What changed, and what is still wrong.

Version 0.4.1 (Breezy Wizard)

Beta 4 Point Release · October 2026

A small patch release. The Mods marketplace now opens at its own address, and the installer is rebuilt and re-notarized. Nothing else about the app changes.

• The Mods marketplace has its own home: opening Mods from OOMU now goes to mods.oomu.ai instead of the old site root. Same listings, same install flow, with its own valid certificate.
• Everything else is unchanged: the same features as 0.4.0, and the same files and settings on your Mac. Update in-app, or download the installer again and replace the app.

Version 0.4.0 (Breezy Wizard)

Beta 4 · October 2026

This release brings Mochi to the desktop, adds Standing Memory, connects Slack and Discord for phone approvals, and rebuilds how OOMU starts so an interrupted install or update repairs itself on the next open instead of getting stuck.

• Mochi on your desktop: Mochi watches the standing interests you turn on, does the first pass of the work, and stages anything it wants to change for your approval. Answers stream in as they are written, and the review surface collects everything waiting for you in one place.
• Slack and Discord, bound to this Mac: Connect once with an owner account and you can check status, list staged items, and approve, deny, or snooze them from your phone. Messages never bypass the same approval screen the desktop uses, and the seven supported commands are spelled out in the docs.
• Standing Memory: While your Mac idles overnight and on power, OOMU distills finished conversations into a local memory. Every saved fact keeps its exact source quote, and you can edit, trace, or forget any of it from the Profile inspector. Memory banks are excluded from cloud-model requests.
• A startup that repairs itself: If an install, update, or launch is interrupted midway, the next open recovers the database from a retained safety copy. Startup problems now say what they actually are (a missing key, a wrong key, a busy lock, a damaged file) instead of a generic error.
• Two changes from Beta 3 worth knowing: scheduled workflows now send file writes to Ready for review instead of writing directly, and Quick Launcher moved to Command-Control-Shift-Space so it never collides with a Beta 3 copy.
• Upgrading from Beta 3: one-way and lossless. Your Beta 3 data comes across with a backup kept beside it, model files are reused without re-downloading, and provider keys are carried over.

Version 0.3.10 (Manic Kingpin)

Beta 3 Point Release · October 2026

This update enhances grounded document analysis with in-prompt directory cascading, ensures Automatic agent conversations reliably stay on Apple Silicon, stabilizes user-named Office deliverable generation, and preserves attached documents across multi-turn dialogue until compaction.

• Autonomous In-Prompt Directory Grounding: Specifying folder paths directly in chat (e.g. Search ~/Documents/...) now autonomously lists and reads proposal documents on-device, enabling immediate, grounded multi-document analysis and comparison without requiring manual folder-picker dialogs.
• Deterministic Automatic Agent Routing: Starting a new conversation with an agent configured to Automatic now reliably preserves on-device Apple Silicon execution rather than inappropriately falling back to cloud defaults.
• Robust Office Deliverable Generation: Eliminated brittle planning assertions when generating user-named spreadsheets and documents, allowing models to compile files like Q4_Freight_Rate_Comparison.xlsx directly to your Desktop, Documents, or Downloads folders.
• Multi-Turn Session Attachment Retention: Uploaded spreadsheets, PDFs, and rate cards now remain active across all dialogue turns in a session until context compaction or explicit removal, eliminating amnesia during follow-up questions.

Version 0.3.9 (Manic Kingpin)

Beta 3 Point Release · October 2026

This update introduces dynamic multimodal vision projector auto-discovery and pairing for custom local models, enables seamless on-device image analysis, and provides zero-friction automatic delegation to the built-in vision engine when a text-only local model is selected.

• Dynamic Sibling Vision Projector Pairing: Registered and sideloaded local models now automatically discover and bind compatible sibling multimodal projector files (such as *-mmproj*.gguf) in their directory. Vision models like Gemma 4 E4B now run multimodal image analysis directly on Apple Silicon.
• Autonomous On-Device Vision Delegation: If you attach an image while an active local model is text-only, OOMU automatically delegates the image analysis to the built-in Gemma 4 E2B vision engine on this Mac rather than failing the turn. Responses display clear on-device attribution.
• Improved Model Capability Error Guidance: Eliminated ambiguous verification error banners when unsupported models encounter image inputs, replacing them with clear guidance on selecting a compatible local or cloud vision model.

