Persistent Desklets
Workspaces that keep conversations, tasks and saved memory together. Background agents and file understanding are on the roadmap.
Desklet · personal AI desktop
Desklet turns AI from an endless chat window into a workspace. Persistent Desklets keep your conversations, tasks and saved memory together — so the work continues instead of starting over. Background agents and file understanding are coming later.
Desklet is free in early access. AI model usage is charged separately by your provider.
Workspaces that keep conversations, tasks and saved memory together. Background agents and file understanding are on the roadmap.
Connect OpenAI, Anthropic, Gemini or OpenRouter. Keys are encrypted on the server and never returned to the browser.
Editorial typography, warm paper tones and no notification theatre. The work stays in front of you, not the interface.
Inside Desklet
Keep projects, notes, conversations and next steps together on one desk.

How it works
No setup ceremony. Open a workspace, give it your material, and it stays useful every time you come back.
Start a workspace for a project, a client, a research thread or a side business. It stays put and keeps its own context.
Keep reference notes and workspace memory together. File understanding and link ingestion are planned for a later iteration.
Chat with saved memory and track tasks beside your project. Background agents are planned, not running yet.
Use cases
One Desklet per project, client or part of your life. Here is how that plays out in practice.
Founders and operators
Keep investor updates, pricing experiments, hiring notes and metrics in a single Desklet that already knows your company.
Researchers and analysts
Keep research notes and saved memory together, then continue the conversation with that context. Document understanding is planned.
Writers and studios
Keep style rules, character notes and outlines in memory alongside your writing conversations, so each session can pick up the thread.
Consultants and freelancers
Keep conversations, tasks and memory separate per engagement. Switch clients without losing the thread.
Engineers
Save architecture notes and decisions in workspace memory to carry context into your next chat. File understanding is planned.
Everyone else
Travel plans, home projects, applications and family logistics — each one a small, calm Desklet you can return to.
Capabilities
We label everything honestly. Nothing marked available is a mock-up, and nothing coming soon pretends to run yet.
Persistent workspaces with overview, chat, files, agents, tasks and memory in one view.
Route any conversation to OpenAI, Anthropic, Gemini or OpenRouter models through one interface.
Provider keys are encrypted at rest server-side, masked in the UI, and never exposed to the browser.
Facts you record are injected into every conversation in that Desklet, so context survives the session.
A workspace file list is visible today; uploading and using file contents in chat are planned.
Jump between Desklets and actions with ⌘K, without ever reaching for the mouse.
Lightweight tasks attached to the workspace they belong to, visible next to the work itself.
Long-running workers that research, monitor and draft in the background, then report back to the Desklet.
Recurring routines — morning briefs, weekly digests, watchlists — that run on their own schedule.
Google Drive, Notion, GitHub, Gmail and Calendar as first-class sources inside a Desklet.
A sandboxed environment where agents run code, transform files and produce real artefacts.
Search meaning across every Desklet, file and conversation you own, not just exact words.
Curious about the future of agents? See the AI agent workspace.
Roadmap
An AI computer is built in layers. This is the order we are building them in.
The calm foundation: persistent workspaces, multi-model chat and secure BYOK.
Work that continues when you close the tab.
A real computer behind the desk.
FAQ
Straight answers about models, keys, privacy and pricing.
Create an account and bring your own model key. Desklet is free in early access; your AI provider bills you separately for model usage.