Changelog

New model support, a higher-powered tier, and smarter agents across the platform.

• Added Sonnet 4.6 and Kimi 2.5 as available models for agent tasks
• Launched Max tier plan with increased limits for power users
• 8GB instances added to the warm pool, so larger VMs now boot instantly alongside 4GB
• Custom profile picture uploads: use your own PFP instead of the default Google avatar
• Improved prompt engineering and context injection for better agent performance
• Fixed the chat stop button
• Patched React Server Components CVE and other security vulnerabilities
• Fixed file sync logic across desktops in multi-VM projects

Credits-based pricing, Claude Opus 4.6 agent support with fast mode, and a rebuilt terminal.

• Credits and usage-based pricing: top up credits on demand, monthly auto-refresh, per-call usage tracking, and credit gauge in the dashboard
• Claude Opus 4.6 support: new agent loop architecture with streaming, tool definitions, and fast mode for quicker agent responses
• Bash tool for AI agents: agents can run shell commands on VMs with foreground process support and working directory tracking
• Terminal overhaul: tabbed UI, resize handles, improved auth flow, and connection retry logic
• Navigation bar fix: important actions (New Computer, Files, Templates, API key) no longer pushed offscreen
• Code tab improvements: dark theme, line numbers, horizontal resize handle
• Chat upgrades: speech synthesis for composer input, better message rendering, thread context
• Bug fixes: workspace creation, account linking, free tier credits updated to 500

More reliable agents, faster startup, easier debugging.

• SSH directly into your VM to debug your agent without API overhead
• Long-running commands now finish, with no more processes killed mid-task
• Snappier desktop UI: the display loads instantly after a 500ms boot
• Playground chat remembers context, so follow-up questions work now
• Cleaner credentials UI: API keys are easier to find and copy
• Fixed memory issues causing random session crashes
• Removed confusing RAM options (2GB gone, 8GB blurred on free plan)
• Files tab navigation fixed
• Light mode color fixes

Major performance improvements and SDK reliability updates.

• 4x faster page loads (from ~2s to ~500ms) with optimistic updates and request bundling
• Added auto-retry to SDK for improved request reliability
• Workspace links now work: visit computers directly from SDK-generated URLs
• Fixed scroll position bug that hid computer IDs when navigating from workspace page

Larger instances, higher tier limits, and improved agent tooling.

• Added 32GB RAM instances for heavier workloads
• Increased tier limits: 5 computers on Developer, 50 on Startup
• Improved computer.prompt() for more reliable agent performance
• Refreshed TypeScript package for cleaner npm install experience
• Fixed file upload bug affecting UI sync
• Updated documentation for faster developer onboarding

Core infrastructure improvements for faster boot times and enhanced stability.

• Optimized desktop images with 2x faster p50 boot times and reduced memory footprint
• Added Google Chrome support as the default browser for desktop instances
• Fixed critical API stability issue that caused intermittent crashes
• Resolved terminal command blocking that previously caused system hangs

File export capabilities and platform stability improvements.

• Added file export to download files directly from virtual desktops
• Updated desktop backend with file transfer support
• Multiple platform stability improvements and bug fixes

Backend infrastructure improvements and developer documentation.

• Rolled out new proxy backend which reduces latency on the computer action API by 75%
• Added new documentation for embeddable virtual desktops
• Added llms.txt and llms-full.txt to documentation for LLM compatibility
• Platform improvements for reduced latency on virtual desktop live streams
• Multiple UI bug fixes and improvements

Embeddable desktops, new plans, and developer tooling.

Currently building file download/upload for workspaces, improving computer use agent capabilities, and working towards general availability for Windows.

• Virtual desktops that are embeddable into your own application
• New plans supporting larger resource limits and deployments
• Predefined applications and configurations for new desktops
• Several bug fixes and stability improvements

Product improvements for developer productivity and UX.

Currently building file download/upload for workspaces, developer-defined templates to preboot computers with custom applications and configurations, and embeddable virtual desktops.

• New global search to find workspaces and account resources
• UX: Navigation layout supports larger workspace counts
• UX: Chat interface displays suggestions on submission
• UX: Live stream maintains proper aspect ratios
• Improved invite flow for teammates to share compute resources
• Improved terminal reliability with automatic retry
• Additional platform stability improvements