Claude Opus 5, Reasoning Visibility & Codex Catalog
Claude Opus 5 is now available, each message shows its reasoning effort level, and Codex dynamically discovers which models your subscription supports.

NucleusNucleus · Changelog
Every Nucleus release — features, fixes, and the small craft work in between. Written down so it doesn't disappear into a commit log.
Claude Opus 5 is now available, each message shows its reasoning effort level, and Codex dynamically discovers which models your subscription supports.

Agents can now create, list, read, edit, and run automations directly from a chat session.

Thinking Machines is now available as an AI provider alongside Claude, Codex, and Terramind.

Run multiple code reviews simultaneously across uncommitted changes, branches, and specific commits — all without interfering with each other.

Agents can now render and preview HTML directly inside chat. MCP tools load faster and render more reliably across all providers.

Agents can now access your team's shared docs, channels, and calendar via the new Spaces MCP server. Codebase indexing status is visible in the sidebar and now works correctly in git worktrees.

Better handling of rate limits during indexing and sync, with automatic retries so background operations complete reliably.

Chat changes now sync reliably even if you go offline or the app closes unexpectedly — nothing gets lost.

Projects now connect for codebase indexing automatically when you send your first message — no manual configuration required.

Load your entire conversation history without truncation, and agents can now spawn new terminal panes on demand.

Add context notes to selected code before sending to the agent, inspect elements inside iframes in the HTML preview, and sub-chats now correctly fire their opening messages.

Agents can now spawn and message their own sub-chats, loops are a first-class chat mode with per-loop controls, and a new Insights dashboard shows your usage patterns and team analytics.

Under-the-hood stability improvements.

Drag images directly into chat from thumbnails, and unavailable models are now clearly marked so you always know what's accessible.

Reduced memory usage in long agent sessions by stripping tool outputs more efficiently.

Under-the-hood stability and reliability improvements.

Long conversations now load incrementally for better performance, parent chats auto-rename when you rename a sub-chat, and MCP authentication is more reliable.

Track your Claude Code and Codex usage in profile stats, view your public profile without leaving the IDE, and download attachments from automation results.

Browser mode is now Magic Mode with richer capabilities. MCP configs sync to the cloud automatically, attachment cards are redesigned, and Discord OAuth is now supported.

A new Pipeline Inspector lets you dive into security analysis details per commit, auto-review remembers your state, and GitHub app setup is now seamless.

Rate limit errors are now correctly detected, and the file viewer stays in sync on Windows and macOS symlinked paths.

Voice transcriptions are now automatically cleaned up and polished, and file watching is more reliable across all platforms.

Cloud-triggered agents now run in a dedicated read-only mode so you can monitor them live. Fork any cloud run into a full local workspace to continue work independently.

Your default agent model now syncs between the web and desktop apps so your preference follows you everywhere.

Fixed a crash when approving plans in certain views.

Chats load faster without premature loading screens, and sidebar widgets now persist correctly between sessions.

Chats no longer appear empty after syncing from the cloud, and UI state is more consistent across sessions.

Roll back and redo with confidence using git checkpoints, render OBJ 3D models natively, and see at a glance which collapsed chat groups have running agents.

Review individual commits with AI, manage your subscription directly in-app, and instantly switch providers when you hit usage limits.

Studio 3D models render with better lighting and materials, the sidebar is smarter about when to toggle, and dev servers are more responsive on first run.

Under-the-hood stability and reliability improvements.

Duplicate tool outputs are now automatically cleaned up on startup, and migration errors are handled more gracefully.

Generate high-quality 3D models and preview them directly in chat. Terramind Studio is now fully integrated into Nucleus.

A new Glass Mode brings a transparent, modern aesthetic to Nucleus. Widget preferences now persist per project, and keyboard shortcuts show native OS symbols.

Authentication syncs reliably across sessions so you stay signed in, and chat input no longer disappears when switching workspaces.

Chats load faster with preloading and lazy-loading, memory usage is significantly reduced, and sync is more reliable with bulk synchronization and crash recovery.

Nucleus now detects vulnerabilities in your code, explains each finding, and uses AI to generate and verify fixes automatically. Re-fix confirmed-but-missed vulnerabilities with new approaches guided by verification feedback.

Companion now syncs with your IDE with real-time updates, chat windows are draggable to split-view, and Claude reads your project's git history.

Your chosen model now stays locked across chat sessions, and AI reasoning gets an upgrade with the latest model versions.

Browser mode now has full access to all tools — files, shell, searching — so you can automate complex workflows without mode-switching.

Video seeking lets you scrub through recordings frame-by-frame, browser recordings auto-save with timestamps, and the CLI is now fully self-contained.

Agent Browser is now built into Nucleus with live previews, security settings are clearer and more powerful, and authentication stays fresh automatically.

New rolling 6-hour credit windows give you finer usage tracking, pricing plans are more accessible, and hotkeys no longer trigger while you're typing.

Pick your exact reasoning level for Claude, click-to-toggle voice recording, and enjoy smoother preview navigation with no lost terminal output.

Force a complete resync of your data from the cloud on demand, and push Claude to its deepest thinking level with a new max reasoning effort.

Account switching now refreshes all data, new users get their full history on first install, and update notifications follow you everywhere in the app.

A new extended-reasoning variant of Opus 4.7 is available for the toughest problems where you want the model to think longer.

Opus 4.7 is now the default model, preview toggles with a keyboard shortcut, and a quick-create button lets you start chats straight from any project.

Large requests are now automatically compressed, so long conversations with heavy context no longer hit size limits.

Agents now run read-only tools in parallel for significantly faster workflows, with locked models during runs and reliable message persistence.
