Look at the changes.
Readable diffs and guarded file restoration help you inspect what happened.

Codex, Google Antigravity, Claude, cloud APIs, and local models.
One native workspace for the things you want to make.
Work on files. Share a browser view. Inspect changes. Turn a review into Tasks, Bugs, and Features you can follow through a report’s Kanban board.
Readable diffs and guarded file restoration help you inspect what happened.
Browser captures and previews give you and your assistant a shared view.
Stable action references follow the work across models and chats.
Open + beside Projects and choose Import from Codex. Search chats by project, keep their history, and link the original folder or choose its new location.
Restore covers captured file changes, not external effects. Parallel workers share project files.
A coding engine you already use. A cloud model for the bigger picture. GalaxyDesk Native on your own machine, or another local runtime. Give each conversation the intelligence that fits.
Use Google Antigravity through your installed agy CLI, alongside Codex and other coding engines. See the account allowances reported by supported CLIs.
Devin CLI is available as a preview.
Use your provider accounts and the models they make available.
Check configured models and prices automatically, every 30 minutes by default. Completed usage keeps the rates recorded at the time.
Open Model Library in the shared workspace pane to browse live GGUF listings, compare hardware and memory estimates, download a verified file and load it natively. You can also connect a separately running local server, including compatible API endpoints.
GalaxyDesk Native is included. External accounts and runtimes are supplied separately. Model capabilities, hardware needs, and provider charges vary.
You shouldn’t have to start the story again every time you try another model. Keep saved decisions and conversations together, with recalled sources you can look inside.
Add, edit, and remove the specifics that guide the work.
Search saved messages after compaction and return to the source.
Inspect recalled snippets before building on them.
Memory supports continuity; it does not guarantee a model recalls or interprets every detail correctly.
Pair a Primary with a Secondary. Give one the lead and the other focused implementation work, then bring the result back for review.
Choose a pairing that fits the task and your budget, including a local worker when your hardware supports it. Delegation and review depend on the models you choose.
A Rust agent engine and a Tauri desktop form the foundation. Your models, your tools, and your personal touch make it a place to build.
Named connections and provider credentials use protected OS-keyring storage.
Choose project and chat access. Inspect activity, steer supported engines, and stop managed work.
See token usage, supported account allowances, and estimates with the model rates recorded at the time.
Create native Git worktree chats with isolated folders and branches, shared history, and controls to review or archive the work.
Meet Emma, your new assistant, alongside five familiar companions for your desktop.
Emma, Violet, Penelope, Merlin, Clippy and Jeeves are all included. Emma is selected for a new profile; your existing choice stays yours. Choose a companion in Configure → Settings → Pets and turn on Pet Assistant.

Emma works at her laptop, types while you build, and gives a thumbs-up when the work is done. Violet glasses, a knowing smile, and a little Red Bull.
Included with GalaxyDesk
A knowing violet-eyed rabbit with a carrot, a little attitude, and a front-row seat to whatever you build.
Included with GalaxyDesk
A fluffy, blue-eyed panda with her favourite bamboo. Soft company for your next big idea.
Included with GalaxyDesk
The blue-robed wizard is back at your desk. A familiar face for a new generation of projects.
Included with GalaxyDesk
The inquisitive silver paperclip returns, bringing that unmistakable expression to your workspace.
Included with GalaxyDesk
The unflappable butler has come out of retirement. Quietly composed, even when the build is not.
Included with GalaxyDeskLive bubbles name the project and conversation. Multiple chats stack, with a checkmark when work finishes. Click a bubble to return to that chat.
Slide from the original size to twice as big and see the change immediately. Your companion grows; bubble text stays comfortably readable.
Drag your pet into place, switch companions, or turn the overlay off. Custom pet packages remain supported, and your saved choice carries into updates.
Choose your platform. Connect your intelligence. Open a project.
Choose the installer for your operating system and processor. Connect your own provider account or local model runtime, then open a project and start building. Your saved projects and conversations stay in application data. Already using GalaxyDesk? Update and restart in the app for the latest version.
EXE installer
Windows 10 or later; Windows 11 recommended.
DMG disk image
Apple Silicon Mac with macOS 13 Ventura or later.
