Changelog

Skills become installable from a shared registry, the Journey page gains a contribution graph that tracks your daily usage, and the Vault sidebar splits into six focused center panels you can open independently. On the reliability side, auth gets a diagnostic toolkit and a handful of edge-case fixes.

Skills Marketplace

Skills are no longer limited to what ships with Solo or what you write yourself. A new marketplace tab lets you browse, search, and install community skills from a public registry.

• Three-tab layout — the Skills section is now split into Installed, Marketplace, and Forks tabs. Each tab has its own empty state and list view.
• Registry search — keyword search with a suggestion banner above the composer helps you discover relevant skills without leaving your workflow.
• Install and uninstall — one-click install from the registry, with safety guards that prevent overwriting local edits. Uninstall cleanly removes the skill and its tracking metadata.
• AGENTS.md auto-generation — Solo now generates and maintains a workspace-level AGENTS.md file that indexes installed skills, with SKILL.md fallback for older formats.
• Fork tracking — installed skills carry a .solo-origin.json manifest so Solo knows where a skill came from and whether your local copy has diverged.

Usage heatmap

The Journey page now includes a GitHub-style contribution graph showing your daily activity over time.

• Daily activity store — a new solo-stats module records pushed commits and other actions into per-day usage buckets, synced to the server.
• Contribution graph — a heatmap component renders between the stats summary and leaderboard, with color intensity proportional to a computed usage score.
• Server hydration — the heatmap loads historical data from the server on mount so your activity is persistent across devices.
• Blurred future tiers — tiers beyond your current level are now blurred by distance with the progress bar removed, keeping focus on what you have unlocked.

Vault panels

The Vault sidebar sections have been promoted to standalone center panels.

• Six panels — Skills, Memory, Tasks, Current Vault, Plugins, and Connectors each open as their own center panel instead of being crammed into the sidebar.
• Find-or-focus — opening a panel that already exists brings it to focus rather than creating a duplicate. Panels are registered as singletons with BUILTIN_PANEL_TYPES constants.
• Shared header chrome — all six panels share a consistent header treatment so the experience is uniform regardless of which section you open.
• Sidebar as launcher — the Vault sidebar now acts purely as a launcher surface, dispatching panel opens rather than rendering content inline.

Auth diagnostics

A new diagnostic toolkit makes it easier to troubleshoot sign-in issues.

• auth:diagnose command — run bun run auth:diagnose to probe your auth setup from the terminal. The command checks token state, Cognito configuration, and callback URLs.
• Manual URL fallback — if the automatic OAuth redirect fails, a copy-URL button and retry action are now available on the sign-in screen.
• Diagnostic button in-app — a diagnostics button surfaces verbose auth tracing so you can see exactly where a flow is stalling.
• Token lookup fix — resolved a bug where a double github: prefix caused pending-token lookups to miss.
• File-first auth keys — auth-critical keys now prefer the on-disk file over shell environment variables, fixing flaky behavior in some terminal configurations.

Settings polish

All ten Settings tabs now use canonical uppercase section headings for visual consistency: AI, Account, Editor, Files, General, Plugins, Shortcuts, Skills, Terminal, and Voice.

UI refinements

• Scrollbar hider — a reusable .scrollbar-none utility is now available app-wide, applied to the commit message textarea and worktree setup commands.
• Hairline split divider — the panel split divider collapses to a 1px hairline with an invisible hit zone for a cleaner look.
• Rounded content pane — the main content area now rounds its left corners where it meets the sidebar.
• Traffic light spacing — macOS window controls get proper inset and a gap between the traffic lights and the sidebar toggle.

Fixes

• Git push now correctly calls bumpLocal (renamed from recordCommits) to update the local commit count.
• Bundled ONNX and Sherpa dynamic libraries now ship with the correct rpath in debug builds.
• Skills direct-edit tweak modal correctly flips the origin to fork.

A polish-heavy day. The Vault gets a full visual refresh, the sidebar sheds its heavy chrome for whitespace and flat tabs, the agent input dropdowns are rebuilt to match Codex's single-line style, and the design system consolidates around a single tokens pipeline — removing hundreds of lines of dead CSS in the process. Plus, Solo now remembers which chat you were in when you quit.

Session persistence

Solo now restores your last-active chat session on startup. Close the app mid-conversation and pick up exactly where you left off — no hunting through the session list.