Version 0.3.8 (Manic Kingpin)

Beta 3 Point Release · October 2026

This update introduces an ergonomic overhaul to the Settings > Profile Memory management surface, replacing raw scrolling text with compact, collapsible memory cards, live keyword search, scope filtering, and bounded viewport scrolling.

• Compact Collapsible Memory Cards: Saved memories are now organized into clean, structured cards with three-line preview clamping. Multi-paragraph notes and structured logs no longer flood the screen; an inline toggle expands content smoothly into an inset, bounded reader.
• Instant Keyword Search & Scope Filtering: Find memories instantly with live text filtering and switch seamlessly across All, Everywhere (Global), and Projects using segmented filter tabs with live memory counts.
• Bounded Viewport Ergonomics: The memory list is now bounded with internal scrolling, ensuring the Profile personalization form and memory management stay perfectly balanced and accessible on all Mac display sizes.

Version 0.3.7 (Manic Kingpin)

Beta 3 Point Release · October 2026

This update introduces autonomous tool loop prevention to stop repetitive tool invocations gracefully, adds native Apple Silicon hardware profiling, and restores exclusive tool-calling authority to the generative model, ensuring conversational discourse and negative instructions never trigger unintended actions.

• Autonomous Tool Loop Prevention: Repeated or identical tool invocations are automatically suppressed after two iterations. When autonomous step limits are reached, OOMU synthesizes a helpful response from collected receipts rather than canceling the turn or rescinding the chat bubble.
• Native Apple Silicon Hardware Profiling: Added native macOS hardware profiling via system.hardware. You can now ask OOMU for this Mac's specifications to view its chip model, CPU core layout (performance and efficiency cores), unified memory, and macOS version.
• Content Model Tool Authority: Restored exclusive tool-calling authority to the generative content model, eliminating speculative fast-path bypasses. Conversational feedback and negative instructions (such as "do not check the battery") are recognized naturally without triggering unwanted tool calls.

Version 0.3.6 (Manic Kingpin)

Beta 3 Point Release · October 2026

This update ensures local model integrity remains rock-solid against routine APFS filesystem metadata fluctuations, eliminates route execution desynchronization between the UI tuning panel and backend runtime, and expands the cloud model catalog with Google Gemini 3.1 Pro, Anthropic Claude 5.5 and Mythos 5, and OpenAI GPT-6.1 Sol.

• APFS Filesystem Metadata Tolerance: Local model verification now accounts for routine macOS APFS nanosecond timestamp jitter and background indexing, preventing installed models from spuriously downgrading from Ready to Changed status.
• Deterministic Local Route Execution: Fixed a desynchronization between the chat tuning panel and backend execution, ensuring that chats faithfully execute the exact model selected in the tuning panel.
• Expanded Frontier Model Catalog: Updated the cloud provider catalog to version 1.0.5, bringing native support for Google Gemini 3.1 Pro, Anthropic Claude 5.5 (Sonnet and Opus), Anthropic Claude Mythos 5, and OpenAI GPT-6.1 Sol, while retiring obsolete entries.

Version 0.3.5 (Manic Kingpin)

Beta 3 Point Release · October 2026

This update ensures multi-turn conversation context is fully preserved when using the @ste100 clarity tag, stabilizes cloud model inference parsing, and guarantees that your chosen route remains locked to each session across turns and reloads.

• Multi-Turn Context with @ste100: Asking follow-up questions with the @ste100 clarity tag now retains the complete conversation history. The tag functions as a pure conversational style modifier rather than an isolated tool call.
• Relaxed Cloud Inference Parsing: Cloud models paired with @ste100 now complete turns reliably without response errors, gracefully handling follow-up suggestions and conversational prose.
• Session-Scoped Route Persistence: When you select "On this Mac" or "Cloud", that choice is permanently saved to the active session and will never flip back to default routes upon turn completion. Starting a new chat cleanly presents the configured default mode.

