Release history
Review what AGI Cockpit added, improved, and fixed in each published version, together with its release date and distribution status.
Some screens and steps may differ in the latest version.
Published AGI Cockpit releases appear here in reverse chronological order. Use this page to inspect the changes in a specific version or compare it with the version you currently run.
How to read the release history
Each entry includes a version number, release date, change type, and main changes. latest marks the current release. When a release remains downloadable, its link points to a platform that was actually distributed for that version.
The release data is also available as machine-readable JSON.
All releases
This list contains every published release since v2.0.
- Added creating child tasks from the child-tasks panel, either starting fresh or branching from the parent.
- Added a CLI command to change the model of a running task.
- Added quoting a preview or HTML surface location into the task instruction.
- Added editing of a Fleet node's agent, model, and reasoning effort from the node detail panel.
- Added answering a multi-question Ask with one free-form reply.
- Added a remote PWA text size setting.
- Changed project pinning so it is available from the project menus.
- Fixed the workspace picker overflowing the viewport on short windows.
- Fixed Fleet row controls so they sit on the right edge of the task list row.
- Added pinned answers for Cockpit agent tasks, with quick navigation on Desktop and the PWA.
- Added text search in Desktop chat and side panels.
- Added an overview of open browser tabs in Settings.
- Added per-tab browser audio muting.
- Added project-specific icons to project lists on Desktop and the PWA.
- Added CLI controls for changing project icons.
- Added per-recipient delivery status for Talk Room messages.
- Added sign-in actions for accounts requiring authentication on the Usage screen.
- Improved usage cards to make account usage easier to review.
- Improved shared agent settings organization and default-agent selection.
- Added a CLI setting for sending follow-ups as queued messages or steering the current turn.
- Fixed scrolling in the Skills settings screen.
- Improved app responsiveness during startup and shutdown on macOS.
- Improved responsiveness in long conversations.
- Fixed detection of Codex credit exhaustion as an account restriction.
- Fixed wrapping and overflow in the Autorun execution target selector.
- Added search by name or task ID when choosing an Autorun target task.
- Added rings showing the remaining allowance for each quota in the usage overview.
- Added countdowns to quota resets in the usage overview.
- Improved visual-mode queued messages to start separate turns after the previous response completes.
- Fixed cached model information display when switching agents.
- Improved model information placement on the PWA task creation screen.
- Added persistent task search queries in the PWA so searches can continue after reopening the screen.
- Added recent task searches in the PWA.
- Improved the PWA to show each question from the top when switching between questions.
- Added all reasoning effort choices for Codex models missing from a custom model catalog.
- Added visibility into custom model catalog usage in the Codex model list.
- Improved screen loading by reusing previously retrieved model lists.
- Improved mode selection and the layout of options on the new task screen.
- Fixed ordinary conversations and errors from other agents being incorrectly treated as account usage limits.
- Fixed exhausted usage allowances disappearing from the usage display.
- Fixed usage details extending beyond the screen on smartphones.
- Added mobile PWA controls to check, download, and install host updates, with confirmation before restarting the host.
- Added /compact conversation context compaction for Antigravity, Cursor, and Qoder.
- Added model list sources, retrieval status, and a reload action.
- Added guidance for agent CLI updates needed for new models, with CLI updates available from Desktop.
- Improved Autorun validation to check that the selected model and reasoning effort are available before saving.
- Fixed viewport height and composer positioning after pairing or reconnecting in the PWA.
- Improved selected-text capture on macOS by waiting for held hotkey modifiers to be released before copying.
- Improved Cockpit Agent so new tasks and autoruns use the models and reasoning levels supported by the connected provider.
- Improved Codex so the latest models available to the selected account can be chosen in New Task, task settings, and autoruns.
- Improved cockpit task create and task run so they reconcile an already-created task after a lost response and avoid creating duplicates.
- Improved project deletion so you can delete all tasks in a project, including running tasks, after confirming the impact.
- Added links to each agent's provider status page when a service outage is detected.
- Improved the in-app Browser so covering overlays show a page snapshot instead of a blank white surface.
- Fixed Antigravity Visual so High reasoning effort is kept after a turn completes instead of snapping back to Low.
- Improved Cockpit Ask so JSON summaries with choices are converted into structured Asks instead of showing raw JSON.
- Fixed Codex usage and rate-limit retrieval for Codex CLI 0.153.
- Improved Antigravity by isolating authentication and quota state per Cockpit profile and adding accounts logout with a safe target preview.
- Improved Antigravity turns to stay active while background commands are pending and track them through the final result.
- Improved Fleets so running nodes remain active during Auto account switches.
- Improved new-task composition by preserving drafts when navigating between screens.
- Added per-attempt Fleet command-gate logs with failed files and complete output available through the CLI and panel.
- Improved full-page browser screenshots for tall pages and pages with inner scrolling containers.
- Added the OpenCode Zen provider and its dedicated API key setting to Cockpit Agent.
- Added clear expired-account states and validation of every Fleet node account before a Run starts.
- Added browser uploads through file choosers created after a click and support for pasting image-only clipboards.
- Fixed false usage-limit detection during Claude Auto account switches and added task re-evaluation after quota recovery.
- Improved task creation acknowledgments when starting tasks through the CLI.
- Improved onboarding and landing-page guidance to clarify that each agent requires its own plan and sign-in.
- Improved Antigravity image attachments by staging them safely inside the workspace so they work in Supervised mode.
- Improved Auto account selection to prioritize the longest usage window and make better use of long-term quota that expires sooner.
- Added Claude Fable 5.1 to the built-in Claude Visual model options.
- Improved Ask windows with resizable layouts and adjustable text sizing.
- Improved startup performance by loading persisted browser tabs only when they are needed.
- Added a Reload Window menu action that preserves the selected task, draft text, and side panel.
- Changed Codex model choices to sort by capability for easier selection.
- Fixed browser actions failing to click controls on wide pages and large card links.
- Added navigation from the Command-K palette to app screens such as Settings and History.
- Changed Remote Access and PWA tab titles to show the connected machine name first.
- Added originating-task data and filtering to Fleet run hooks.
- Added cockpit browser evaluate for running JavaScript in a browser tab's main frame.
- Changed the special AGI Labo skill name to wings and now treats legacy names as regular skills.
- Fixed the app icon appearing as a black square with the macOS dark icon style.
- Added an option to prevent idle sleep while Remote Access is running.
- Fixed hidden browser tabs on macOS covering other apps or appearing as black windows.
- Fixed the app becoming unresponsive when Git diff requests surge on Windows.
- Fixed extra windows opening when AGI Cockpit is launched while already running.
- Improved Cursor Visual context usage and limit display when the runtime does not report token usage.
- Added Discord Ask relay so users can receive and answer Asks from Discord.
- Added Slack Ask relay so users can receive and answer Asks from Slack.
- Added a dedicated Ask notifications settings screen for managing Discord and Slack connections and channels.
- Improved cockpit ask so AskUserQuestion JSON renders as a structured Ask with questions and choices.
- Added local file attachments to answers sent with cockpit ask answer.
- Fixed PWA task deep links so they open correctly at startup and after reconnecting.
- Added detection for Visual Runtime sessions that repeat the same error and an option to resume in a fresh session.
- Fixed /goal in Codex Visual so it starts an actual turn instead of only saving the objective.
- Fixed Antigravity Visual scheduled continuations so they sync back to the original turn after a restart.
- Fixed task status notifications so clicking one brings the target task to the foreground.
- Improved task cleanup so descendant processes detached from their parent are reclaimed after the task stops.
- Improved Visual Runtime history restoration to report omitted conversation turns accurately and explain omitted output in the PWA.