Vault visual refresh

Every Vault component has been reskinned to use Lucide icons and the shared design token system.

• VaultPanel — icons, shadows, and spacing now pull from @solo/ui tokens instead of bespoke values.
• VaultEntryDrawer — updated to Lucide icons with token-based shadow treatment.
• VaultUnsortedTray — reskinned with Lucide and token utility classes.
• VaultDropZone — icons and border radius now use Lucide and token values.
• VaultEntryCard — card icons migrated to Lucide for visual consistency.
• VaultScopeToggle — scope toggle icons and shadow updated to the shared set.

Sidebar breathing room

The sidebar continues its march toward calm. Heavy borders and boxes are gone, replaced by whitespace and subtle affordances.

• Ghost-style toolbar — toolbar buttons and action pills are now ghost-styled, reducing visual weight.
• Flat tab underline — the Dev/Vault pill box is replaced with a flat tab underline that takes up less space.
• No more bordered repo card — the bordered card wrapper on the repo header is stripped; whitespace does the grouping.
• Sessions breathe — the outer sessions box is removed, letting content flow without a container.
• Collapsed worktree groups — all worktree groups collapse by default so you see what matters first.
• Cleaner worktree headers — the "Grouped by worktree" subtitle and branch-name text are removed; the icon is enough.
• Visible drill-in affordance — worktree drill-in now shows a tree + chevron chip so the click target is obvious.

Agent input dropdowns

The mention (@) and slash (/) dropdowns have been rebuilt from scratch in the Codex single-line style.

• Single-line layout — both dropdowns now render entries in a compact single-line format instead of multi-line cards.
• Menu token alignment — dropdown styling matches the @solo/ui Menu component tokens for radius, padding, and color.
• Concentric radii — nested rounded corners use mathematically concentric radii so inner and outer curves feel harmonious.

Design system consolidation

The CSS layer got significantly thinner.

• 583 lines removed from index.css — dead rules, overrides, and hand-rolled tokens are replaced by @solo/ui/styles.
• Single tokens.css import — styles.css in the UI package now re-exports a single tokens.css file that drives the entire Tailwind theme via @theme inline.
• Select shim — the local select wrapper now delegates to the @solo/ui component.

Keyboard shortcuts

• Cmd+B toggles the sidebar — a standard shortcut, now wired up and registered in the keybindings map.

Fixes

• Dialog and AlertDialog shadow fallback — replaced an undefined shadow-glass variable with shadow-xl so dialogs render their shadow correctly.
• Dialog auto-close button restored in the local shim.
• Settings NumberInput rewritten as a segmented control to prevent value overflow on small screens.

A big round of agent surface work. Skills become first-class citizens inside the chat, the mode pill grows a fourth state with a dedicated Debug mode, project settings and plans move to an on-disk store under .solo/, and the sidebar finally reflects git state in real time. Most of this is plumbing you will feel rather than see.

Skills

Skills are callable behaviors that live alongside your project and can now be referenced directly from chat. They work across agent runtimes.

• Inline skill chips — type / anywhere in the input and pick a skill from the slash menu. It is spliced into your message as a SkillChip decorator node instead of plain text, so the chip is part of the message itself rather than a floating attachment.
• Source badges in the picker — every entry in the slash menu shows where the skill comes from so you know whether you are calling a project-local skill, a user-scoped one, or one from an adapter-contributed catalog.
• Claude and Codex adapters — skills resolve through polyglot adapters for both Claude and Codex agents. When the same skill is contributed by multiple sources, a source-priority dedup step keeps a single entry in the catalog.
• Ordered parts at submit — messages now capture the interleaving of chips and text, so a message like "refactor with /tidy then /test" is preserved in order on both the send and recall paths.
• First-launch onboarding — a SkillsOnboardingDialog opens the first time you start a new install, giving a tour of where skills come from and how to turn specific sources on or off.
• Settings → Skills — a dedicated Skills tab in Settings (Zap icon) lets you toggle sources and browse the merged catalog without hunting in config files.

Mode pill, now with Debug

The mode cycle has been rebuilt around a proper permission pipeline.