AppImage
Modern 64-bit glibc desktop; Ubuntu 22.04+ or Debian 12+ recommended.
8 GB RAM is recommended for cloud-connected work. Local models need additional RAM or VRAM depending on the model and context size.
Allow around 1 GB for the app, plus free space for updates, projects, attachments, and any local models.
Use a glibc desktop with WebKitGTK 4.1 and an X11 or Wayland session. AppImage mounting requirements depend on your distribution. If mounting is unavailable, use the extract-and-run option.
The desktop uses WebKitGTK. Shared browser tools use a separately installed compatible browser and its security requirements. Encrypted credentials also need an unlocked desktop keyring.
Install Git, Node.js, Python, compilers, or other dependencies when the project needs them. GalaxyDesk Native is included; other local model servers and coding CLIs are installed separately. Cloud work needs internet and the relevant provider account. Account allowance indicators require a signed-in, supported Codex or Claude Code CLI. Smart auto-complete needs the app open and unlocked.
x64 also means x86_64/AMD64. ARM64 also means AArch64; M-series Macs use ARM64.
Sources: Tauri system dependencies · System web renderers · Local model context and memory.
Explore the connections, controls, and capabilities behind the workspace.
Choose a model for each chat, or pair a Primary and Secondary in Multi-Model Mode. Keep your profiles, preferences, and saved work together.
Plus compatible local API servers.
Use their own installed CLI and sign-in.
Quick setup detects Ollama and supported installed CLIs. Models, runtimes, and provider accounts are supplied separately; their availability, tool support, and charges vary. Unsupported reasoning-effort controls are disabled.
Bio, skills, and global/project Specifics grow only as needed beneath adjustable caps. The conversation and tools receive most of the remaining space.
Default split of the remaining working budgetNew-workspace caps: bio 2,560; global Specifics 1,024; project Specifics 1,536; skills 4,096 tokens. Existing saved allocations carry forward. These layers use at most 25% of available input; unused room can return to the conversation.
Set an automatic compaction percentage or use /compact in a project chat. Search saved messages after compaction, recover tool results, and refer back to image OCR and available visual descriptions.
Keyword retrieval can miss a detail. Project-chat originals remain available to inspect. Assistant keeps a rolling history and clears older completed turns; its submitted packet is capped at 16,000 tokens or your model’s smaller capacity. Native coding engines manage additional internal context; supported Claude Code windows start at 100,000 tokens.
Emma’s personality and portrait are included. In Personalization, choose Custom to edit your assistant’s name, bio and avatar; your saved Custom profile stays available when you return to Emma. Thirteen bundled help guides stay available offline with either choice, with relevant sections supplied as you ask.
Follow lifetime, project, and per-response usage—including tool rounds, compaction, image descriptions, and delegated work.
Reported counters and estimates stay separate. Available cost information is labeled with its source; missing prices remain unavailable.
These are usage records and optional estimates, not reconciled provider invoices. Raw native-terminal activity stays separate from the app ledger.
Token-free reads show remaining Codex and Claude Code allowances every minute, with reset tooltips. At low allowance, Codex additional credits appear as a nominal USD equivalent; a positive remainder below 1% displays as <1%. Claude Code extra usage shows spending and a monthly cap—not a prepaid balance; the live usage feed has no prepaid-balance data.
For OpenRouter, an optional encrypted Management Key reads account credits separately from the API-key budget. Without one, an ordinary API key reporting no budget (not a known account balance) shows ∞ with this help: “No budget allowance set, usage is based on Credit Balance, configure the optional Management Key to see your current Credit Usage.” Clicking a known balance opens a browser choice for the credits page.
Turn on Smart auto-complete to pause near your chosen limit and resume the same task after a verified reset. Saved progress stays in the conversation; the continuation appears as “Resuming…”.
Account reads use no model tokens; regular model work can still consume provider allowances or credits. The app must be open and unlocked. Periodic readings can lag consumption, and automatic continuation uses your normal model allowance.
Your files, next ideas, and active conversations—powered by a new native Rust engine.
Select Assistant above New chat for one ongoing conversation across your registered projects. Review saved progress, discuss priorities, and open a project chat when you want to change files.
Automatic source checkpoints run in Rust before and after chat work, with no Git installation needed. Review commits and earlier files without touching your existing repository.
