+F. Quickly locate that one function or string in larger files opened from your context without leaving the editor.

🤖 New LLM API Provider

AiderDesk now ships with a built-in LLM API provider. Add your API key (or set the LLMAPI_API_KEY environment variable) and your available models load automatically - with pricing and context information included. One more way to connect the models you want to the agent workflow you already have.

🧩 Extension API: Message Dropdown Menu Actions

Extensions can now attach custom actions to the message dropdown menu via the new task-message-bar-menu UI placement. This opens the door for extensions that operate on individual messages. And to make custom extension commands more powerful, images attached to the prompt are now passed to custom extension commands.

Other Changes

• Prompt Autocompletion: Added a /resume command for quickly resuming previous tasks.
• Command Palette: Added actions to show or hide archived tasks.
• Project Tabs: Added a context menu action to open the project folder directly from project tabs.
• Model Library: Updated default models for built-in providers to their latest versions, and added a low reasoning effort option for the Z.ai plan provider.
• Worktrees: Clarified the worktree to local mode switch dialog for shared worktrees.
• Task Sidebar: The in-progress spinner now propagates to the closest visible ancestor task when parent subtasks are collapsed.
• Code Highlighting: Falls back to plain code when the auto-detected language yields no highlighting matches.

Bug Fixes

• Fixed a crash during message streaming (React maximum update depth exceeded) caused by duplicate message ids.
• Fixed duplicate keys in git branches popup rows.
• Fixed committing deleted files by retrying staging with git add -A -f.
• Fixed partial commits including unrelated staged files - only the selected files are now committed via pathspec mode.
• Fixed in-progress tasks being evicted from the task mount pool when switching tasks.
• Fixed the find in file shortcut being swallowed by inactive background project task views.
• Fixed the wrong scroll offset in the updated files diff modal when tracking headers were not found.
• Fixed provider options leaking into task context messages.

🧩 Extensions

• Cursor SDK: Improved context message persistence and MCP server config retrieval.

💡 We'd love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

0.84.0

Highlights

📁 Git Support for Every Project

Projects that aren't git repositories can now join the party! AiderDesk will automatically initialize a git repository for non-git projects, unlocking everything that comes with it: branches, worktrees, and version tracking. No more sitting on the sidelines - all of AiderDesk's git-powered workflow is now available for any project you open.

⚡ Major Performance Improvements

This release is full of speed-ups across the board:

• Workspace file trees are now virtualized (migrated to headless-tree), so large projects render smoothly regardless of how many files and folders they contain.
• Updated files diffs are loaded lazily on demand instead of being computed upfront with the file list - the diff modal opens noticeably faster.
• Background projects stay lightweight: a task is only loaded when you first activate the project, reducing memory usage while you work elsewhere.
• Tooltips render faster across the app.

🛡️ Safer Git & Worktree Workflows

We've hardened the git and worktree experience with a series of reliability improvements:

• Switching to local mode is now blocked when the worktree has uncommitted changes, so you'll never silently lose work again.
• Creating a task worktree whose branch is already checked out in another worktree now creates a unique sibling branch instead of failing.
• Task branches are no longer renamed when a worktree is shared with other tasks.
• Ignored files are now staged with --force on retry, so ignored files no longer fail to commit.
• Uncommitted diffs work in freshly initialized repositories with no commits yet (diffing against the empty tree, including untracked files).

🔐 OpenAI Zero Data Retention Support

A new "Store Responses" setting for OpenAI lets organizations with Zero Data Retention (ZDR) requirements use AiderDesk by running requests in stateless mode. If your organization doesn't allow OpenAI to store request data, this setting keeps you compliant.

✅ Bulk Tool Approval for MCP Servers

Approving MCP tools one by one is history: a new "All" select in the server tools list lets you bulk-approve tools per MCP server. Combined with the new keyboard navigation in the git branches dropdown (arrow keys, submenu controls, and Escape to close), power workflows just got a lot smoother.

Other Changes

• Task Management: Added an ephemeral indicator for transient message groups, and task IDs are now stored in the messages database, enabling per-task usage tracking.
• Agent Reliability: The agent no longer stops prematurely when a model finishes with stop but produces no messages (or only reasoning content) - the step is retried automatically.
• Updated Files Diff Modal: Files are now ordered to match the sidebar tree, commit errors show clean, selectable messages (no more IPC wrapper noise), and full old/new file contents are fetched on demand.
• Markdown Rendering: Numbered lists starting mid-paragraph now render as proper lists instead of plain text.
• In-Chat Search: The messages scroll container is now reliably detected via a dedicated callback.
• Model Library: The temperature column now fits its values.
• UI Polish: Toast notifications got a more visible border (errors use the error color), and the extensions search field is auto-focused when opening extensions settings.
• Extension API: projectDir is now passed to extension component wrappers on header placements.
• Platform: Updated Electron to 44.3.0.