• Four states, not three — the old Plan / Agent / Ask cycle is now a 4-state PermissionMode that adds a dedicated Debug mode.
• Debug mode — surfaces tool decisions before they run, so you can see what the agent is about to do and why. The bridge pre-decides tool status and re-evaluates pending prompts whenever the mode changes mid-run.
• Shared permission pipeline — solo-core ships a tier-table + rule-parser permission pipeline; agent-bridge mirrors the exact same pipeline in TypeScript so the UI and the runtime never disagree about whether a tool is allowed.
• Preview decisioning — previewPermission lets the UI know what a tool call would resolve to without committing to running it.
• Mid-run mode toggle — switching modes while the agent is working no longer cancels the turn; the in-flight prompt is re-evaluated under the new rules.
• Session goal capture — the bridge now emits a session_goal_captured event so the surface can show what the agent thinks the turn is about.

Project settings on disk

Settings and plans now live in your project, not just in memory.

• .solo/settings.json — a real on-disk settings file with a loader that merges user, project, and local scopes. Frontend subscribes via useSoloSettingsStream and the soloSettingsStore Zustand store.
• .solo/plans/ — plan files are atomically written to the project root rather than a global path, so each repo owns its plans. The plans module in solo-core handles the I/O; plan_commands wraps it for the frontend.
• IPC for all of it — new settings_commands, plan_commands, and debug_mode Tauri handlers expose the surface to the UI, with a settings:changed broadcast event when the file changes on disk.

Git

The sidebar now tracks git without you asking.

• .git/ ref watcher — a new debounced GitRefWatcher in solo-fs installs on the workspace root and emits a change event whenever refs change. Fetches trigger an immediate post-op refresh so the UI reflects the new ahead/behind count right after the network returns.
• Live "commits ahead" on the repo card — the sidebar repo card reads the live commitsAhead value for the active repo instead of a snapshot.
• Stale fetch invalidation — the git store now invalidates in-flight fetches that were superseded by a newer one and widens the refresh that runs after local ops.

Queued messages

Input is no longer blocked while the agent is streaming.

• Per-session queue — every session has its own queue of pending messages; type while the agent is running and your message enters the queue instead of getting dropped or forcing you to wait.
• QueuedMessagesStrip — a pill above the input shows what is queued, with quick actions to edit or drop a queued message.
• Interrupt tidy-up — interrupting a turn now also closes the task overlay tied to that turn, so stopping the agent leaves the surface clean.
• Unified Escape/stop — Escape and the Stop button route through a single ChatInputContainerHandle so every "get me out of here" gesture behaves the same way.

Layout and chrome

• Collapsed agent tabs — inactive agent tabs collapse to pills and expand on hover, with a pulse animation when a collapsed session needs attention.
• Titlebar is fully draggable — the WorkspaceSwitcher has been removed from the titlebar and the whole header is now a drag region.
• Icon pass — a sweeping migration swaps bespoke icons for Radix/Lucide across the sidebar, settings, source control, welcome screens, markdown renderers, and dialogs. The visual set is now consistent end-to-end.

Fixes

• RepoCard reflects the live ahead/behind for the active repo instead of a cached value.
• Stale git fetches no longer clobber newer ones.
• The task overlay no longer lingers after its source turn is interrupted.

A batch of UX and editor refinements focused on making the agent workspace quieter, the sidebar calmer, and the editor feel more like home. No new APIs — every change here is about getting out of the way while you work.

Agent chat

The chat surface now borrows its flat layout from Codex and uses it consistently across tool calls, intermediate traces, and task lists.

• Codex-style flat tool cards — tool invocations render as wider, reading-optimized cards with a neutral secondary surface instead of stacked bubbles.
• Collapsed intermediate trace — the step-by-step trace now folds behind a compact "Worked for Xm Ys" header so finished turns stay scannable.
• Sticky tasks pill — an active todo list now pins just above the input, and the inline TodoToolWidget is automatically suppressed when the sticky overlay owns that list. No more duplicate task views.
• Neutral tool widget surfaces — tool widgets adopt the shared Orbit motion and neutral surface treatment. Expand/collapse regions now share one ExpandRegion primitive and one motion constants file, so animations match everywhere.

Sidebar

The left rail gets a round of polish aimed at reducing jitter and making worktrees easier to live with.

• Toggleable branch switcher — swap branches directly from the sidebar, and right-click any session for a context menu of actions.
• Calmer worktree rows — worktree names render smaller by default and expand to full width only when the row is hovered, keeping long names out of the way.
• No more width-shake — releasing the drag divider no longer snaps or jitters the sidebar width.

Editor

Monaco's find widget is finally behaving, and markdown now renders through a single pipeline.