Version 0.3.4 (Manic Kingpin)

Beta 3 Point Release · October 2026

This update introduces rapid sequential chat deletion with independent LIFO recovery windows and deploys the native @ste100 Simplified Technical English clarity tag for ultra-concise, token-efficient local generation.

• Rapid Sequential Chat Deletion: You can now delete multiple chat sessions in quick succession from the sidebar without experiencing a 10-second button lockout.
• LIFO Recovery Stack: Every deleted chat preserves its full, independent 10-second Undo recovery window in local storage. Clicking Undo seamlessly unwinds the deletion stack in reverse order.
• Native @ste100 Clarity Tag: Added the native @ste100 explicit tag chip to the message composer. Applying this tag constrains on-device models to the European aerospace ASD-STE100 Simplified Technical English standard, eliminating conversational filler and cutting generation token volume by 40% to 60%.

Version 0.3.3 (Manic Kingpin)

Beta 3 Point Release · October 2026

This update delivers critical stability and persistence improvements, ensuring permanent chat deletions across updates and restarts, restoring the ergonomic Cmd+Shift+Space Quick Launcher shortcut, and eliminating main-thread dismissal friction.

• Chat Deletion Persistence & Right to Erasure: Staged chat deletions are now flushed and committed immediately across application updates and unexpected restarts, guaranteeing that deleted conversations are permanently purged from local storage and never resurrect.
• Ergonomic Quick Launcher Shortcut: Restored the clean, single-handed Cmd+Shift+Space global shortcut from Beta 2, eliminating the four-finger keyboard claw.
• Quick Launcher Responsiveness: Re-engineered background shortcut reconciliation onto non-blocking asynchronous routines, ensuring instant, seamless dismissal without interface freezing.

Version 0.3.2 (Manic Kingpin)

Beta 3 Point Release · September 2026

This update introduces native macOS continuous spellcheck across all text entry surfaces, turn-level message copy actions, one-click Markdown session export, and a refined zero-scroll onboarding layout designed for compact displays.

• Native macOS continuous spellcheck: text inputs in the main composer and Quick Launcher now support continuous spellchecking, native red squiggly underlines, and contextual spelling corrections.
• Turn-level message copy actions: added dedicated, accessible copy buttons directly beneath prompts and assistant responses with instant visual confirmation.
• One-click Markdown session export: export complete chat histories into clean, portable Markdown files with YAML frontmatter for Obsidian, Bear, Logseq, or Apple Notes via native macOS save dialogs.
• Refined zero-scroll onboarding layout: restructured the onboarding setup screen into an adaptive two-column layout during model downloads. Fits cleanly within 640px to 800px window heights on 13-inch MacBooks and scaled displays with zero clipping and zero scrollbars.
• Quick Launcher keyboard input recovery: resolved an issue where keyboard input could be blocked upon re-opening the Quick Launcher floating window.
• Starter prompt retrieval binding: connected chat starter chips directly to native vector search and project folder summaries for faster contextual bootstrapping.

Version 0.3.1 (Manic Kingpin)

Beta 3 Point Release · September 2026

This update introduces seamless automated migration from Beta 2, real-time local model synchronization across settings and chat, agent model inheritance, and Google Gemini 3.8 Flash catalog defaults.

• Seamless Beta 2 data migration: automatically detects existing Beta 2 installations and imports your custom agents, chat histories, workflows, and projects into Beta 3 on first launch, while keeping your original database safely preserved as an encrypted backup.
• Automatic keychain credential transfer: securely migrates your saved API keys and provider connections from Beta 2 into Beta 3's secure macOS Keychain container without requiring re-entry.
• Live local model synchronization: adding, changing, or setting default local models in Settings now updates the Chat surface model selector immediately without requiring an application restart.
• Chat session agent model inheritance: selecting a custom agent in chat now automatically binds the conversation to that agent's configured model and route preferences.
• Google Gemini 3.8 Flash default: updated Google provider defaults to Gemini 3.8 Flash for faster inference and higher reasoning limits.
• Resolved model switching error banner: fixed an issue where switching from cloud models to local models could trigger a false-positive configuration alert.