🧩 Extensions

• Git Log: Added per-file diff sections with scroll-to-file, a commit context menu, and smarter project directory selection that prioritizes the current project.
• Task Scheduler: Added one-time scheduling - schedule a task to run once at a specific time.
• SearXNG Search: Now talks to the SearXNG API directly instead of spawning a CLI process - more reliable and faster.
• tps-counter: Improved streaming and subagent token rates (Thanks @pcfreak30)
• Cursor SDK: Fixed the vendored ripgrep path configuration.
• Reasoning Selector: Added an extension icon.

💡 We'd love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

0.83.0

Highlights

🖥️ Terminal Sessions That Stick With You

Terminal sessions now persist when you switch between tasks or projects. Come back to a task and your session is reattached instantly with full output replay, so you never lose track of what happened. Sessions restart automatically after they exit, and touch input is now supported - a much more reliable terminal experience all around.

⚡ Instant Task and Project Switching

Recently active tasks now stay mounted in the background, making switching between tasks and projects nearly instantaneous. No more waiting for everything to reload every time you jump around your workspace.

📝 Bash Command Descriptions

Modern models love using shell commands for all kinds of actions, and sometimes it can be hard to understand what a command actually does. Bash tool commands can now carry an optional description shown right in the tool message, so you can see at a glance what the command is supposed to do.

🔐 Interactive Git Authentication

Git operations that require authentication now prompt you interactively for credentials and passphrases instead of failing silently. Whether it's cloning a private repository, pushing, or pulling - when Git needs credentials, AiderDesk asks.

🛡️ Automatic Renderer Crash Recovery

If the renderer process crashes, AiderDesk now recovers automatically by reloading, with crash loop protection to prevent endless restart cycles. Event delivery has also been hardened to gracefully handle sending to destroyed or crashed windows. A crash should no longer take your whole session down with it.

🌳 Worktree Post-Create Command

You can now configure an optional post-create command that runs in a freshly created worktree, with project, worktree, task, and branch environment variables available. Perfect for initializing submodules, installing dependencies, or any other setup your project needs before work begins.

⌨️ Task Management in the Command Palette

The command palette is now a task management hub: create subtasks and sibling tasks, rename tasks, change their state, and copy task IDs (Thanks @pcfreak30) - all without leaving the keyboard.

Other Changes

• Updated Files Diff View: You can now select which changed files to include when committing, and file selection is properly restricted to uncommitted files.
• Copy as Image: Chat messages can be copied as an image to the clipboard.
• Custom Commands: Queued custom commands now run directly from the queue instead of appearing as user prompts, and they queue correctly while another prompt is running.
• Chat Auto-Scroll: Auto-scroll now pauses only when you scroll upward and resumes reliably when you return to the bottom - no more false pauses from internal scroll adjustments.
• Worktree Reliability:
  • Fixed worktree rebase silently dropping commits when the tracked base commit was stale.
  • Using merge-base for worktree flat diffs to avoid reverse diffs.
  • Uncommitted changes are carried over when switching to local after a checkoutless merge.
  • Fixed a subtask inheriting the parent worktree when its working mode was overridden to local.
  • The branches dropdown refreshes correctly after a branch switch or rebase completion.
  • The git branches tooltip now shows icons and a rebase hint when a worktree is behind its base branch.
• Git: Adding a file to Git now updates the file state optimistically without refetching, and checkout no longer detaches the main repository branch.
• Prompt Field: The Escape key no longer interrupts task processing while you're typing.
• UI Polish: Tool message status icons no longer shrink when message content is long, and password fields got improved toggle alignment with new visibility translations.
• Performance: Reduced the diffs worker pool size and AST cache, lowering memory usage.

🧩 Extensions

• BMAD: Fully rewritten around bmad-method v6, bringing the latest BMAD methodology workflows to your projects (Thanks @777marvin). Also fixed the default Welcome message showing outside of BMAD mode.
• OpenAI Codex: Added a 5-hour usage window display, and token storage was moved to the extension folder with legacy migration.
• Cursor SDK: Updated @cursor/sdk to the latest version and improved tool message error handling and result formatting.

💡 We'd love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

0.82.0

Highlights

🌿 Full Git Branch Management, Safer Pushes & AI-Powered Recovery

This release turns AiderDesk into a complete git workstation. The new branch management UI lets you work with branches without ever touching the terminal - and when something goes wrong, your AI agent is one click away from fixing it.

• Branch management: list, create, checkout, rename, and delete branches directly from the UI.
• Safer pushes: a confirmation dialog now shows you exactly which commits are about to be pushed, with an opt-in force push option when you really need it.
• Smarter pulls: when branches have diverged, you can choose the pull strategy that fits your workflow.
• Automatic upstream setup: pushing a branch without a remote tracking branch now sets up the upstream automatically.
• Checkoutless merge/squash for worktrees: merge or squash worktree changes into the target branch without ever checking it out - faster and safer for worktree-based workflows.
• Resolve with AI: when a git operation fails (conflicts, interrupted rebases, and friends), a new "Resolve with AI" action hands the problem to your agent to analyze and fix the error for you.
• Command palette integration: all git actions are now available in the command palette with configurable hotkeys in Settings.
• Reliability fixes: worktree commit tracking and operations now correctly use the base branch, integration checks no longer fail when a worktree directory no longer exists, rebasing onto your own branch is prevented, and git state refreshes automatically.