• Find widget stays on-screen — the Monaco find widget is no longer clipped at the top of the editor.
• Tooltips overlay the breadcrumb — the editor status bar is gone; find-widget tooltips now float cleanly above the breadcrumb bar without being cut off.
• Preview mode uses Streamdown — markdown preview now shares the same MarkdownPreview component as the rest of the app, so streamed and static markdown render identically.
• File explorer: single-click to open — one click opens files and toggles folders; no more double-click gymnastics.

Platform and polish

• Global zoom with Cmd+/– — zoom level is persisted across sessions via the new useAppZoom hook, wired into the Tauri set-webview-zoom capability.
• Bundled Geist fonts — Geist is now shipped with the app and the entire react-markdown stack has been removed in favor of Streamdown. Faster cold starts, one fewer dependency tree.
• Compact icon buttons — IconButton sizing and radius were tuned down to feel less heavy in dense toolbars.
• Theme nits — secondary buttons drop their extra border, and the worktree action pill gets its intended background.

Fixes

• macOS keychain auth — authentication now shells out to /usr/bin/security directly for keychain operations, fixing a class of failures on some macOS configurations.
• Chat feed alignment — the message feed now reserves vertical space equal to the sticky tasks pill height, so the last message never sits under the pill.
• Editor warnings — silenced a duplicate-extension warning by disabling StarterKit's bundled Link extension in favor of the project's own.

Version 0.1.0 is the first release that treats the worktree as a first-class workflow. Instead of bouncing between the agent, the diff view, and your shell, you can now plan a change, run it against its own worktree, review the diff with AI assistance, and push it — all without leaving the window.

Plan mode, rebuilt

Plan mode is no longer just a toggle on the agent's behavior. It now has its own persisted artifact and a dual enforcement mechanism.

• Plan file tracking — the agent writes its proposed plan to an actual plan file that you can read, edit, and re-run against. No more losing the plan to a scroll event.
• Dual enforcement — plan mode is enforced both in the UI (the mode pill) and by the agent runtime, so the two can't drift out of sync.
• Synced UI mode state — switching modes mid-conversation now updates the plan file and the mode pill atomically.

Agent chat, overhauled

The agent chat picks up real feedback signals, not just streamed text.

• Loader and running timer — every turn shows a live indicator and the elapsed time so you know the agent is working, not hung.
• Clarifying questions surface inline — the agent can stop and ask a question, and the question renders as a first-class element in the thread rather than being buried in prose.
• Mode cycling — cycle between Plan / Agent / Ask modes from a single control above the input.

Worktree workflow

The worktree sidebar graduates from a read-only list into an actual workflow surface.

• Push from the sidebar — kick off a push to the remote from the worktree changes view without dropping to the terminal.
• AI review on your diff — launch a dedicated review session for the current diff. The agent reads your changes and produces a review with line-level feedback before you push.
• Review session launcher — a single entry point to start a code review against the current set of changes, scoped to the worktree.
• Git staging and commit UI — stage individual hunks, write commit messages, and advance through file tabs inline.
• RepoRail overflow fix — long repo paths no longer push the sidebar out horizontally.

Welcome screen

The first 500ms of the app now feels intentional rather than empty.

• Splash gate on launch — the splash screen now waits until you've picked a project instead of auto-dismissing on a 5-second timer. Your first interaction is always the project picker.
• Smooth recent-projects animations — recent projects fade and stagger in rather than popping.
• Stars background — a subtle animated starfield behind the welcome screen, plus a round of sidebar visual refinements.

Installation

• Signed macOS DMG — v0.1.0 ships as a signed .dmg built via GitHub Actions. Apple codesigning is gracefully skipped in environments without a certificate configured.
• TypeScript bindings generated in CI — protocol bindings are generated before the desktop build so the frontend always typechecks against the Rust contract that actually ships.

Fixes

• Desktop auth config now works against Supabase-built binaries.
• Settings migrated from ad-hoc storage to the SkillInfo IPC type pipeline.
• Miscellaneous TypeScript build errors resolved with a bindings barrel.

Solo IDE now supports the latest generation of frontier language models across all major providers. Every model is available through the same unified interface — switch between providers without changing your workflow.