Version 0.3.0 (Manic Kingpin)

Beta 3 Initial Build · September 2026

Welcome to OOMU Beta 3. This release introduces our Cooperative Dual-Model Engine, pairing an ultra-fast on-device neural sensor with your chosen language models for sub-second responses and zero-hallucination document workflows.

• Cooperative Dual-Model Engine: introduced our specialized System 1 on-device neural router (Bosun v3.1). Running entirely locally on Apple Silicon unified memory in 22 milliseconds, it classifies user intent instantly and eliminates the seconds-long preflight delays of monolithic models.
• Direct Plan Bypass: conversational questions, coding queries, and reasoning tasks now bypass multi-step action planning entirely, streaming answers in under half a second (493ms time-to-first-token) while preserving deep planning strictly for tasks that require real tools.
• Zero-hallucination attachment processing: guaranteed attachment precedence prevents models from hallucinating disk-search tool calls when analyzing PDFs, spreadsheets, or images you drop into the chat. What you attach is what the model works with immediately.
• Native macOS system automation: added direct native execution for Apple Reminders and hardware battery status, executing in under 135 milliseconds without launching external terminal subshells.
• Over-the-air neural model supply chain: System 1 routing models now update independently of application binaries, checking our canonical model repository automatically every 24 hours with user-consented background downloads and atomic verification.
• Automatic disk space reclamation: safely detects and prunes obsolete Beta 2 router weights upon upgrading, instantly recovering up to 1.2 GB of SSD storage while preserving your private chat history and local model library.

Version 0.2.19 (Apocalyptic Captain)

Beta 2 Final Build · September 2026

• Enterprise governance and OOMU Control support: added client-side connectivity for managed macOS deployments. Connect your workstation to your organization's self-hosted OOMU Control plane to automatically receive verified enterprise security guardrails and policies.
• Native Apple MDM configuration discovery: automatically detects management profiles and enrollment settings deployed via Apple Device Management, Jamf, or enterprise postinstall scripts without requiring manual setup.
• Cryptographic policy verification: enforces tamper-evident Ed25519 signature checks on all incoming organizational policy updates, ensuring complete cryptographic trust.
• Instant Air-Gap lockdown: immediate hardware network isolation when commanded by organizational security policy, severing external connections and dropping credentials while keeping local models operational.
• Independent personal installations: workstations without enterprise MDM profiles remain completely private, with zero background fleet synchronization and full user control.

Version 0.2.18 (Apocalyptic Captain)

Beta 2 · September 2026

• General-purpose spreadsheet creation: extract data, comparisons, project timelines, and structured facts from any attached document directly into a clean, formatted native Microsoft Excel (.xlsx) workbook, complete with auto-fitted column widths, styled header rows, and clear gridlines.
• Resilient Google Gemini structured responses: improved transport recovery when creating documents and complex deliverables with Google Gemini models. If a provider rejects rigid formatting rules, OOMU automatically recovers and delivers your requested spreadsheets and data without interruptions.
• Refined macOS menu bar icon: replaced the temporary status item with pixel-perfect, high-DPI native template assets derived directly from the official OOMU vector emblem, ensuring crisp rendering and automatic color inversion across macOS Light Mode, Dark Mode, and active selection states.

Version 0.2.17 (Apocalyptic Captain)

Beta 2 · September 2026

• Date-ordered imported memory: imported notes and historical discussion records now preserve their original calendar dates and chronological sequence across your timeline, preventing recent updates from being obscured by older notes.
• Fair-share memory recall: memory projection windows now allocate capacity evenly across all matching records, guaranteeing that an oversized note cannot crowd out the rest of your conversation history.
• Resilient model recovery: if a connected model produces an empty answer, pauses unexpectedly, or enters an internal reasoning meta-loop, OOMU detects the stall immediately and automatically triggers a clean conversational recovery before showing you the response.
• Fault-tolerant document generation: if a cloud provider experiences an unexpected connection issue while creating an Excel spreadsheet or document deliverable, OOMU falls back gracefully to synthesising the full factual analysis in conversational text, preventing silent turn cancellations and giving you the data immediately.

