Your canvas. Your agents.
Your next great idea.

Yes. The Verynize app will always be free. Any future paid offerings will be limited to optional cloud and AI services. You will not need a paid Verynize plan to use the app.

Yes. Deploying Verynize to your own cloud remains free, with no Verynize license or subscription fee. You only cover any infrastructure or AI usage charged by your chosen providers. The collaboration service has been tested locally; no production-hosted service is available here yet.

You can run the macOS app from the product source. Original Verynize code is MIT-licensed. An older locally signed app build was tested, but the package has not been rebuilt for all recent changes and there is no notarized public installer available here yet.

Canvas editing works independently of an agent. Agent tasks use your installed Claude or Codex and its existing account; provider access and usage charges are managed by that provider. One real reply-only turn has passed through each integrated runtime, including event streaming and MCP tool discovery. Results on your own coding tasks still depend on the provider and the task.

Switch to Code to work with repository chats, files, and project memory. New chats in repositories with a commit get their own Git worktree; older chats and uncommitted repositories can use the project root. Files, previews, and the agent follow the active chat. Send a canvas selection into an implementation task, then inspect diffs and commit or push from the Changes panel. This is an agent-assisted workflow, not a guaranteed design-to-source round trip. Real GitHub pull-request creation remains unverified.

Rewind can fork a provider session at a kept turn, with a transcript fallback. Second opinion sends the latest answer and diff to the other provider for read-only review. Both controls are implemented, but neither has been exercised live from inside the app yet.

The composer offers 31 named styles and twelve styled starter briefs, backed by design guidance and saved preferences. Style checks cover structure, contrast, and fonts; they do not guarantee the quality of generated designs. Preference changes have approval controls in Claude, while allowed Codex chats still rely on a guidance rule that has not been live-verified.

Agent-authored SVG illustrations, configurable image-provider adapters, and smooth bitmap-to-vector tracing are implemented. Real image-provider APIs still need live-credential validation. The agent design harness itself has no image-generation tool; it uses stock photos, SVG, shaders, and gradients.

Accounts, organization workspaces, room roles, shared cursors, and operation-based collaboration are implemented and tested locally. Publishing supports frozen design snapshots and versioned static apps with rollback. The collaboration service has not been deployed to production; live email, sign-in, and a separate apps domain remain unverified. Published apps are sandboxed without cookies or localStorage.

Previously opened cloud files can keep edits in a persistent outbox and send them when the desktop reconnects, including after an app restart. This was tested in Electron; the shared web implementation has not been separately exercised. Shared-image display while offline depends on the browser cache.

Figma REST import, clipboard decoding, .fig import, and OAuth are implemented against local fixtures. Real Figma OAuth and clipboard data remain unverified; encrypted .pen files are refused. URL capture supports cross-origin images, embedded fonts, region selection, and source mapping. Shadow DOM, iframes, pseudo-elements, animation keyframes, and off-viewport content remain outside capture support.