Stella, in detail.
A private, open-source desktop app that gives you one ongoing chat for your computer. Ask once, keep talking, and Stella figures out which agent, app, file, browser, model, or tool should handle the work.
The unusual part is not just that Stella can use your computer. It is that the desktop app itself can change. Stella can learn your preferences, adjust the interface, add workflows, and turn the app into something closer to your own operating space.
A desktop app, not just a chat box
Stella lives with your files, apps, browser, and local state, so it can help with the real work on your machine instead of only answering questions in a web tab.
You can use Stella for research, writing, spreadsheets, PDFs, Word documents, browser tasks, computer control, image generation, video and 3D workflows, media prompts, scheduling, reminders, dictation, realtime voice, and connected apps.
Those capabilities are table stakes now. Stella's bigger bet is that all of this belongs in one personal desktop app, one chat, and one interface that can keep adapting.
You keep talking in the same place
Most agent products make you choose a mode, start a new thread, pick a specialist, then remember where everything went. Stella keeps the top-level experience continuous.
Behind the scenes, Stella can split work into smaller jobs, run specialized agents, keep track of active threads, and bring the result back into the conversation. The point is simple: you should not have to become a project manager for your assistant.
What Stella can do
Use your computer
Inspect the screen, click, type, open apps, navigate windows, and work with what is actually in front of you.
Use the web
Browse, search, read pages, fill forms, and use browser context when it helps.
Work with files
Read, write, organize, summarize, and transform documents, spreadsheets, PDFs, presentations, images, and generated outputs.
Create media
Help make images, video, audio, 3D assets, small apps, games, mockups, and visual artifacts.
Listen and speak
Use in-app dictation, OS-wide dictation, read-aloud, and realtime voice. Wake-word activation is optional.
Run routines
Create reminders, recurring check-ins, scheduled work, and local automations from plain English.
Connect apps
Use supported services and messaging apps, including mobile, Discord, Telegram, Slack, Teams, Linq, Google Workspace, and Store-backed integrations.
Choose your model
Use Stella's managed provider, bring your own keys, use local models, pick OpenRouter-style options where supported, or select Claude Code as the engine.
Ways to reach Stella
Full desktop window
Chat, display, settings, history, Store, media, files, and everything else in one place.
Quick access
Capture, chat, add context, or start voice from the app or page you are already using.
Mini window
Keep a smaller Stella surface nearby for fast asks without taking over your screen.
Voice and dictation
Dictate into Stella, dictate into other apps when enabled, or talk to Stella in realtime.
Phone
Pair the mobile app with your desktop so your phone can message the Stella running on your computer.
Messaging apps
Message Stella from supported apps. Full desktop-powered execution depends on pairing, connection settings, and your desktop being available.
Local-first, with clear exceptions
Your normal desktop chat history, files, memories, generated local artifacts, and app state live on your computer. We do not keep your desktop conversations or files on our servers by default.
Some features need a backend: sign-in, billing, plan limits, managed model access, connected app setup, mobile pairing, Store catalog data, push notifications, and optional cloud features. The important boundary is that Stella does not need a cloud copy of your whole desktop life to work.
What we store
Account and billing records
Sign-in identity, billing profile state, Stripe customer and subscription references, usage credit records, and payment metadata needed to run paid plans.
Usage metadata
For managed model calls: owner ID, model, agent type, token counts, duration, success or failure, estimated cost, billing plan, and timestamps.
Anonymous limit counters
A salted hash of the device or client identifier, request count, first request time, and last request time. Current retention is seven days after last use.
Device and pairing metadata
Device IDs, device names where provided, platform, presence timestamps, mobile pairing records, pairing secret hashes, push tokens, and bridge registration URLs.
Connected app metadata
The minimum connection records needed to know which account is linked to which Stella user and provider. Some connection secrets are encrypted.
Remote delivery state
When you message Stella from a phone or connector, the backend may store request text, delivery metadata, request state, and routing info so the desktop can claim, cancel, complete, and deliver the work.
Optional cloud content
Cloud backups, Store publishing, social or collaboration surfaces, and other hosted features store the data required to provide those features.
What we do not store by default
- Your local desktop files.
- Your normal local desktop chat database.
- Your local memory markdown and runtime state.
- Your local provider API keys.
- A persistent copy of managed-model prompts and responses for the Stella Provider path.
- BYOK model traffic when the model call goes directly from your device to your provider.
Use Stella, BYOK, local models, or Claude Code
Stella has a simple path for convenience and a private-control path for people who want to bring their own providers.
Stella Provider lets you install the app and start using strong models without setting up accounts everywhere. Requests pass through Stella's infrastructure in transit so billing and limits can work, but prompt and response text is not persistently stored for that model path.
You can also add your own provider credentials, use local runtimes, and use Claude Code directly as the assistant engine. In those paths, Stella is acting as the desktop app and runtime you control, not as the model vendor.
The launcher starts a local runtime
The installed launcher handles setup, updates, recovery, and startup. The desktop app itself is a local repo-style runtime.
The launcher downloads the current desktop release archive and native helpers, writes the local environment file, creates a launch script, installs dependencies as needed, initializes the local Git state, and launches the desktop with bun run electron:dev.
That is intentional. Stella is open source, inspectable, and changeable. The app can update itself, but you can also inspect the repo, keep your local changes, and recover from bad self-changes.
How self-change works
When you ask Stella to change the app, an agent edits the local desktop code. Stella tracks the files involved, coordinates live updates, then applies visible renderer changes through Vite HMR when possible.
If a change affects routes, shell structure, config, dependencies, native helpers, or deeper runtime code, Stella may need a reload or relaunch. The morph overlay covers visible refreshes so the change feels intentional instead of like a broken page reload.
If a self-change breaks startup, the launcher can show a recovery view and, when the latest commit is an agent-authored self-change, offer an undo path.
A running changelog
Stella ships preview updates almost daily. The full log, grouped by date with new features and fixes, lives on its own page so you can scroll the history without leaving Learn More.
Stella is your desktop app.
Private, open source, one continuous chat, and flexible enough to reshape itself around how you work.
Get Stella