- Added Cockpit Hooks for automatically running commands from task, Ask, Autorun, Fleet, app, and hotkey events.
- Added controls in Settings to enable or disable Cockpit Hooks, inspect recent runs, and delete hooks.
- Improved Fleet runs with a progress bar and gate details that show exit codes and approval results.
- Improved Fleet definition load errors with the cause, target path, Open file action, and Reload action.
- Fixed delayed Visual Runtime errors and composer content so they remain attached to the originating task after switching tasks.
- Fixed empty sessions so they show models for the correct agent and late model-list responses do not revert the current selection.
- Improved PWA Autorun creation and editing so failures keep the dialog open and show the reason and relevant guidance.
- Fixed PWA Ask image panning so it does not switch Remote tabs unintentionally.
- Improved Cockpit CLI errors so they follow the app's selected language.
- Fixed Cockpit CLI internal communication in projects that use type: module.
- Fixed account selection so the account chosen when creating a task is used consistently in the display, saved state, and runtime.
- Improved Auto account selection so it detects exhausted Claude usage credits and switches to another account.
- Fixed task shutdown so descendant processes are cleaned up reliably instead of leaving background work running.
- Fixed Fleet groups so they continue to show the correct running state while gate nodes are being evaluated.
- Improved Fleet gate timeout reports so running, failed, and slow tests can be identified.
- Fixed Antigravity Visual so conversations recover and turns continue after tool errors.
- Improved Antigravity Visual so turns finish after starting background commands and show the process state.
- Added a Command/Ctrl+K task palette for searching and navigating across projects by task name or instruction text.
- Fixed the in-app browser covering settings and dialogs when they are open.
- Fixed unintended reloads caused by reload shortcuts in the production app.
- Fixed app crashes caused by task-status monitoring scanning large temporary directories.
- Fixed the file-preview toolbar so it no longer covers document content and remains usable in narrow side panels.
- Improved Cockpit CLI so it automatically switches to file IPC in restricted environments and diagnoses the active connection path.
- Improved Codex Visual so /goal commands are retained as user messages in conversation history.
- Improved PWA task creation so it recovers when the desktop screen is unresponsive and avoids duplicate tasks on retry.
- Improved desktop and PWA task lists so they can collapse to the initial count after using Show more once.
- Fixed unnecessary authentication, device, and version checks being repeated when switching languages.
- Fixed AGI Cockpit itself being terminated when stopping tasks or ending Fleet gates.
- Improved Cockpit Ask so users can navigate to the task that raised a question.
- Improved Cockpit Ask so files can be attached to answers by drag and drop.
- Improved the in-app browser to explain when no passkey matches Touch ID and show alternative sign-in methods.
- Added a setting to choose whether Follow-ups wait in the queue or steer the current response.
- Improved queued messages so they can be edited before sending.
- Added an option for new-task Autoruns to complete the task they created after the run finishes.
- Added passkey authentication with Touch ID on macOS and Windows Hello in the in-app browser, with clear reasons and alternatives when unavailable.
- Added support in cockpit browser for elements inside Shadow DOM and locators based on accessible names and roles.
- Added project-menu actions to complete all incomplete tasks and delete all completed tasks in a project.
- Improved new-task project selection with current task-list priority and search.
- Improved the new-task screen so it can be used without waiting for agent CLI verification to finish.
- Improved the PWA so the next message can be queued while a task is running.
- Fixed /compact in the PWA so it compacts the conversation context.
- Added the ability to rename Fleet Run titles from the UI and CLI.
- Improved Fleet groups by showing progress for each node.
- Improved Fleet Run recovery after a pull request is merged manually.
- Improved Fleet notifications by suppressing per-node alerts and notifying once when the Run ends.
- Improved Fleet command gates by automatically retrying temporary failures.
- Improved parallel tool-call groups by showing the number of failed calls when only some calls fail.
- Improved updater diagnostics by saving logs and detecting and retrying an unresponsive installation.
- Fixed Autorun management failures caused by a session-refresh race even while signed in.
- Improved Auto account recovery by detecting exhausted Codex workspace credits.
- Fixed Antigravity usage and account details so they come from matching credentials and current observations.
- Fixed context usage reporting for Cursor Visual.
- Fixed unbounded memory growth during high-volume streaming output from Qoder, Cursor, and other ACP agents.
- Changed supported agents to use a unified /goal entry point and goal-status display across desktop and PWA.
- Added paste support and input support for contenteditable editors in the in-app browser.
- Improved recovery for unresponsive in-app browser tabs by recreating them with the same session and URL.
- Improved the cockpit CLI so every command consistently accepts the --json flag.
- Improved long-running visual tasks by rotating older conversation history and making it accessible from desktop, PWA, and CLI.
- Improved message handling during Claude context compaction so queued messages are neither duplicated nor lost.
- Improved save and load efficiency for large task collections while preserving changes made during loading.
- Improved app stability by bounding memory use for growing history and usage logs during long-running sessions.
- Improved diagnostics by saving a report when the app terminates because of an out-of-memory or similar fatal error.
- Fixed a crash in the PWA new-task screen on HTTP connections and older WebKit versions.
- Fixed hidden in-app browser input so it no longer shows a black window or switches macOS Spaces.
- Improved task and Talk Room CLI connection recovery and made unreachable-instance failures explicit.
- Fixed native terminal failures so they stay within the affected task and show the error reason without terminating the whole app.
- Improved History stability by reducing memory usage when loading large histories.
- Improved new-task creation so the new task is selected immediately and its terminal connects reliably.
- Improved validation and user confirmation when opening external links from the in-app browser.
- Improved chat attachment opening by limiting it to safe formats and managed locations.
- Improved isolated agent profiles so they can safely use developer credentials from the macOS Keychain.
- Fixed transient Antigravity tool errors so they stay within the affected action, and improved tool-result presentation.
- Fixed task account selection so it switches correctly between Auto and a fixed account.
- Fixed automatic account switching and task resumption after a usage limit in Auto mode, with a clear reason when recovery is unavailable.
- Fixed ACP tasks remaining in the running state after answering a permission request that arrived after the turn ended.
- Fixed unlimited Qoder organization credits so they appear correctly and count toward available usage.
- Fixed Cursor Grok 4.6 models so they run with the selected reasoning effort.
- Improved Qoder BAD_REQUEST errors with diagnostic details and a terminal resume command.
- Added error details to text copied from chat messages.
- Added multiple Antigravity accounts with automatic account selection and switching for tasks and Autoruns.
- Changed agents added later to a Talk Room to receive mention-only notifications and added per-participant notification indicators.
- Fixed Escape to stop only the task in the focused pane when multiple panes are open.
- Added retries for failed Fleet loop nodes while preserving completed work from the current iteration.
- Added completion indicators for tasks while the task list is frozen in place.
- Changed Cursor Models and Other Models to use separate usage windows for display and account selection.
- Added standard-input and file-input support to cockpit CLI task create, task send, ask, and talk say.
- Changed Talk Room task calls to stop waiting and return a timeout reason when the task does not respond.
- Added the affected model and estimated reset time to Grok Visual usage-limit errors.
- Fixed Fleet failing to recover tasks interrupted by usage limits.
- Added preflight validation for Fleet prompts that reference output fields not guaranteed by the contract.
- Added detailed causes and recovery guidance for Grok Visual and Antigravity Visual startup and runtime failures.
- Added per-participant Talk Room wake modes for all messages, @mentions only, or no notifications.
- Added @mention wake-ups for Talk Room participants who have reduced notifications.
- Changed Fleet command gates to run serially across Runs in the same repository, preventing collisions over resources such as ports.