Version 0.2.16 (Apocalyptic Captain)

Beta 2 · September 2026

• Durable conversation compaction: context summaries created during long chats are now saved securely in your Mac's local storage. Returning to previous chats or restarting the app reuses existing summaries instantly, eliminating redundant processing and keeping your work fast and responsive.
• Single language authority across your Mac: selecting a non-English language now persists cleanly across all screens, native macOS notifications, and generated documents. English wording no longer leaks into non-English profiles.
• Continuous workspace navigation: the top navigation bar is now permanently visible on the Chat screen, letting you switch between Chat, Agents, Workflows, and Activity with a single click. The Command Palette (Cmd+K) now indexes quick actions and all primary destinations.
• Clear and scoped Air-Gap Mode: Settings now describe connection boundaries with total transparency, clearly explaining that cloud models, external web searches, and remote connections are paused while all processing remains on this Mac.
• Helpful control guidance: buttons that require prior setup (such as saving custom agents or connecting external services) now display clear explanations beneath them describing exactly what is needed before they can be clicked.

Version 0.2.15 (Apocalyptic Captain)

Beta 2 · September 2026

• Native PDFKit fallback: opening and analyzing complex, non-standard, or unusual PDF documents now extracts text reliably by falling back to native macOS Apple PDFKit when standard document parsers encounter formatting issues. Image-only or scanned documents without a selectable text layer now clearly explain that no text is present rather than producing an internal system error.
• Fast Apple Mail automation: searching your local Apple Mail inbox now retrieves header summaries directly, dropping search and enumeration times from over thirty seconds to under half a second without freezing or timing out.
• Actionable macOS permission alerts: when an action needs access to your Calendar or Mail, OOMU now clearly names the specific permission required and provides an immediate one-click button to open that exact System Settings pane or configure it in Connections.
• Composer focus and error clearing: the chat composer automatically reclaims keyboard focus the moment an answer finishes generating, ensuring your follow-up keystrokes are never lost. The file attachment error banner now clears instantly as soon as you type, dismisses after six seconds, and includes a direct close button.
• Vision routing output preservation: image analysis answers and generated text are now permanently preserved on screen even if an optional follow-up suggestion check experiences a connection issue.

Version 0.2.14 (Apocalyptic Captain)

Beta 2 · September 2026

• Local context compaction: conversations that exceed local model memory limits now automatically summarize earlier discussion history on-device, keeping your work running privately on Apple Silicon without dumping your questions to the cloud.
• Multi-monitor menu bar polish: clicking outside the raven menu bar tray menu on any display now dismisses the menu immediately. Clicking the raven icon a second time also cleanly closes the menu.
• True turn-by-turn routing: removed sticky cloud session persistence. Every single prompt in Auto mode is evaluated independently by our on-device router, ensuring follow-up questions and simple calculations stay on your local Mac.
• Agent cloud provider precedence: custom assistants configured with specific cloud providers (such as DeepSeek Direct) now take proper precedence over global profile defaults in Auto mode.
• Updated model catalog: updated provider definitions to version 1.0.2 to ensure reliable, high-speed compatibility with the latest frontier API endpoints.

Version 0.2.13 (Apocalyptic Captain)

Beta 2 · September 2026

• Enterprise fleet controls: added support for organization management, allowing corporate teams to dynamically update workplace safety rules, connect through enterprise web proxies, and use custom security certificates without restarting.
• Cryptographic audit security: policy boundaries and security events now generate tamper-evident audit receipts signed directly by your Mac's Apple Secure Enclave hardware, ensuring organizational compliance with complete data privacy.
• Zero impact on individual users: personal and unmanaged installations continue to run with 100 percent freedom, zero background telemetry, and zero overhead. Prompts, documents, and conversations never leave your Mac.
• Browser Bridge efficiency: enhanced the background connection handler between your browser AI sessions and local OOMU tools, ensuring completely silent operation when idle.

