Veya

Veya is a cross-platform desktop workspace. It ships as a Tauri app on macOS, Windows, and Linux, and as a web beta at veya.sozenta.ai.

Three anchor ideas run through the whole surface. Each one is provable on the shipped app today:

Microsoft 365 Copilot ships on Windows. Notion AI runs in a browser. ChatGPT desktop is a chat window. None of them treat Linux as a real target — and office productivity on Linux has been a second-class experience for two decades. Veya treats Linux as a first-class platform from day one: same surface, same features, same release cadence as macOS and Windows.

The generative-UI claim is not aspirational. The canvases are real files on disk, editable in Settings > Catalog, each with the intents and functions it exposes:

Atlas is the work navigator across all workspaces. The demo install runs seven: AlphaFin, AtlasAlpha, AtlasBeta, BetaFin, MoveDstFin, MoveSrcFin, QA Playground — each with per-page word counts. Import from a zip or a folder; clean up empty pages; + New workspace to start another.

Agents ship with Run / Chat / Schedule on every entry. Install from Vera, import your own bots, or + New Agent from the Agents canvas.

An “Allow untrusted apps” toggle sits at the top of the Apps canvas. Every app carries Open / Pin / Sign in / Re-authorize / Sign out / Uninstall actions.

The right-panel activity log shows the actual prompt → tool_call → result chain, expandable per event:

"what events are on my calendar on July 20 2026?"
  → agent_tools_calendar_list_events_both { on: "2026-07-20" }
  ← 3 events

"search my calendar for Office"
  → agent_tools_calendar_search_events_both { q: "Office" }
  ← 2 matches

Every model call routes through your Vera deployment. Keys, quotas, and policy live in Vera; Veya carries the request. Frontier API today, on-device gemma next, your team's Ollama commons after that, your fine-tune after that — the application stack does not change when the model does.

Stacks — Veya's file canvas — shows what actually sits on the founder's disk right now. 4,919 files in total, broken down by kind:

~/Library/Application Support/Veya/veya-notes.json
~/Library/Application Support/Veya/docs.sqlite3
~/Library/Application Support/Veya/mail.sqlite3
~/Library/Application Support/Veya/photos.sqlite3
~/Library/Application Support/Veya/index.sqlite3

Open any of those in sqlite3. Grep the JSON. If Veya disappears tomorrow, your data does not.

Each section carries Research / Edit / Validate; the right pane shows the raw source.md with veya:section markers, so the file stays readable outside Veya. Every AI-initiated change surfaces an accept / reject / edit affordance before it touches the file. The model proposes; the user commits.