- Added the captured output tail to failed Fleet command-gate Run logs so the cause is visible.
- Added validated structured output to Fleet command gates and until.run as JSON environment variables and execution directories.
- Changed isolated Fleet nodes to create their worktrees in the repository specified by the node.
- Changed Fleet worktree failures to stop before starting a task in the source checkout.
- Fixed Grok Visual failing the initial prompt when a configured MCP server returns an authentication error.
- Fixed interrupted turns with no new answer redisplaying the previous answer or user input as a fresh completion report.
- Improved PWA task-list loading by prioritizing unfinished and attention-needed tasks and streaming older tasks in chunks.
- Added Talk Room support for resuming completed tasks and recalling participants who have left.
- Changed Talk Room messages to wake participants whose turns have ended and deliver the message content.
- Added Fleet --unset support for clearing saved runtime settings during retry and resume.
- Changed Auto accounts to detect plan-restriction responses and switch to another eligible account.
- Fixed Autorun skipping the next scheduled run while the previously created task was still unfinished.
- Fixed Antigravity Visual failing to send messages with agy 1.1.15 and later.
- Fixed an individual MCP tool failure ending the entire Grok Visual task.
- Fixed long-running Grok Visual tool updates causing conversation history to be lost.
- Fixed Grok Visual tool-only updates increasing the unread state.
- Fixed long visual conversations moving away from the position you were reading.
- Added Fleet {{needs.<id>.task_id}} so later nodes can continue the Cockpit task that ran an earlier node.
- Added Antigravity Visual Mode back.
- Changed Antigravity Visual model and reasoning effort into separate selections.
- Added reasoning effort selection for Cursor visual tasks and Fleet nodes.
- Added reasoning effort selection for Qoder visual tasks and Fleet nodes.
- Changed new Claude and Antigravity tasks to default to Native UI without overwriting a saved Terminal preference.
- Fixed Codex spark models sending unsupported reasoning values so those tasks can start.
- Fixed usage fetching for named Cursor profiles.
- Added Cursor and Antigravity account emails to the usage panel.
- Added Qoder account emails to the usage panel.
- Changed task list and child-task entries to reveal 20 at a time with show more.
- Fixed unanswered task Asks disappearing by showing them again.
- Fixed Autorun skipping a scheduled occurrence when its timer fired early.
- Changed Fleet retry --grant-iterations so the Run iteration limit cannot be exceeded.
- Added Run history to Fleet definition views so each Run can be opened immediately.
- Added back and forward navigation across Fleet definitions, Runs, and nodes viewed in the Fleet panel.
- Improved the Fleet layout so parallel groups and node cards are easier to compare.
- Added cockpit fleet wait for reconnectable waiting without missing Fleet attention events.
- Changed Fleet structured output to use cockpit fleet output for immediate validation, persistence, and separation from report text.
- Changed Fleet node output so it no longer counts as unread user messages.
- Improved in-app browser reliability for dialog actions and input sent to background tabs.
- Improved agent availability stability while the PWA task creation screen is loading.
- Fixed empty task lists flashing before the PWA finishes its initial sync.
- Changed context usage to update immediately after a successful /compact.
- Fixed Japanese punctuation and brackets after chat URLs being included in the link.
- Added a combined all-agent account view to cockpit accounts.
- Added node IDs, settings, and gate or loop conditions to Fleet definition flows.
- Improved active Fleet flows with a compact view focused on run progress.
- Added automatic recovery from transient Fleet node runtime errors, including across app restarts.
- Changed Fleet loop retries to preserve cumulative iteration counts and require an explicit grant after the limit is reached.
- Fixed clicks and other input not being reliably dispatched to background in-app browser tabs.
- Fixed Codex approval cards appearing before newer output or locally sent messages.
- Fixed completed reports appearing twice in Codex visual conversations.
- Improved the task view to show the account selected by Auto.
- Fixed Claude terminal input and scrolling becoming unresponsive on Windows.
- Fixed Claude terminal image attachments being delivered as metadata only on Windows.
- Fixed agents being unable to access attachments with Japanese names on Windows.
- Fixed Cockpit hooks for Antigravity tasks failing during PowerShell expansion on Windows.
- Fixed usage not appearing for authenticated Antigravity accounts on Windows.
- Fixed cockpit task send from Git Bash so Japanese and multiline instructions are delivered correctly on Windows.
- Fixed Fleet command gates to use the login-shell PATH regardless of how the app was launched.
- Fixed the in-app browser remaining visible on another screen after Mission Control or Spaces changes on macOS.
- Added one catch-up execution for an Autorun missed while the app was stopped, when it is within 24 hours.
- Added Autorun outcome, last-task, and skip or failure details across Desktop, PWA, and CLI.
- Added Desktop notifications for missed or failed Autoruns that need attention.
- Changed Autorun to remain active after all macOS windows close and re-arm upcoming runs after wake from sleep.
- Fixed task-list, Overview, and right-side-panel horizontal alignment when their widths change.
- Fixed restored messages in resumed native-UI conversations completing out of order.
- Improved Fleet rows so run state, unread results, and actions are available in one line.
- Added a confirmed action to delete a finished Fleet Run together with all of its related tasks.
- Added titles for Fleet Runs so their purpose is easier to identify in task lists, details, and search.
- Improved Fleet flow presentation by distinguishing agents, gates, and loops by shape and state.
- Improved state saving to reduce load when many tasks update and recover safely from write failures.
- Added registration and selection of custom system prompts for Claude, Codex, Qoder, and Cockpit Agent.
- Added Autorun delivery to an existing task as an alternative to creating a new task.
- Added JSON Schema validation for Fleet node results and structured output references in downstream nodes, gates, and loops.
- Added final-adjudication routing when a bounded Fleet loop reaches its iteration limit.
- Added continued agent sessions and per-iteration follow-up prompts for Fleet loops.
- Added Fleet message nodes that send a prompt to an existing Cockpit task and use its next result downstream.
- Added Cockpit CLI commands to inspect and change allowlisted AGI Cockpit settings.
- Added Cockpit CLI commands to create, sign in to, and remove agent account profiles.
- Added Cockpit CLI commands to configure and control Remote Access.
- Added Cockpit CLI commands to check, download, and install updates.
- Added Cockpit CLI commands to inspect and manage CLI and built-in skill setup.
- Added Cockpit CLI commands to open Settings, Usage, Remote Access, and Autorun screens directly.
- Added Fleet enabled and excluded states so runtime arguments can safely omit nodes.
- Improved usage-limit recovery so stopped tasks can switch to another account and resume instead of failing.
- Added Auto account selection for tasks, Fleets, and Autoruns based on the usage remaining across signed-in accounts.
- Added automatic failover so work continues on another account when the account selected by Auto reaches its usage limit.
- Added reasoning-effort selection for supported Grok visual models.
- Changed new Grok visual tasks to default to Grok 4.6 with xhigh reasoning effort.
- Added Cockpit CLI browser tab summaries with each tab's selection state and page preview.
- Added Cockpit CLI commands to close a selected browser tab or every other tab.
- Added select-box option selection to Cockpit CLI browser controls.
- Added scrolling a specified element into view to Cockpit CLI browser controls.
- Added JavaScript and HTML dialog handling to Cockpit CLI browser controls.
- Added double-click support to Cockpit CLI browser controls.
- Added detection and waiting for newly opened tabs to Cockpit CLI browser controls.
- Fixed Cockpit CLI browser clicks so selecting a child element correctly activates its clickable parent.
- Improved task search to match visible task names, project names, and task IDs of at least four characters.