Updated Model Support

Anthropic

• Claude Opus 4.6 — Maximum capability for complex reasoning and multi-step tasks
• Claude Sonnet 4.6 — Balanced performance with improved speed and accuracy

Google DeepMind

• Gemini 3.1 Pro — Enhanced reasoning and extended context over Gemini 3 Pro

OpenAI

• GPT-5.3 — Latest flagship model with improved instruction following and code generation

What This Means

All workspace tabs — Ask, Agent, Browser, and Canvas — can leverage these updated models immediately. Select your preferred model from the model picker and start building.

Previous model versions remain available for compatibility.

This release establishes Solo IDE as a comprehensive Unified Development Environment — a single workspace designed to eliminate context switching entirely. Built on Electron for native performance, version 0.0.2 introduces four specialized workspace tabs and support for the latest frontier models.

Workspace Tabs

Ask Tab

A conversational interface optimized for ideation and exploration.

• Ideation workflows — explore concepts without modifying your codebase
• Read-only tooling — designed for research, documentation queries, and architectural discussions
• Isolated context — no side effects on your active project

Agent Tab

An autonomous coding assistant with three distinct operational modes:

• Plan Mode — The agent analyzes requirements and proposes an implementation strategy before generating code
• Accept Mode — Streamlined execution with approval checkpoints for rapid development
• Critique Mode — Multiple language models review code simultaneously, providing consensus-based feedback

Browser Tab

A fully integrated Chromium instance within the development environment.

• Local development — Test applications without switching windows
• DevTools integration — Full inspector access, visible to the AI agent
• Playwright support — Automated end-to-end testing directly from your workspace
• Web context — Research and reference documentation without interrupting your workflow

Canvas Tab

A visual interface for designing and orchestrating complex development workflows.

• Component library — Reusable UI elements with drag-and-drop composition
• Animation editor — Visual tools for designing micro-interactions
• Contextual notes — Capture and organize thoughts alongside your code
• Multi-agent coordination — Deploy multiple agents with distinct tasks and real-time oversight
• Live editing — Modify running workflows without restart

Model Support

Solo IDE provides unified access to frontier language models through a consistent interface.

OpenAI

• GPT-5 — Latest flagship model
• GPT-5.1 (mini / nano / full) — Optimized variants for different latency and capability requirements

Google DeepMind

• Gemini 3 — Multimodal capabilities with broad task coverage
• Gemini 3 Pro — Extended context and enhanced reasoning

Anthropic

• Claude 4.5 Haiku — Optimized for speed and efficiency
• Claude 4.5 Sonnet — Balanced performance across diverse tasks
• Claude 4.5 Opus — Maximum capability for complex reasoning

Additional model integrations are in active development.

Technical Foundation

• Electron runtime — Native performance with cross-platform compatibility
• Monaco editor — VS Code-class editing with full language server protocol support
• Unified state — Seamless context sharing across all workspace tabs

Roadmap

Upcoming releases will expand Canvas workflow capabilities, enhance browser automation, and introduce support for custom model configurations.

We're excited to announce the first beta release of Solo IDE - your AI-powered development workspace that combines intelligent assistance with professional development tools!

AI Integration

Solo IDE features deep integration with Claude SDK, bringing advanced AI capabilities directly into your workflow.

The platform supports:

• Multimodal interactions for text, code, and visual content
• Built-in tool calls for executing functions
• Multi-session SDK support for managing multiple conversations
• Seamless interaction with external tools and services

Development Environment

At the core of Solo IDE is the Monaco Editor, providing a professional VS Code-style editing experience. The environment includes:

• Syntax highlighting for multiple programming languages
• Fully functional terminal for command execution
• Complete file operations system
• Professional code editing features

Workspace Modes

Solo IDE offers three distinct operational modes to match your workflow:

• Plan Mode - AI helps you strategize your implementation approach
• Accept Mode - Review and approve AI-generated code suggestions
• Default Mode - Standard development workflow

All three modes are fully functional in this beta release.

Specialized Pages

Canvas Page (Beta)

A powerful visualization tool for designing and planning your projects:

• UI components and interfaces
• Frontend workflows and user journeys
• Backend architecture diagrams
• System design and infrastructure planning

Chat Page

A dedicated space for conversations and brainstorming with AI, separate from your coding workspace. This allows you to explore ideas and get assistance without cluttering your development environment.

Authentication & Security

Solo IDE implements secure OAuth integration with intelligent token management. The system reads credentials directly from system storage, eliminating unnecessary API calls while improving both security and performance.

What's Next

This is our first beta release, and we're actively developing new features and improvements based on user feedback. Join us in shaping the future of AI-powered development!