Opt in to local access for phone progress, animated pet bubbles and project chats. Scoped API tokens let your own tools connect with the permissions you choose. Download the integration skill.
Filter Tasks, Bugs and Features as work moves from Open through In Progress to Completed. Open a card for its details, keep report status current, and remove stale items when plans change.
Choose a provider in a clear connection panel, configure it inline, and save and test together. Connect Dropbox, Google Drive, Box, OneDrive, FTP/FTPS or SFTP; browse folders, search and select multiple files.
Choose a chat for a temporary text reference or download an editable project copy. Publish edited outputs explicitly; changed remote content blocks the upload. Sidebar browsing also offers local uploads.
Pet bubbles show the project and chat, refresh their reply excerpt, and stack while several chats work. Click a bubble to open its conversation; dismiss finished ones individually.
Send follow-ups while the assistant works. Edit, delete, or drag them into order. Five stacked rows stay in view; the rest scroll above the composer.
Steer the active Codex turn, or interrupt and redirect Claude Code with your queued instruction. Other providers run queued prompts in order after the current task.
Turn on Smart auto-complete in Usage limits. Choose a remaining percentage, pause near it, and resume only after a fresh account reading confirms the reset.
Remote connections are optional. Connect GitHub, GitLab, Bitbucket, or Azure DevOps, then link a connection to each project. Clone, inspect changes, manage branches, and commit specific files.
Connect Magnitude’s local inference service and discover installed models with reported context and reasoning controls. Magnitude runs on Windows through WSL.
A left-hand menu keeps configuration organized. GalaxyDesk, Lime and Strawberry Orange Burst palettes carry through the workspace. Adjust UI/control size and chat text independently, with live previews before saving.
Save a project review as Action Items. Follow up with another model, preserve earlier assessments, and attach a specific item to the conversation.
Read and edit project files, review clear diffs, and restore captured changes when it is safe. Later edits are checked before a restore.
Use the same browser page as your assistant. Captures stay inline as clickable thumbnail grids, between the replies and tools that produced them. Full access chats can open localhost previews directly.
Tool-call budgets reset when the assistant posts a visible update. Suspected loops show their evidence and let you decide whether to stop or continue. Adjust the limit in Execution settings.
Enable subagents and choose a shared limit of one to eight tasks. Review each task’s result and activity; stopping the parent stops its children.
Configure global or project MCP servers. Add Markdown skills, turn individual skills off, and reuse the instructions that fit your projects.
Emma joins Violet, Penelope, Merlin, Clippy and Jeeves as an included desktop companion. Turn on Pet Assistant and choose your company. You can add your own Codex-format pets, too.
Stacked bubbles identify each project and chat. Their text updates as new replies arrive; a checkmark marks completion. Click the bubble to open that conversation and dismiss its completed notification.
Pets are optional and hide when GalaxyDesk is locked. Imported pets stay on your computer.
Subagents share project files. File restore covers captured changes, not external effects or a complete workspace checkpoint. Browser sessions are isolated by chat and end when closed; some embedded-frame controls and model uploads are unavailable.
We rebuilt the machinery behind GalaxyDesk. The desktop host and agent engine are native, with Rust doing the heavy lifting.
158 MB → 24 MB
2,725 MB → 460 MB
Windows · macOS · Linux
Native Rust host + engine
1 Windows ARM64 installer, 0.6.0 versus 0.8.1. Download reductions vary by platform and version.
2 Demonstrated startup readings on one Mac: 0.6.5 “app” versus 0.7.0 “host + agent.” The latter excludes system WebView memory; these are different process scopes, not a total-RAM benchmark. Results vary.
The agent loop, provider connections, tools, storage and background services run in Rust. No Electron shell. No bundled Node service. Just a focused native foundation for your work.
Tauri uses your system’s web renderer instead of shipping another Electron runtime. Your familiar interface stays, with less machinery packed around it.
Rust’s ownership and type system catch whole classes of memory and concurrency mistakes during development. Managed tasks also have explicit cancellation and shutdown paths.
Multi-Model Mode directs the Primary to delegate implementation to the selected Secondary and review its work. This is guidance, not a guarantee that every model will follow it. Pair models that suit your task and budget, including a local worker when your hardware supports it.