- Improved the sidebar so task ordering stays stable while a task is hovered.
- Improved rich-text paste handling so text is inserted instead of creating unnecessary file attachments.
- Fixed unread notifications for visual-agent tasks in the PWA.
- Fixed file-type tooltips and Chrome session-import errors appearing in Japanese when the app is set to English.
- Added a setting to choose Enter or Cmd/Ctrl+Enter as the chat send key.
- Added in-place rename for sidebar tasks from the row overflow menu.
- Improved the sidebar and child-task action menus so both surfaces share the same actions.
- Improved Fleet graphs to show unset models as the agent default instead of an unspecified warning.
- Improved Fleet loop containers to show a back edge that makes the whole loop body repeat.
- Improved Fleet loop bodies to show parallel ranks as parallel clusters, matching the top-level graph.
- Removed automatic child-to-parent report-back; use wait commands to collect child results.
- Fixed per-message attachment limits for long-running tasks and made the remaining count clearer.
- Fixed task action menus closing when unrelated panes scroll.
- Fixed sidebar task order so it stays stable while a row menu, rename, or delete confirmation is open.
- Fixed navigation from a Fleet Run back to its definition.
- Fixed terminal tasks sometimes recovering with a stale runtime state.
- Fixed header and window controls so they stay usable while the task list preview is open.
- Fixed Claude visual model options listing the same model more than once.
- Fixed Codex visual tasks so plugin-install confirmation prompts are handled correctly.
- Fixed Grok visual progress not updating when a tool turn arrived late.
- Improved child-task header navigation so header actions no longer collide with the body.
- Added model selection for Grok Build visual tasks created through the CLI or Fleet.
- Fixed Grok Build, Cursor, and Qoder task views showing usage for the default account instead of the selected account profile.
- Improved bounded Fleet loops to display as loop containers with their internal nodes, dependencies, and progress.
- Fixed Fleet completion conditions and command gates failing to evaluate real multi-line reports correctly.
- Fixed memory usage spiking when reopening Codex visual tasks with long histories.
- Fixed parts of a Codex visual task conversation sometimes being missing after reconnecting or restoring history.
- Fixed saved conversation history not being restored for Qoder visual tasks.
- Fixed new visual-task messages sometimes disappearing when they arrived while the PWA was loading history.
- Improved the Codex credit reset control to clearly appear disabled when reset is unavailable.
- Fixed Cursor and Qoder Native UI tool calls sometimes losing results or target files, or displaying internal data instead.
- Fixed Claude Native UI sometimes failing to identify the model behind Claude Code's Default option.
- Fixed saved tasks potentially being lost when task data could not be read at startup or an unexpected empty task list was received.
- Added multiple account profiles for Grok Build, Cursor, and Qoder, with per-task, Autorun, and Fleet selection and switching.
- Added bounded Fleet loops that repeat work until a completion condition is met.
- Fixed Fleet nodes being marked complete before the user answered a request from the agent.
- Improved Fleet validation to catch invalid model, reasoning-effort, and service-tier selections before a run starts.
- Fixed Fleet tasks sometimes running more than once when stop, resume, and retry operations overlapped.
- Fixed native-UI tasks with long histories continually increasing memory usage and sometimes stopping the app.
- Fixed Qoder usage retrieval failing depending on how the app was launched.
- Fixed Claude sign-in and authentication checks ignoring the account selected for the task.
- Fixed retrying after Claude authentication recovery restarting from the beginning without the existing conversation.
- Fixed connection settings being shared between Codex account profiles.
- Fixed authentication screens sometimes appearing behind other screens and becoming unusable.
- Added the new Fleet feature and the Fleet skill for designing and running Fleets.
- Added safe diagnostic data export from the Help menu without prompts or credentials.
- Added Qoder usage reporting alongside the other agents.
- Improved the Autorun list by moving enabled schedules to the top so active schedules are easier to find.
- Fixed Codex model discovery sometimes not completing before task creation, preventing selection of the latest models.
- Fixed long-running Claude context compaction timing out and failing the task.
- Fixed API-key authentication being selected even when Grok native authentication was available.
- Fixed browser actions sometimes failing to verify the response or page change after a click.
- Fixed fallback file-change diffs sometimes being attributed to the wrong task.
- Added a reconnect action for detached or stale App Surface targets.
- Fixed the chat account switcher menu being clipped at the edge of the window and becoming unusable.
- Fixed a crash on Linux.
- Improved the task header and panel-switching controls while the right side panel is open.
- Added Qoder as an agent for tasks and Autoruns.
- Changed the app to a three-column layout for the task list, workspace, and settings, making key information and actions easier to access.
- Added direct editing and automatic saving of text files from the file preview, including files outside the project.
- Changed task creation to show only agents that are available.
- Fixed Grok Build device-code sign-in not completing.
- Fixed unstable text selection and external-link handling in Markdown previews.
- Fixed account, usage, and Git information below the composer breaking on narrow layouts.
- Fixed renewed HTTPS certificates not being applied to the running remote-access server.
- Improved execution safety by preventing video-generation providers from recursively invoking cockpit gen-video.
- Improved site compatibility by having the in-app browser present its native Chromium identity to websites.
- Changed the app platform to Electron 43 with a newer Chromium engine.
- Added App Surface so agents can inspect and operate a connected Android emulator, Android device, or booted iOS Simulator.
- Added isolated Browser Identities for tasks and Autoruns so sign-in state and site data can be kept separate.
- Added Browser Identity creation, editing, data clearing, and removal in Settings and the CLI.
- Added post-click checks to the in-app browser for URLs, visible text, element state, and network responses.
- Fixed controls not being clickable in fullscreen HTML Surfaces.
- Fixed chat auto-follow stopping while new output was still arriving.
- Fixed approval-mode changes not being saved in Settings.
- Fixed Linux update notifications downloading an AppImage for the wrong architecture.
- Fixed app-server settings interfering across multiple Codex Native UI tasks.
- Fixed Grok Build rate limits not being reflected correctly in task failure state and usage.
- Fixed ARM64 AppImages failing to start on Ubuntu 24.
- Fixed the Project Explorer not refreshing automatically after file changes.
- Improved authentication errors with clearer Japanese and English messages.
- Fixed controls not being clickable in the image lightbox.
- Changed Settings with a refreshed design and moved it to the lower-left corner.
- Changed child-task reports to a dedicated view.
- Added per-device names and color icons to identify connected PWA devices.
- Improved in-app browser clicks, typing, scrolling, and key actions by verifying their results.
- Improved the in-app browser side panel by making its resize area easier to grab.
- Improved update checks to retry temporary network errors and show failure reasons with manual update guidance.
- Fixed remote access not recovering after sign-in or membership status changes.
- Fixed Autorun stopping incorrectly after expired authentication or temporary membership verification failures.
- Fixed Cursor Visual task history not restoring correctly after resuming a task.
- Fixed Grok Build Visual task history and active workflows not restoring correctly after resuming a task.
- Added native UI support for running Cursor.
- Added first-run onboarding for sign-in, agent setup, and integration configuration.
- Changed the cockpit skill and cockpit command to be configured automatically for available agents.
- Improved the in-app browser to preserve its state when switching tasks.
- Added cookie and localStorage support when importing sign-in sessions from Chrome.
- Added support for opening links displayed in Ask.
- Added support for opening folder links in Project Explorer.
- Added a button in PWA chat to jump to the latest message.
- Fixed an unnecessary scrollbar appearing while selecting chat text.
- Improved the reliability of copying text from chat details.
- Added per-Autorun model, reasoning effort, service tier, system prompt, approval mode, and account settings that can be reviewed and edited from Desktop, PWA, and CLI.