Version 0.2.12 (Apocalyptic Captain)

Beta 2 · September 2026

• Cloud reasoning capacity and stability: resolved an issue where asking complex questions using cloud models could abort before sending the request due to an overly strict local hardware limit. Cloud models now use their full context capacity, allowing long conversations and active browser connections to run smoothly.
• Clearer error reporting: when a request experiences a setup or connection issue before reaching a model, OOMU now accurately reports the routing error rather than displaying an incorrect execution failure notice.
• Preserved local memory safeguards: on-device AI models continue to run with strict, verified memory bounds tailored directly to your Mac hardware.

Version 0.2.11 (Apocalyptic Captain)

Beta 2 · September 2026

• Update transparency: when an update is ready to install, OOMU now displays the top changes directly in the update prompt so you can see what is new before restarting, along with a direct link to these full release notes.
• Grok browser connection: added native host support for Grok in Browser Bridge. You can now connect active Grok browser sessions alongside ChatGPT, Claude, and Gemini to use frontier models with your local OOMU workspace.
• Browser extensions in progress: companion browser extensions for Chrome and Brave are now in active testing, providing private, token-efficient communication between your web browser AI sessions and your local Mac without API keys.

Version 0.2.10 (Apocalyptic Captain)

Beta 2 · September 2026

• Zero analytics and tracking: completely removed background analytics and diagnostic tracking. OOMU transmits zero diagnostic data, clears existing anonymous tokens from previous builds on startup, and keeps your work entirely on your Mac.
• Workplace policy controls: organizations can now define clear boundaries for OOMU across work computers. If your team requires strict offline use, limits which cloud providers can be reached, or restricts which folders the app can touch, administrators can place an immutable policy file on the machine to enforce those rules automatically.
• Stricter folder boundaries: file tools now enforce folder permissions and company policy rules before touching disk. This keeps sensitive system directories protected and prevents unauthorized file changes before an action runs.
• Browser Bridge mod: introduced the Browser Bridge capability, which lets web browser sessions (such as ChatGPT, Claude, and Gemini) connect directly with your local OOMU workspace on your Mac. The companion browser extensions for Safari, Chrome, and Firefox will become available over the next week.
• Connections screen polish: cleaned up the Connections page with balanced card heights, clearer setup guidance, and full translations across all 12 supported languages.

Version 0.2.9 (Apocalyptic Captain)

Beta 2 · September 2026

• European sovereign cloud integration: added native Mistral AI Direct support, giving you direct access to European-hosted frontier models including Mistral Large, Codestral, and Pixtral directly from your Mac.
• Instant chat switching: returning to previously opened conversations now reuses cached on-device model memory, cutting evaluation time in half and eliminating sluggish restarts when multitasking across chats.
• Multiple custom AI endpoints: you can now connect, latency-test, and manage multiple independent OpenAI-compatible custom servers directly in Connections, complete with dedicated status badges and one-click removal.
• Sovereign hybrid fact retrieval: on-device models now answer civics, historical, and conceptual questions directly from memory while automatically invoking verified public citations when you ask for exact verbatim quotations of statutes and formal texts.
• Unified connections and credentials: removed redundant API key prompts from Settings. All cloud keys, local models, and service connections are now managed in one unified place under Connections.
• Dedicated GDPR and EU AI Act pane: added a transparent regulatory disclosure surface in Settings detailing on-device router specifications, machine-readable export metadata, and complete data subject erasure controls.

