What's New in Glass.

A chronological timeline of iterations, improvements, and architectural evolutions deployed to Glass.

The complete agent workspace

Glass 2.0 brings your models, apps, tools, and live projects into one focused workspace. Connect Claude, Codex, Gemini, or Grok, then let Glass build, run, inspect, and refine your work without breaking your flow.

Four first-class AI providers: Grok joins Claude, Codex, and Gemini, with provider-aware model discovery, reasoning controls, context limits, and more reliable session routing.
Simpler account setup: Guided sign-in, clearer connection status, automatic runtime setup, and self-updating provider runtimes make it easier to get each model working and keep it current.
Connected apps: Find and connect tools such as Gmail, Google Calendar, Drive, Notion, Slack, WhatsApp, Outlook, Sheets, Canva, and YouTube directly from Settings.
One home for API keys: Save keys once in Glass, manage them from Settings, and make them available across projects while still allowing project-specific values to take priority.
A Preview agents can use: Glass can open your app, inspect the page, click, type, navigate, capture screenshots, and verify its own work from the built-in Preview.
Built-in development servers: Glass can start, detect, list, recover, and stop local development servers, with their live output and Preview URL kept together.
A better Skills experience: Search your Skills, distinguish provider-native capabilities at a glance, and use expanded Codex plugin and app support without managing a wall of toggles.
A guided first project: The new interactive tutorial walks through account setup, safety, models, working folders, Skills, connected apps, API keys, Preview, Files, and building a complete local project.
Clearer chat and tool activity: Streaming Markdown, syntax highlighting, Mermaid diagrams, grouped tool results, cleaner file changes, and improved terminal output make agent work easier to follow.
More dependable long-running work: Improved message queues, send protection, session recovery, model persistence, image handling, and provider transport recovery reduce interrupted or duplicated turns.
A sharper Glass workspace: Refined navigation, tabs, Files, Settings, onboarding, dark styling, Low Power Mode, a new welcome experience, and refreshed application icons bring the whole product together.
Security and release hardening: Stronger tool approvals, safer credential storage, verified runtime downloads, tighter environment handling, and mandatory pre-release checks protect every supported provider.

A broader Glass foundation

Glass 1.7.5 expands model choice, brings Skills and MCP Servers deeper into Settings, and sharpens the chat, update, and workspace flows across Claude, Codex, and Gemini.

Expanded model picker: GPT-5.5 and Claude Opus 4.7 were added to the unified Glass model picker.
More durable model sessions: Glass remembers the selected model per chat session and can restore the provider binding when reopening or continuing a session.
Shared Skills in Settings: Shared Glass Skills now live in ~/.glass/skills and work across Claude, Codex, and Gemini, with Settings controls to create, open, enable, disable, and delete Skills.
Provider-native Skills discovery: Glass also surfaces Claude and Codex native Skills, while cached Skill lists remain visible during refreshes so temporary provider errors do not blank the UI.
MCP Server management: Configure MCP servers for Claude, Codex, and Gemini from Glass Settings, backed by ~/.glass/mcp.json and ~/.glass/mcp.env.
Provider runtime support: MCP server definitions can be compiled into provider-specific runtime configs with support for stdio, HTTP, SSE, env files, required servers, and tool include/exclude filters.
Richer chat rendering: Chat now has cleaner Markdown and code rendering, Mermaid diagrams, clearer tool call blocks, collapsible thinking streams, image attachments, better titles, session restore, and conversation compaction.
Smoother approvals and modes: Provider questions and approval prompts now appear inside Glass, while Safe Mode, Autopilot, and Full Send have clearer behavior for different levels of tool access.
Workspace polish: The main workspace and Settings received usability improvements that make daily navigation and setup feel cleaner.
Under-the-hood upgrades: Glass now builds on a VS Code 1.116 foundation with updated Electron/dependencies, packaging preservation, extension compatibility cleanup, broader tests, and more reliable Claude, Codex, and Gemini provider transport.

Fixed a critical issue where Intel Mac users could not authenticate with AI provider CLIs (Claude, Codex, Gemini). Login flow now works correctly on x86_64 builds.

Reliability across updates, chat, and CLIs

Glass 1.7.3 focused on reliability: cleaner app updates, more dependable CLI login, better Codex reasoning output, Gemini rate-limit fixes, and a smoother Glass Chat surface.

Better update experience: The titlebar can show update status, Settings now includes an Updates section, and the update flow has clearer checking, downloading, preparing, and restart states.
Cleaner chat startup and reasoning: Assistant turns show a lightweight loading state while providers start responding, and thinking blocks now stay visible only when real reasoning content is available.
More reliable Codex sessions: Glass preserves provider thread continuity more carefully when switching Codex model variants in the same workspace and runtime mode.
CLI login fixes: Codex CLI detection avoids the desktop app bundle, stale executable checks refresh correctly, shell startup files are left untouched, and Codex login avoids double-opening the browser.
Gemini rate-limit hardening: Gemini requests include additional provider metadata to reduce 429 failures and make Gemini turns more dependable.
Workflow polish: Agent is now labeled AI, Canvas is now labeled Preview, the welcome screen behaves better on smaller windows, and bug reports now route into the Glass support pipeline.

The Genesis Release

Welcome to Glass. This initial release marks the beginning of the most fluid, aesthetic, and AI-native development environment ever built. We've torn down the legacy constraints of traditional editors to craft a workspace designed for pure velocity.

Liquid Glass Aesthetic: Stripped away all generic Microsoft visual boilerplate in favor of a stunning, refractive UI with immersive depth, blur effects, and geometric precision.
Native Next-Gen AI Integration: Sovereign agentic models are built directly into the engine's core architecture, rather than being layered on as a bolted-on extension.
Frictionless OTA Pipeline: A custom-engineered update infrastructure using Cloudflare R2 means you get instant, silent background capability upgrades, bypassing clunky traditional update prompts.
Distraction-Free Orchestration: Total elimination of jarring color palette flashes (like the default Microsoft blue/purple status bars) in favor of sleek, coherent neutral tones that never break your focus.
Zero Microsoft Telemetry: Total independence from Microsoft's pervasive data telemetry, tracking hooks, and cloud polling logic for a genuinely private, sovereign environment.
Focus-First Canvas: Ruthlessly minimized default sidebars, action bars, and widget clutter to prioritize code visibility. The tool gets out of your way.
Premium Typography: Hand-tailored default font weights, syntax highlighting curves, and line-height configurations optimized for extended deep work and modern Retina rendering.
Pre-Optimized Perfection: No more spending hours configuring hundreds of generic JSON settings just to make the editor bearable—Glass looks and feels perfect on boot.
Seamless Authentication Layer: Custom license validation gateways that protect the build without blocking your offline workflows or requiring complex wrestling during a session.
Built for Creators, Not Mechanics: Glass is designed to let you describe what you want and watch it instantiate, rather than forcing you to constantly manage your local environment configuration.