- Added secure task attachments for text, source code, JSON, CSV, PDFs, audio, video, and office documents.
- Added secure Ask answer attachments for text, source code, JSON, CSV, PDFs, audio, video, and office documents.
- Added quoting of selected chat text into the composer.
- Improved Terminal by integrating it as an independent workspace in the right side panel.
- Added source-task labels and navigation to inter-task messages.
- Changed the history dashboard to be available in guest mode.
- Improved agent-specific settings by unifying them across Desktop, PWA, CLI, and Autorun and identifying unsupported combinations before creation.
- Fixed Codex CLI detection and installation sometimes failing on Windows.
- Improved the update screen with localized release notes and download destinations.
- Added explicit confirmation and a persistent warning for unencrypted local Wi-Fi connections.
- Changed HTTPS startup to fail instead of silently downgrading to HTTP when a certificate is unavailable.
- Added AGI Labo membership verification for Autorun across Desktop, PWA, CLI, API, and scheduled execution.
- Fixed chat text selection being lost during task updates.
- Fixed the HTML Surface switcher leaving the side-panel header.
- Improved child-task listing, creation, and review with a dedicated workspace in the right side panel.
- Added keyboard controls to the in-app browser.
- Added file uploads to the in-app browser.
- Added Cockpit CLI commands to list, answer, and close Asks.
- Fixed file previews interfering with other controls.
- Improved file previews by centering the requested line when opened.
- Improved thinking-level controls by disabling unavailable levels and showing guidance when thinking is off.
- Fixed Claude Visual tasks sometimes failing to start on Windows.
- Fixed Codex Native UI sometimes failing to start on Windows.
- Fixed the production OAuth endpoint being configured incorrectly on Windows.
- Fixed Grok Visual tasks sometimes failing to start when the side panel was closed.
- Fixed running workflows ending when switching to a Grok task.
- Fixed Grok background work appearing to remain active after task completion.
- Improved Codex usage resets to show each reset window's expiration alongside the number of available resets.
- Fixed subagent output appearing in the parent task conversation during a Grok workflow.
- Improved selection by placing Grok Build near Claude and Codex on the new-task and Settings screens.
- Fixed Codex tasks failing to start when the Windows user-folder path contained spaces.
- Added a per-file button for reverting changes from the diff side panel.
- Fixed Grok Visual tasks appearing to remain active after background work completed, and added workflow progress and status details.
- Fixed selection of Codex models that support Fast mode on mobile.
- Fixed the confirmation view losing focus during Japanese text input.
- Fixed text input in the in-app browser and file previews using paths relative to the home directory.
- Improved the task list's child-task layout and the composer's usage and context display.
- Fixed Codex tasks sometimes failing to start on Windows.
- Improved the in-app browser by preserving sessions, tabs, and sign-in state after task completion and app restarts, and by refining page control, waits, full-page screenshots, and Chrome-profile cookie imports.
- Added OAuth sign-in popups to the in-app browser.
- Added automatic Claude Visual model discovery from Claude Code and reflected new models, including Claude Opus 5, in desktop and mobile selectors.
- Fixed the Visual agent model picker failing to open on the new-task screen and fixed usage and mobile composer layouts on narrow screens.
- Changed Claude Code authentication guidance for the current subscription allowances.
- Added Fast mode to Codex tasks and a default speed setting.
- Added imports of site cookies from Chrome on macOS so the in-app browser can reuse sign-in state.
- Added an in-app branch switcher and improved mobile Git-status visibility, including changed-file counts.
- Added task search by task ID.
- Fixed notification task navigation after browser loading and missing in-app browser tabs.
- Improved the chat composer with AI usage, time-to-reset, manual refresh, and account switching.
- Improved model, thinking level, and skill selection operations on desktop and mobile, making it easier to use even on narrow screens.
- Improved Cockpit Agent history recovery so visible conversations remain intact after a temporary retrieval failure.
- Added OpenCode Go provider support to Cockpit Agent.
- Added Grok tasks to Native UI by default and displayed Grok Build authentication status in Settings.
- Fixed issues with terminal display stopping and Japanese input flickering, moving tasks from notifications, and Git status display on mobile.
- Improved update checks and downloads so the latest release remains available after an earlier update was downloaded.
- Improved synchronization between update notifications and Settings, and prevented accidental dismissal actions.
- Fixed past conversations failing to reappear after switching Claude accounts.
- Fixed an issue where Codex tasks could keep returning to running after waiting.
- Improved the usage screen with Grok Build account, plan, sign-in, and retrieval-error details.
- Improved usage loading so a slow service does not delay results from other services.
- Improved the stability of Claude account display and Grok Build usage and credit retrieval.
- Improved work-status visibility by grouping branch, pull-request, and diff information near the composer.
- Added multiple Codex accounts with selection during task creation, execution, and Autorun.
- Added per-account Codex usage and Cursor usage to the usage screen, and improved scannability.
- Improved Grok Build tool readability by showing context usage and background work.
- Added PWA messaging to Grok Build and improved copy reliability.
- Improved retention of sign-in state and remote-access settings across updates.
- Improved in-app browser authentication compatibility for more reliable access to sign-in-required sites.
- Improved task and report notification destinations to prevent incorrect navigation and notification loops.
- Improved Antigravity completion detection, Claude background logs, Codex session detection, and notification handling.
- Fixed the working-folder picker disappearing from the new-task screen in the light theme.
- Added image attachments to confirmation answers and improved large-media transfers in Desktop and the PWA.
- Added PWA control of right-side-panel terminals and automatic resumption of stopped tasks.
- Added OS notifications and Dock or taskbar badges for tasks that need attention.
- Added task-list filtering by agent type.
- Added Codex goal progress to desktop and PWA task headings.
- Added error reasons for process startup failures and fixed terminals failing to start after prolonged macOS use.
- Improved terminal copying with a context menu and failure guidance.
- Improved Claude permission checks and terminal waiting detection to prevent duplicate notifications and status flickering.
- Added working-folder selection from the project name on the new-task screen.
- Changed Linux update notifications to open the download page for reliable installation.
- Fixed bugs related to formula display, external links in display HTML, history loading, Windows file selection, and parent-child task structure.
- Added readable inline and block LaTeX rendering to AI chat.
- Added automatic setup of Creative Studio picture-book tools on Windows and Linux.
- Fixed the Cockpit Ask window changing size as its content changed.
- Added OpenRouter credit balance and usage for Cockpit models to the usage menu.
- Improved formula rendering in AI chat.
- Added project renaming, folder opening, and name reset actions to the task list.
- Added simultaneous review and answering of multiple Cockpit Asks in separate windows.
- Fixed accidental Ask submission with Enter during Japanese input and delivery of Ask answers to Grok Visual.
- Added Talk Room for real-time conversations among multiple agents and users.
- Added Talk Room access from the PWA for reviewing and joining conversations away from Desktop.
- Added Ultra reasoning to Codex Visual Sol and Terra models for automatic delegation of complex work to subagents.
- Fixed Grok Visual Cockpit actions targeting the wrong task when several tasks used the same folder.
- Added Native UI launch and in-app progress and result review for Grok Build.
- Added Grok Build usage to the usage popover.
- Changed the in-app browser to follow Cockpit appearance settings.
- Improved long-running stability and immediate display of Visual history.
- Added shared agent skills to every agent.
- Added persistent agent confirmation requests that can be answered after an app restart to resume the task.
- Improved hierarchy and progress visibility by arranging child tasks vertically.
- Added screenshots of off-screen tabs in the in-app browser.
- Fixed Max reasoning-effort selection for supported Codex Visual models.