Version 0.2.8 (Apocalyptic Captain)

Beta 2 · September 2026

• Gmail inbox intelligence and automatic detail fetching: asking OOMU about unread email or recent message subjects now automatically fetches full subject headers, sender names, dates, and snippets in a single turn, eliminating raw ID outputs and multi-step delays.
• Universal direct cloud provider compatibility: connecting directly to DeepSeek, Zhipu AI / Z.AI, Moonshot AI / Kimi, and custom third-party endpoints now adapts structured responses automatically to native JSON mode, eliminating HTTP 400 format errors on prompt submission.
• Enhanced clipboard privacy protection: Quick Launcher and composer pasteboard inspection now actively checks for password manager markers (including 1Password, Bitwarden, and Keychain ConcealedType tags), ensuring passwords and sensitive credentials never leak into prompt context or cloud models.
• Hardened filesystem boundary defense: file reading and writing tools now enforce strict confinement rules blocking access to sensitive dotfiles, cloud credentials, SSH keys, and persistence hooks in the user profile, while preserving seamless access to standard Documents, Desktop, and project files.
• Secure credential migration: eliminated cleartext fallback files during Keychain synchronization, ensuring all credentials remain permanently encrypted inside the macOS DataProtectionKeychain.
• Sanitized terminal execution prompts: terminal confirmation dialogs now enforce strict character limits and newline sanitization, preventing multiline spoofing and formatting anomalies before commands run.

Version 0.2.7 (Apocalyptic Captain)

Beta 2 · September 2026

• Autonomous health diagnostics: you can now run a deep on-device audit of OOMU by tagging a prompt with @diagnose. It gathers a structured health snapshot of your local credential store, mod listeners, model catalog integrity, and macOS sandbox permissions. This diagnostic capability is strictly gated to the @diagnose tag, ensuring standard conversation and implicit tool invocation remain completely uninterrupted.
• Native file creation in chat: asking OOMU to write or save a Markdown, text, or documentation file now uses a dedicated native file-writing tool. This eliminates command truncation and formatting issues when creating detailed documents.
• Smooth tool tag suggestions: typing @ in the prompt box now opens an unclipped menu showing five suggestions cleanly, with smooth scrolling for the rest and a refined layout that stays contained in compact windows.
• Seamless file inspection: asking OOMU to verify or read back files you just discussed or saved now binds the file path reliably, without parameter errors.
• Google Workspace connection reliability: fixed an issue where connecting a Google account in release builds could fail to detect desktop OAuth configuration settings.
• Continuous daily activity tracking: anonymous lifecycle telemetry now cleanly renews across day boundaries and after waking from sleep, keeping daily active usage visible in private product metrics without collecting personal content or interaction data.

Version 0.2.6 (Apocalyptic Captain)

Beta 2 · September 2026

• On-device voice dictation: clicking the microphone button now records and transcribes speech entirely on your Mac using Apple's built-in speech recognition framework. It is completely private, runs 100% offline, and uses zero cloud tokens or model memory.
• Multi-step tool execution for local models: tagging prompts with tools like @Terminal or @Files now tells on-device models to execute actions immediately and chain subsequent steps together smoothly, without conversational hesitation. Untagged prompts keep local models safe and conversational, while cloud models continue to infer intent and run multi-step tools automatically.
• Reliable startup windows for large local models: importing and running large models (up to 30 GB or more) on high-memory Apple Silicon Macs now scales the initialization timeout dynamically, eliminating false-positive timeout errors during cold starts.
• In-app update notifications: an overt notification banner alerts you when a new release is available, letting you restart and install immediately or defer while finishing your current work.

Version 0.2.5 (Apocalyptic Captain)

Beta 2 · September 2026

• Reliable document and spreadsheet reading: asking OOMU to inspect or analyze files using natural English sentences (such as ending a file path with a period) now correctly binds the path instead of failing with an argument validation error.
• Robust terminal command synthesis: executing commands that generate large listings (such as listing hundreds of files in your Downloads folder) now cleanly formats and returns the complete output, eliminating generic internal processing errors during response repair.
• Canonical document preview helper links: the "Get LibreOffice for Previews" button in Settings now links directly to the canonical LibreOffice download page instead of a broken 404 URL.

