All releasesVelork logo

Velork Alpha 0.1.7

A reliability, UX, and tooling release: auto-update fixes across all platforms, a reworked AI oversight model, subagent drill-in, terminal sessions in the fullscreen panel, and a new MCP server for using Velork's design tools from external AI clients.

Fixes

  • macOS auto-update "app is damaged": a packaging bug was corrupting the app bundle's symlinks during zip-based updates; the archive now preserves them correctly so Gatekeeper accepts the update
  • Silent update notifications: the "Update ready" toast had been accidentally disabled during a debugging session and is now reliably shown again
  • Linux auto-updates: Linux clients were requesting a platform key that didn't exist on the server, so updates silently never appeared
  • Same-version hotfixes: if a fix ships under the same version number with a new build, the updater now correctly offers it instead of treating it as already up to date
  • Windows / PowerShell terminal: fixed AI-run commands failing to detect completion or capture correct output
  • Billing "Upgrade" button: no longer shown to users already on a Team plan
  • Tool-call sequencing: fixed certain tool-call sequences (especially with OpenAI-compatible models) getting out of sync and causing a response to fail outright
  • Network hiccup recovery: brief drops during a long AI response (laptop sleep/wake, wifi switching, VPN reconnects) are now automatically retried instead of failing the whole response
  • Conversation memory after reload: for Claude and Cursor-powered chats, reloading the window or restarting the app no longer resets the AI's context even though the history was still visible
  • Cursor-provider tools: shell commands, file search, and subagent delegation no longer silently fail; they fall back to Velork's own execution path when needed
  • Canvas persistence on reload: shapes, text reflow, and images now persist correctly across window reloads
  • Canvas pan/zoom on Windows and Linux: now as smooth and responsive as on macOS
  • Chat/Task Board performance: reduced background overhead during rapid multi-step AI work
  • Subagent sessions in sidebar: child sessions from agent tasks no longer appear as independent top-level entries; they are only visible by drilling into the parent card
  • Accept/Undo file spinner: the editor overlay now shows a busy spinner and disables controls while an accept or undo operation is saving, preventing double-clicks
  • Chat panel initialization loop: fixed a layout re-initialization cycle that made the editor progressively slower the longer the chat panel had been open
  • Sidebar toggle blocked by fullscreen: you can now toggle the primary sidebar at any time without having to exit fullscreen first
  • Tool activity card reflow: the collapsed "Explored N files" rows no longer jump or rebuild the entire chat list on every streaming update — the list updates incrementally

Improvements

  • The AI uses subagents far more aggressively for research, codebase exploration, and independent multi-part tasks; click any subagent card to open its own session
  • Clarifying questions now use an interactive prompt with a recommended option instead of guessing silently or burying questions in a paragraph
  • Removed generic AI filler phrases ("Sure!", "Happy to help!") in favor of direct, teammate-like communication with short status updates before long batches of work
  • New reasoning effort control: pick Low / Medium / High / Extra-High directly from the model picker for models that support extended thinking
  • Tool activity summaries now group by category (file reads, searches, shell commands, web lookups, design reads, skills) so you can see what the AI was doing without expanding every card
  • Accept All / Undo All only appear when there are multiple files pending review, removing the redundant button when there's only one file
  • Your account avatar (photo or initials) now appears in the activity bar when signed in, replacing the generic accounts icon
  • Title bar layout reorganized: sidebar toggle and mode tabs now stack vertically, giving the tabs more room

New

  • Approval selector: choose AI oversight level — "Ask for approval", "Approve for me", or "Full access" — directly from the chat input
  • Plan usage banner: a banner appears above chat when you're running low on (or have used up) your plan's request allowance, with a quick upgrade link
  • Terminals in fullscreen panel: open and manage full terminal sessions alongside chat sessions in the fullscreen view; includes a "New terminal" shortcut (⌃⇧`) and per-terminal close button
  • Subagent drill-in: click any subagent card in the transcript to open that agent's session in a read-only view with a breadcrumb to navigate back to the parent conversation
  • Velork MCP server: a local MCP proxy lets Claude Code, Claude Desktop, OpenCode, and other MCP clients use Velork's design canvas tools (get_pages, design_add_layers, design_batch_layers, and more) directly from their own interfaces
  • Sidebar toggle button in the title bar for quickly showing/hiding the primary sidebar

Install

Download from velork.com/download, or update in-app once you see the "Update ready" notification.

July 2026. The Velork team.