- Changed the official website messaging and release history to make information easier to find.
- Added model and reasoning-effort selection to cockpit CLI task creation and exposed both values in task information.
- Added header back and forward buttons for navigating task history.
- Added import of external Claude Code and Codex sessions as Cockpit tasks.
- Added remote Autorun control from the PWA and cockpit CLI.
- Added Claude account profile selection to Autorun.
- Improved PWA task creation and messaging with desktop options including model, effort, approval mode, branch, profile, and skills.
- Added a recent-instructions bar to the PWA with navigation to remote task views.
- Added Fast fallback for GPT-5.6 models in Codex Visual.
- Added Cmd+Backspace line deletion to macOS terminal input.
- Removed unread badges from running tasks.
- Changed Terminal UI and PWA conversations to follow the chat font-size setting.
- Improved Codex usage display and reset handling.
- Changed preview and report links to open in the operating system default browser.
- Improved status display while running /compact or /clear and fixed missing status after switching tasks.
- Added npm installation instructions when installing Codex CLI.
- Fixed Tailscale appearing uninstalled while it was running.
- Fixed global shortcuts failing to persist across restarts and Meta or Windows key recording on Windows and Linux.
- Fixed an issue that could cause a crash when opening a terminal.
- Fixed text overflow in terminal.
- Fixed automatic conversation scrolling interrupting manual scrolling.
- Changed the directory picker to stay in front of the main window.
- Removed background browser windows from Mission Control.
- Added available usage-reset actions to the Codex usage popover.
- Added GPT-5.6 Sol, Terra, and Luna model selection to Codex Visual and improved required CLI-update guidance.
- Fixed the agent Settings tab overflowing its container.
- Added Grok Build task creation, startup, trust confirmation, and terminal control.
- Added multiple Claude account profiles with per-task selection and switching.
- Improved per-account Claude status and usage visibility in the header and usage popover.
- Added Claude account listing and per-task account switching to Cockpit CLI.
- Improved in-app browser address selection and Grok Build terminal scrolling.
- Added skill completion in the chat composer and resolution to the correct skill reference on send.
- Improved in-app browser tab handling with compact display for many tabs and less rendering of hidden tabs.
- Changed environment guidance for agents to stabilize decisions on in-app browser usage and branch creation.
- Added system-prompt presets to Native UI tasks with per-agent defaults in Settings.
- Fixed an issue where the cockpit CLI could not be used from agents running in a sandbox.
- Improved Cockpit CLI guidance for agents and organized cockpit help by command.
- Added in-app release history.
- Added /compact to the Claude Visual slash menu.
- Added a chat minimap and terminal prompt-history rails for navigating long conversations and output.
- Improved Claude and Codex usage display to match OAuth sign-in information.
- Fixed an issue where completion reports were sometimes not sent for self-resumed tasks.
- Fixed occasional app freezes when opening external links, including support links.
- Fixed the app stopping after interaction when launched in the background from a terminal.
- Improved display in light mode.
- Fixed an issue where browser login could not be started on Linux.
- Added login URL copy for environments where the browser cannot be opened automatically.
- Fixed per-task model selection overwriting the default model in Settings.
- Added Linux support with x64 and arm64 AppImage and .deb packages.
- Added cockpit task run and wait for command-driven task creation and completion waiting.
- Improved parent-task delivery of child-task results by persisting task reports.
- Added Native UI and Terminal selection to Creative Studio task creation.
- Improved diagnostics by saving main-process exception logs.
- Improved chat readability in the light theme.
- Added Native UI and Terminal selection to new-task creation.
- Fixed Sonnet 5 missing from the Claude Visual model picker.
- Added Fable 5 weekly usage, remaining runs, and reset time.
- Improved Antigravity usage retrieval.
- Added shortcuts for new-task creation and terminal opening and closing.
- Fixed a blank view after reopening the cockpit browser in the right side panel.
- Improved review of long output by retaining more PWA terminal history.
- Added character-sheet and confirmation-frequency settings to Creative Studio picture-book generation.
- Improved closing image previews in cockpit ask.
- Added bulk expand and collapse actions to the child-task panel.
- Added bulk resumption of child tasks that require resumption.
- Fixed completed Visual Runtime tasks reverting to an earlier state.
- Added adjustable task-list sidebar width.
- Added resizing and collapsing to the child-task panel.
- Added decision-source information to cockpit task waiting reasons for clearer status checks.
- Improved cockpit task conversation output, duplicate Visual Runtime entries, and Visual Runtime history display.
- Improved cockpit task waiting-reason detection.
- Fixed image attachments missing from the first task created in the PWA.
- Improved Antigravity usage and quota display.
- Improved cockpit skill guidance for image and video generation workflows.
- Fixed display and controls in Creative Studio, the model picker, and Cockpit popovers.
- Improved Codex Visual by carrying priority-execution settings to Fast-compatible models.
- Added the cockpit gen-image command for Codex, Antigravity, and Grok Build.
- Added the cockpit gen-video command for Grok Build.
- Added KDP Paperback publishing settings to Picture Book Studio.
- Improved KDP PDF package generation and review-file reliability.
- Improved Picture Book Studio packaging and task display.
- Added a picture-book production mode to Content Studio.
- Improved task creation, review, and PWA display for picture-book production.
- Added the OpenRouter Fusion model preset.
- Improved the readability of image-preview prompts.
- Improved the organization of task action buttons and Content Studio UI.
- Fixed Claude usage display and terminal paste behavior on Windows.
- Fixed Markdown numbered list display.
- Added agent CLI installation-status checks and an in-app installation guide.
- Improved guidance for uninstalled agent CLIs during task creation.
- Added Terminal task creation to Master workspaces.
- Improved right-side-panel controls so they do not shift focus to the task.
- Improved resizing of the right side panel and terminal tabs while displaying iframes and webviews.
- Improved terminal support for copy shortcuts.
- Added Creative Studio for creating and reviewing 3D-model production tasks.
- Added preview, animation playback, and model-download options to the 3D-model review screen.
- Added managed CadQuery and Blender toolchains.
- Added output exports by version and exports that preserve the file structure.
- Added Creative Studio output review to the PWA.
- Added Creative Studio actions to Cockpit CLI.
- Improved detection of Claude Code trust confirmation dialogs.
- Fixed date and time editing for one-time Autoruns.
- Improved help for the cockpit display command.
- Added Claude Fable 5 selection to Claude Visual.
- Added legacy-format support to Codex priority-execution settings.
- Added the cockpit display command.
- Added update checks and installation to Settings.
- Added a collapsible task list.
- Added shortcut settings for the right side panel and task list.
- Improved right-side-panel open and close controls and tooltips.
- Improved Cockpit CLI authentication and side-panel controls on Windows.
- Improved accuracy of restoring Codex sessions.
- Fixed the task diff view showing outdated changes.
- Removed the application menu from the Windows version.
- Added Project Explorer.
- Added file editing and previews to Project Explorer.
- Improved terminal tabs and send actions in the right side panel.
- Added the cockpit side-panel command.
- Added cockpit task goto and current commands.
- Added the in-app browser through cockpit browser.
- Improved the chat message UI.
- Fixed theme rendering in the Ask window.
- Improved waiting-state detection in the Claude terminal.
- Improved light-theme rendering across the app.
- Changed status sorting to place running tasks first.
- Improved navigation to parent tasks.
- Fixed the appearance of screen display and task name editing in light mode.
- Improved running-task display in the PWA.
- Fixed the AI message copy-button position.
- Improved light-theme rendering and terminal theme settings.
- Added creation-date and update-date sorting to the task list.
- Improved task activity display and attention scoring.