🔍 Search Gitignored Files with the grep Tool

The grep power tool now accepts an ignoreGitignore input parameter, letting your agent search inside gitignored files. When the answer hides in build outputs, vendored dependencies, or generated code, your agent no longer comes back empty-handed.

🧠 Delete Stored Memories on the Spot

Agents build up memories as they work with you - and now you have immediate control over what sticks. When the store_memory tool saves something you don't want to keep, you can delete it right from the tool message the moment it's created. No digging through settings, no letting bad memories linger.

Other Changes

• Task Management: Improved the create_task tool description and worktree inheritance for new tasks.
• Subtasks: Distinct "Create subtask" label when creating subtasks via the create task tool message.
• Reliability:
  • Atomic context saving with corrupt recovery, so a crash can no longer corrupt your task context. (Thanks @pcfreak30)
  • Automatic retry for transient network errors during window load prevents the dreaded blank window.
• Performance: Sped up loading of available extensions. (Thanks @pcfreak30)
• Providers: Added the x-opencode-session header to the OpenCode Go provider.
• Bug Fixes: Fixed the working mode switch dialog staying open and appearing frozen while switching.

🧩 Extensions

• OpenAI Codex: Dedicated sign-in UI with device flow authentication for a smoother Codex account setup, plus the new gpt-6-astra model. (Thanks @pcfreak30)
• TPS Counter: Diagnostics logs moved to debug level to reduce noise. (Thanks @pcfreak30)

💡 We'd love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

0.81.0

Highlights

🌳 Nested Subtasks

Complex work no longer has to be flattened into a single level of subtasks. Subtasks can now have subtasks of their own, letting you (and your agent) break work down into a true hierarchy - epics, phases, and individual steps - all visible as a tree in the task sidebar. In agent mode, the task creation tool supports nesting natively, so your agent can organize bigger endeavors on its own.

⌨️ Activate Skills with the /skill: Command

Skills are now just as easy to trigger as any other command. Type /skill: in the prompt field and pick the skill you want from the autocomplete list, optionally followed by a prompt to run it with. No more digging through menus - your skills are a few keystrokes away.

🔐 Clone Repositories over SSH

Adding a project from a remote repository now supports SSH URLs alongside HTTPS. If you work with private repos or already have SSH keys set up, cloning is now seamless - no tokens required.

🧩 Discover Popular Extensions

The extensions gallery now shows install counts for available extensions, so you can see at a glance which extensions the community relies on most. You can also sort by most installed or by name - perfect for finding proven favorites or discovering new tools worth trying.

🖼️ Queued Prompts with Images

Queuing prompts while your agent is busy keeps the momentum going - and queued prompts can now include images too. Attach screenshots or mockups, and they'll show up with an image indicator in the queued prompts list so you always know what's waiting in line.

Other Changes

• Worktree Reliability: Improved worktree pruning, and fixed worktree merges failing on in-progress rebases or colliding with untracked stash files (Thanks @sbodese).
• Auto-Compaction: Fixed auto-compaction not triggering for models without a known context window - it now falls back to the configured token threshold. For Ollama models, the context window is now read from the model's reported context_length, enabling the token usage bar and percentage-based compaction (Thanks @sbodese).
• Task Management:
  • The task creation tool now has clearer create only, wait for finish, and run in background modes.
  • The workspace sidebar size is now stored on the project level instead of per task.
  • Subtasks of archived parents are now hidden at any nesting depth when archived tasks are hidden.
  • Added loading indicators for workspace sections.
  • Added a highlighted background for the task sidebar filter and search icons when active.
• Voice: Added support for the gemini-3.5-transcribe-live transcription model, now the default for Gemini voice.
• Extensions: Project started events are now re-delivered to extensions that load or reload after a project has started, fixing a startup race; fixed queued prompts for extension custom agent runs.
• Bug Fixes:
  • The MCP servers initialization loading message is no longer shown without prompt context.
  • The agent loop no longer silently re-sends the same prompt when step processing fails.
  • Usage reports no longer fail when provider models fail to load.
  • Fixed context size undercounting that could skip compaction when tool results were appended after the last usage report.

🧩 Extensions

• Learn: A new extension that adds a /learn command to create reusable skills from any source - directories, URLs, conversation history, or pasted notes.
• Provider Quota: Added OpenCode Go provider quota support.
• Cursor SDK: Agent run cancellation now uses an abort signal for cleaner stops.

💡 We'd love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