Version 0.2.4 (Apocalyptic Captain)

Beta 2 · September 2026

• Unshackled cloud model selection: connecting a cloud provider (such as Google AI Studio) now immediately unlocks all catalog models under that provider (such as Gemini 3.8 Flash, 3.6 Flash, and 2.5 Pro) for pinned and automatic chat sessions, eliminating provider-lockout errors.
• Synchronized Settings and active defaults: changing your default cloud model in Settings now immediately updates the provider card header and reliably persists across application restarts.
• Full agent memory recall: asking personal agents to inspect or review their memories now delivers complete factual records directly into their conversation context so they can cite and discuss stored notes accurately.
• Ambient working memory: agents automatically enter new conversations with active awareness of recent stored project context and facts without requiring manual lookup queries.
• Hermetic production packaging: the application now compiles exclusively as a 100% standalone, offline-embedded bundle with zero local development server dependencies.

Version 0.2.3 (Apocalyptic Captain)

Beta 2 · September 2026

• Autonomous on-device tool execution: on-device models now execute local directory and terminal requests immediately rather than getting trapped in conversational permission loops asking if you would like to proceed.
• Seamless affirmative follow-ups: replying with natural confirmations such as "yes, please do it" after discussing a file or command now directly resolves context, performs the action, and delivers the verified output.
• In-chat file and document analysis: reading local documents, spreadsheets (.csv), and Markdown files by pasting or typing their absolute paths now executes reliably without parameter validation errors.
• Responsive over-the-air model catalog: the catalog now checks for published updates on Cloudflare R2 on every launch using conditional headers, eliminating the 24-hour waiting lockout.
• Fresh install catalog currency: the initial setup screen now immediately reflects newly published models as soon as background sync completes. In addition, the latest 14 providers and 107 models are compiled directly into the binary for complete offline currency.
• Manual model refresh: added a dedicated "Refresh models" button in Connections across all 12 supported languages.
• Composer menu polish: the message options popover ("Queue message", "Steer reply") now closes cleanly whenever you send, queue, click outside, or press Escape.

Version 0.2.2 (Apocalyptic Captain)

Beta 2 · September 2026

• Local file and folder browsing: asking OOMU to inspect your Downloads or project folders now lists and reads local files directly without hesitation.
• Cloud document and spreadsheet reading: local PDFs, spreadsheets (.xlsx), and Word documents (.docx) dropped into chat now pass securely to cloud models for fast, accurate summarization and data extraction.
• Confirmed terminal execution: OOMU can now run shell commands directly on your Mac, always presenting the exact command and working directory for your explicit confirmation before execution.
• Connected Google Workspace in chat: your connected Google account is now directly accessible in conversation, enabling search and lookup across Gmail and Google Drive.
• Agent memory graduation: personal agents now reliably commit important preferences and facts to local encrypted memory during chat sessions and recall them in new conversations.
• Signed over-the-air model updates: the catalog of available AI models now updates automatically over the air with cryptographically signed releases, adding frontier models without requiring full app reinstalls.
• Polished chat notices: the persistent green notification banner after editing a chat name now smoothly dismisses itself after four seconds and includes an instant close button.

Version 0.2.1 (Apocalyptic Captain)

Beta 2 · September 2026

• Fixed document attachment summaries: dropping a PDF or report into chat and asking for a summary now reads the text reliably instead of asking you to attach a document you already provided.
• Restored hardware-backed audit signing: enterprise audit exports now sign and verify cleanly against the Apple Silicon Secure Enclave in packaged release builds.
• First-run license terms now display reliably in all 12 supported languages.
• Added visual navigation arrows to introductory feature cards so they are easier to browse.
• Search follow-ups now carry the subject forward in English: asking to search for good choices after discussing a place or topic preserves the name.

Version 0.2.0 (Apocalyptic Captain)

Beta 2 · September 2026

• First public release of OOMU Beta 2. The codename honors Shuyue.
• Rebuilt with a pure agentic harness: removed synthetic classification layers and restored true native Apple Keychain security.
• Streaming verified model acquisition directly from upstream checkpoints with local file-pointing affordance.
• Decoupled optional document previews with native LaunchServices viewer fallback.
• A calmer, zero-idle workspace that stays responsive while work runs.