- Improved restoration when opening completed Terminal tasks.
- Improved the selected task-group display in the PWA.
- Fixed PowerShell CLI submission and Claude resume status on Windows.
- Improved Windows behavior and packaging.
- Added task and workspace pinning to the Cockpit CLI.
- Added a child-task action menu.
- Improved state retention and diff display in the right side panel.
- Improved handling of completed Terminal tasks.
- Improved default model selection in Claude Visual.
- Added Claude Opus 4.8 support.
- Improved file-change diff display.
- Improved performance.
v4.11.3
patch2026-05-30
Unavailable- Improved Codex Visual task startup speed.
v4.11.2
patch2026-05-30
Unavailable- Added collapsing to project headers.
- Improved file-change updates in the right side panel.
- Changed the new-task menu to close when clicking outside it.
- Fixed bugs.
- Changed HTML Mode to a free-form design.
- Fixed the HTML Mode toolbar in maximized and inline views.
- Fixed sent text remaining in the terminal input field.
- Added status sorting to the task list and PWA search.
- Improved workspace selection for master tasks.
- Added folder creation to the directory picker.
- Improved performance by sharing the Codex Visual app server across tasks.
- Added per-agent defaults for approval mode.
- Improved task status detection in Cursor.
- Added Cursor CLI as an agent.
- Improved the agent selector on the task-creation screen and restored default-agent selection.
- Improved agent icon presentation.
- Fixed chat scroll tracking, Cockpit model-selector stacking, AVIF image attachments, Cursor CLI status restoration, and Master task directory behavior.
- Fixed Visual task completion detection, Claude usage retrieval, the file-change diff panel, and HTML Mode code-block display.
- Changed Gemini agents to Antigravity throughout the app.
- Added descriptions to Cockpit Ask choices.
- Added free-form input alongside multi-select Cockpit Ask questions.
- Improved Cockpit Ask UI and selection controls.
- Fixed Antigravity history and usage dashboards.
- Added Antigravity CLI by integrating the Gemini agent as Antigravity.
- Added multi-select Cockpit Ask questions and Shift+Enter for free-form input.
- Added a task-specific terminal to the right side panel.
- Added Cmd+J and Ctrl+J shortcuts for the terminal and right side panel.
- Added QR-code configuration for PWA connections.
- Improved per-task retention and restoration of right-side-panel state.
- Improved long-press behavior for PWA tasks.
- Improved PWA task display width.
- Improved Autorun badge display by simplifying it.
- Fixed child-task rail scrolling, duplicate Git summaries, and long inline-code wrapping in the PWA.
- Added HTML Mode for rendering assistant responses as HTML and returning button actions to tasks, including in the PWA.
- Added Attention Overview for an overlay summary of tasks that need attention.
- Added Master Agent to guest mode.
- Removed Kanban and Pixel modes.
- Added multiple-question support to Master Agent and Cockpit Ask.
- Added skill picker to PWA terminal mode.
- Added agi-worker to AGI Labo skills.
- Changed Visual Runtime history to paginated loading for large histories.
- Added zoom controls to Settings.
- Improved task display and attention-ranking actions in Overview and Focus modes.
- Improved PWA task layout.
- Improved chat drawing performance.
- Fixed Visual Runtime stalls, completion status, tool approval, history persistence, overlapping headers, and long-URL wrapping.
- Changed the mobile PWA task-creation screen to match Desktop features, including Master Agent switching, file attachments, and skill selection.
- Changed the default display to Focus mode.
- Changed display-mode selection for Terminal and Visual to Settings.
- Removed completed child tasks from the right side panel.
- Improved Claude background-task visualization by excluding regular Bash results, retaining pre-start notifications, and fixing occasional missing completion states.
- Added Claude Visual background Bash task status and output to composer chips and the log panel.
- Added completed tasks to Focus mode attention ranking with bulk deletion.
- Added follow-up messages to completed tasks.
- Improved Focus mode by preserving group expansion and preferring the next task in the same directory after closing a task.
- Added line-number support to file-preview links.
- Fixed the task-name editor position on the task-creation screen.
- Fixed image pasting from the clipboard in Visual mode.
- Added initial Visual-task attachments to chat.
- Improved Visual-mode performance with large diffs or image data.
- Improved the recency score in attention ranking.
- Fixed AskUserQuestion answers failing to send in Claude Visual mode.
- Improved stability of equal-rank task ordering in attention ranking.
- Added task pinning to Focus mode.
- Fixed missing result reports while Codex was idle.
- Fixed duplicate reports during status playback.
- Changed Autorun to run only while AGI Cockpit is open and removed operating-system scheduler registration.
- Changed the task-creation screen design.
- Added Visual and Terminal mode selection to Autorun.
- Fixed Visual-mode Autoruns failing to start or detect completion correctly.
- Fixed Autorun starting through the operating-system scheduler after AGI Cockpit exited.
- Fixed AGI Cockpit moving to the foreground during Autorun execution.
- Added Autorun badges to Focus mode attention ranking.
- Added links to plugin-provided skill chips in chat.
- Improved visual-chat display and resumption performance with large command output.
- Added local voice input to the chat composer when mlx-whisper and ffmpeg are installed.
- Added Anthropic API-key authentication to Claude Visual.
- Improved large-image handling in Codex and Claude image generation and previews.
- Fixed the usage popover remaining black.
- Improved directory display during task creation by abbreviating the home directory as ~.
- Added first-launch onboarding for Cockpit skill and CLI setup.
- Added task unread-state synchronization between the PWA and Desktop.
- Added clickable file paths in inline code and improved command-line detection.
- Improved the PWA task-details layout.
- Removed horizontal task swiping from the PWA and standardized navigation through the task list.
- Fixed PWA task synchronization status failing to update.
- Added image, video, and PDF file previews.
- Added a Finder action to file previews.
- Added collapsible tree previews for JSON files.
- Added a scroll-to-bottom button to chat.
- Improved file-name and path recognition for Unicode characters, spaces, and folder paths.
- Changed the Visual context display.
- Improved attention-ranking display in Focus mode.
- Improved skill-reference display and grouping.
- Fixed Visual agents failing to start with certain command settings.
- Fixed conversation retrieval in Gemini tasks.
- Fixed Codex Visual turn ordering after compaction.
- Fixed slash-command message display.
- Fixed a false “Claude Code Not Found” message during Claude sign-in.
- Changed Claude Code sign-in to a terminal flow for more reliable authentication.
- Fixed Claude Visual showing incomplete task-wide diffs.
- Fixed Claude Code appearing missing when its configured command included arguments.
- Added Claude Code and Codex sign-in management to Settings.
- Added immediate sending of the next message while an agent is responding.
- Added Native UI for Codex, Claude, and Gemini with readable tool execution, file diffs, and approvals.
- Changed Native UI to the default while retaining Terminal in Settings.
- Changed the Settings screen design.
- Added cockpit skill installation to Settings.
- Added parallel work on multiple tasks in separate windows.
- Added file diffs and previews to the right side panel.
- Added chat file-name and path links that open previews.
- Added inline Codex image-generation results with one-click download and Finder actions.
- Added a chat font-size setting.
- Added queued messages while an agent responds, persisted per task across restarts.
- Added parent-child result reporting beyond Master Agent tasks.
- Added Native UI for Codex, Claude, and Gemini to the mobile PWA.
- Added authentication to Tailscale access.
- Added Japanese and English localization.
- Fixed Claude Agent resume failures.
- Added left and right arrow navigation back to task cards.
- Improved attention ranking by grouping child tasks under the Master task.
- Added AGI Labo account sign-in through OAuth SSO.