The compact Multi-Model pill keeps its styling while busy. Hover or keyboard-focus it to reveal model details beside two live token charts: the current task, including review continuations, and the entire chat, with Primary and Secondary role totals across earlier tasks and models. Each Secondary assignment counts once. GalaxyDesk keeps the pairing in one conversation.
Saved file changes live in local archives and remain available when needed. The interface loads a compact view instead of hauling every historical file snapshot into chat.
Signed updates are verified before activation. A new version must confirm that it started successfully; a failed startup can recover the prior version and saved state.
Keep your projects, chats, settings, cloud connections and companions. This is a new foundation for the workspace you already made your own.
Built with Rust and Tauri. Rust does not make approved commands an OS sandbox. External coding CLIs, browsers and project tools keep their own requirements.
Choose what your assistant may do, where inference runs, and how your workspace is protected.
Keep cloud tokens and server passwords encrypted with your OS keyring. Project Secrets also support HTTPS, FTPS, SFTP and SSH tool connections without placing passwords in chat.
Set a password, lock manually, or require it at startup. Locking clears private UI, cancels managed activity, and suspends allowance checks and automatic continuation while the rest of your desktop stays usable.
Set a project access ceiling and stricter chat limits. Full access covers ordinary web and connected tool calls; Ask mode still requests approval. Saved secret-connection actions keep their own approval.
Profiles, chat history, queued prompts, attachments, and search records are stored on your computer. Cloud references remove their temporary download after use; editable copies stay in your project until you manage them. OCR runs locally. Allowance checks use the signed-in native CLI; the app does not expose account credentials in chat.
Choosing a cloud provider sends the context needed for the request to that provider. An image description uses your selected image-capable model.
Approved commands run as your OS user. Project access rules do not create an operating-system sandbox, and local MCP programs may have broader machine access.
The app lock protects the app’s view; use your OS account and disk protection for project files and the rest of your computer.
Secrets require an available OS keyring. SSH/SFTP host keys need explicit trust, TLS certificates are checked, and tools stay bound to the saved connection. Downloads and uploads remain subject to project and transfer limits.
Integrated coding engines use their native authentication. Optional raw Native sessions retain the CLI’s own history and permissions; bridged GalaxyDesk project actions still use the app’s controls.
Choose the workspace around the way you want to work. GalaxyDesk brings model choice, project continuity, and visible tools together.
Explore the site’s seven-product comparison. Coverage, scope, and sources matter; this is the publisher’s assessment, not a quality ranking.
A feature-by-feature view of GalaxyDesk and six alternatives. Product families include the surfaces named below; integrations and partial equivalents are called out. Select a cell for details and its source.
✓ Supported× Not built in◐ Partial / integration? Unverified
Partial includes integrations or narrower support. Unverified is not a confirmed absence. The gaps filter shows where GalaxyDesk has less support than a documented alternative.
| Feature | GalaxyDesk1.0.0 · Native desktop | CodexDesktop + CLI + Cloud | Claude CodeCLI + desktop + IDE + Cloud | CursorEditor + agents + Cloud | ClineEditor + CLI + SDK + Kanban | AiderTerminal workflow | DevinDesktop + CLI + Cloud |
|---|---|---|---|---|---|---|---|
| Files & presence | |||||||
| Cloud file browser with reference, edit and publish | |||||||
| Project-labeled, stacked live pet chat bubbles | |||||||
| Built-in and custom pets with adjustable size | |||||||
| Searchable emoji picker in chat | |||||||
| Models & workspace | |||||||
| Choose among cloud model providers | |||||||
| Local model inference through Ollama | |||||||
| Codex and Claude agents in one workspace | |||||||
| Full built-in code editor and debugging UI | |||||||
| Inline editor code completion | |||||||
| Symbol-aware repository map or semantic code discovery | |||||||
| Conversation & continuity | |||||||
| Queue follow-up prompts during work | |||||||
| Edit and drag-reorder a visual prompt queue | |||||||
| Redirect a running task | |||||||
| Interleaved replies, tools and visual evidence | |||||||
| Custom assistant name, avatar and biography | |||||||
| Global and project instructions | |||||||
| Editable token budgets per context layer | |||||||
| Local semantic recall across project chats | |||||||
| Search original chat after compaction | |||||||