- Changed the first launch after the update to require a one-time sign-in.
- Fixed unread badges reappearing on tasks awaiting confirmation.
- Added attention ranking with unread tracking and Focus mode.
- Added a Focus tab and attention panel to the PWA.
- Added origin validation to the PWA API and WebSocket.
- Fixed OpenCode binary support on Windows.
- Added inline Cockpit Agent conversations to the terminal view.
- Fixed Master task detection and the isYou flag in the task list.
- Added cockpit as an official task, Autorun, and default-agent option for Master Agent.
- Changed Master Agent help and operations guidance to use consistent cockpit agent options.
- Added Cockpit Agent to the same board as Claude Code, Codex, and Gemini.
- Added Cockpit task creation, conversations, and permission responses to the PWA.
- Improved Cockpit session resumption, task-specific settings, master linkage, and completion processing.
- Changed the Settings panel to a two-column grid.
- Added a shortcut settings dialog for task exit, navigation, and completion keys.
- Changed the AI agent selector to a dropdown.
- Improved terminal color-theme settings with a three-row inline layout.
- Added full-screen media previews to the Ask window.
- Added a confirmation dialog for closing with Cmd+W.
- Fixed the server lifecycle on macOS.
- Added hover previews to task cards.
- Added links to Ask URLs in the PWA.
- Fixed the header drag area.
- Fixed Master Agent display in Autorun.
- Added image, audio, and video attachments to cockpit ask.
- Added server-side caching to Tailscale device discovery.
- Changed the in-app API endpoint to api.agi-labo.com.
- Changed release distribution to releases.agi-labo.com.
- Added file and folder attachments to Terminal.
- Improved the Terminal details header layout.
- Changed Pixel Office rooms to group by repository.
- Changed Cockpit tasks to group by repository.
- Added Cockpit server switching to the PWA.
- Added a device-discovery cache to the remote-access panel.
- Added automatic Tailscale network-device discovery to the remote-access panel.
- Fixed terminal scroll jumps.
- Changed the update notification position to the left of the toolbar icon.
- Fixed status handling for resumed tasks.
- Improved PWA task details with larger buttons and automatic text-area resizing.
- Added an exit confirmation dialog to prevent accidental app closure.
- Fixed PWA WebSocket reconnection recovery.
- Added an Ask tab to the PWA and improved mobile responses to pending confirmations.
- Improved Ask context by showing the task title and starting-directory name.
- Improved Cockpit CLI runtime discovery so Ask and Autorun commands work outside Master Agent shells.
- Changed PWA Autorun tasks to use the regular task list and match the Desktop Kanban layout.
- Fixed the macOS menu bar becoming unresponsive while the Ask dialog was open.
- Changed Autorun tasks from a dedicated right-side list to the main Kanban board.
- Added Autorun badges to running and completed tasks.
- Fixed Claude usage retrieval failing in installed builds because of PATH or the initial trust prompt.
- Added an on-demand usage popover for Claude, Codex, and Gemini.
- Improved shell-based agent startup by passing prompts safely as arguments.
- Improved Claude usage retrieval and fixed packaged apps failing immediately after startup.
- Fixed Claude usage retrieval ending immediately in installed builds.
- Changed the usage display to a toolbar popover.
- Changed usage retrieval to on-demand and removed periodic polling.
- Added detailed usage display for Claude, Codex, and Gemini.
- Added task and autorun subcommands to Cockpit CLI.
- Added task creation, listing, details, completion, deletion, renaming, and remote operations to cockpit task.
- Added creation, updating, deletion, toggling, and immediate execution to cockpit autorun.
- Added real-time user confirmations from agents through cockpit ask.
- Added choices and free-form input to the always-on-top floating window.
- Added dynamic window sizing based on content.
- Fixed names not carrying over to tasks created by Autorun.
- Fixed orphan Pixel Office characters not being assigned to rooms.
- Added remote task control through Tailscale device discovery.
- Added persistent device aliases.
- Improved worktree cleanup.
- Fixed zombie PTY processes after task completion.
- Changed the SyncServer body-size limit.
- Fixed new-task click handling.
- Changed Pixel Office button placement.
- Fixed minor bugs.
- Fixed minor bugs.
- Added Pixel Office, an interactive office view that represents tasks as characters.
- Added initial task status based on agent type.
- Fixed a 32-bit setTimeout overflow that duplicated monthly or longer cron runs.
- Changed idle timeouts by agent type.
- Improved Pixel Office usability, performance, and layout density.
- Fixed duplicate Autorun triggers after app restarts or operating-system resume.
- Added an execution lock to prevent infinite Autorun spawning.
- Improved Autorun enabled checks, lock safety, and one-time cleanup.
- Fixed an execution-lock leak after a manual kill.
- Added a CLI installation button to the Settings panel.
- Improved and fixed minor issues.
- Fixed startup freezes with many sessions by making getGitInfo asynchronous and removing countPathLines.
- Added per-agent terminal color-theme settings.
- Improved and fixed minor issues.
- Changed internal structure and version organization.
- Added CRUD prompt templates for Master Agent.
- Added template-variable chips and required indicators.
- Added a task-server watchdog and automatic restart flow.
- Fixed weekday plist generation for macOS launchd cron schedules.
- Removed task-server token authentication and standardized port 19280.
- Changed the settings version check to be non-blocking.
- Improved terminal snapshots by optimizing the dirty flag.
- Added an independent task-server port for development mode.
- Fixed the terminal loading spinner continuing with empty history or a disconnected WebSocket.
- Added a loading spinner while the PWA retrieves terminal data.
- Added --directory and --name filters to cockpit task list.
- Added task-delegation guidance to Master Agent AGENTS.md.
- Fixed Ctrl+V paste support and nested Claude Code detection.
- Added Windows cloud-signing hooks and signature preflight and verification scripts.
- Fixed Windows terminal PATH resolution and the app icon.
- Improved Windows support comprehensively.
- Fixed Windows PATH handling by using EncodedCommand to avoid cmd.exe quoting issues.
- Improved tool detection by reading the Windows PATH from the registry.
- Added a guard against duplicate Autorun execution.
- Added Master Agent support to Autorun.
- Added an isYou marker to the task list.
- Changed Windows Terminal to PowerShell.
- Added npm, pnpm, and cargo to the Windows Terminal PATH.
- Added safeStorage for ARM64 Windows.
- Added a skills panel and folder skill badges.
- Added a skill-details dialog with SKILL.md display and deletion.
- Fixed Windows agent-version detection.
- Removed the Python 3 dependency from the Master Agent script and added a Windows .cmd wrapper.
- Added Windows SystemSounds support for notification sounds.
- Improved Windows support for __dirname paths, status watchers, and hook integration.
- Changed Master Agent commands to use relative paths.
- Added a sound when status changes to waiting_confirmation.
- Added an on/off notification-volume setting.
- Added a child-task panel to the Master Agent terminal dialog.
- Removed temporary directories from the workspace list.
- Fixed blank terminals for completed tasks.
- Fixed duplicate report-back notifications to the parent Master Agent.
- Changed image delivery to child tasks to reset lastNotifiedStatus.
- Added simultaneous management of multiple Master Agents on the Kanban board.
- Added Gemini CLI as an agent type.
- Added guest mode.
- Added report-back notifications for Master Agent.
- Added abbreviated task IDs.
- Added Master configuration management to Cockpit CLI.
- Added persistent cross-session notes through memory.md.
- Added per-agent Kanban card label colors.
- Improved task creation with directory selection and description input.
- Changed Master, task, and Autorun navigation to separate sections.
- Added automatic default-agent selection during task creation.