| Retain screenshot OCR and descriptions for later recall | |||||||
| Usage & control | |||||||
| Codex and Claude allowance windows shown together | |||||||
| Automatic continuation after a usage reset | |||||||
| Configurable low-allowance pause before exhaustion | |||||||
| Lifetime, project and response token accounting | |||||||
| Password-lock just the workspace app | |||||||
| Tools & reviews | |||||||
| Run commands and edit project files | |||||||
| Shared interactive browser | |||||||
| MCP tool integrations | |||||||
| Project review findings with persistent status | |||||||
| Kanban board for project work | |||||||
| Primary model directs a different worker model | |||||||
| Optional subagents / parallel work | |||||||
| Automatic local source history | |||||||
| Conversation & continuity | |||||||
| Phone access to ongoing work | |||||||
| Tools & reviews | |||||||
| External control API or SDK | |||||||
| Review diffs and restore prior edits | |||||||
| Encrypted named project secrets for tool use | |||||||
| Execution & isolation | |||||||
| Managed isolated Git worktrees | |||||||
| OS-level sandbox or isolated cloud VM | |||||||
| Cloud task continues with your computer closed | |||||||
| Execution & comfort | |||||||
| Native Rust desktop host and agent engine | |||||||
| Independent UI/control and chat-text sizes | |||||||
| User review of suspected tool loops | |||||||
43 features · 7 products
New capabilities and improvements, release by release.
Local predictions require a separately downloaded model. ChatGPT plan connections require a running CodexAppEngine service.
Prediction performance is currently qualified on Apple Silicon. Vulkan device discovery on older Windows GPUs remains under investigation.
We thought keeping the Shared Browser bar visible would make its controls easier to reach. We didn't consider how frustrating it could be when the bar wouldn't go away. We're sorry. You can now hide or show it, and we'll be more considerate about changes to your workspace.
We thought keeping the Shared Browser bar visible would make its controls easier to reach. We didn't consider how frustrating it could be when the bar wouldn't go away. We're sorry. You can now hide or show it, and we'll be more considerate about changes to your workspace.
GalaxyDesk Native is included. External accounts and runtimes are supplied separately. Model capabilities, hardware needs, and provider charges vary.
Use the 0.9.4 installer once to refresh the launcher. This release focuses on Emma and a lighter workspace; VM workflows are not included.
Cost estimates use recorded input and output token counts where available. Codex and Claude Code amounts represent comparable API value, not charges to your subscription; unknown costs remain unpriced.
A signed small update. Your model context, saved work and access settings stay under your control.
Available as a signed small update for the 0.8.4 desktop launcher. Fresh installations use the 0.8.4 installer, then receive this update through the same channel.
This repair changes the desktop launcher and requires the full 0.8.4 installer. Your chats, settings and projects are retained. Future compatible updates continue through the same update channel.
Handoffs contain reasoning actually returned by the provider, not inaccessible internal thoughts. Project memory remains local and editable; selected context is sent to the model you choose. Compatible desktop installations receive a signed in-app update.
Inspired by MemPalace’s preservation of original evidence. Storage and search run locally; selected context is sent to the model you choose. Available as a signed in-app update for the current desktop installation.
Install the full 0.8.1 version to receive the corrected launcher. A small update cannot replace the launcher already installed on your computer. Your chats, settings and projects stay in place.
Local access is opt-in and requires the desktop app to remain open. LAN access uses HTTP on your trusted private network. Local history is a bounded source record, not a full backup or isolated worktree. Available as a signed in-app update; new users install the current desktop download, then update.
Older conversations without steering-position markers keep their saved layout.
A persistent provider outage can still prevent completion. Work is not automatically replayed after tools have started.
This release uses user-supplied access tokens or server credentials. Reference supports UTF-8 text up to 2 MB; editable copies and uploads support files up to 25 MB. Native Google documents can be referenced through supported text exports, but must be exported before round-trip editing.
Enable Smart auto-complete in Configuration → Usage limits. The app must be open and unlocked to continue. Readings depend on the installed CLI and signed-in account; periodic checks can lag fast consumption. Continued work uses your normal model allowance.