Browse the docs
Desktop App & Local Mode
The Sidekick Forge desktop app can run parts of your sidekick on your own computer: its thinking, its voice, its hearing and a copy of its knowledge. In Local Mode, conversations stay on your machine, and you can mix local and cloud however you like.
On this page
Why run locally? Getting the desktop app Local, hybrid and cloud What can run locally Hardware and the "not ready" screen Switching minds in chat "This computer" for other abilitiesWhy run locally?
- Privacy: with thinking, hearing and speech all set to local, the conversation never leaves your computer.
- Cost: local turns don't use cloud model credits.
- Choice: keep the parts you care about in the cloud (a voice you've paid for, say) and run the rest locally.
Local Mode is optional. Every sidekick always has a cloud setup, and phones, tablets and the website always use the cloud.
Getting the desktop app
The desktop app is available on request. Email help@sidekickforge.com and tell us which operating system and graphics card you have. Once it's installed, sign in with your normal Sidekick Forge account. Your sidekicks, knowledge and conversations are all already there.
Local, hybrid and cloud
On the Sheet's Mind and Body tabs, each part of your sidekick has places it can run. Switch on the places you want, then pick what runs there:
| Place | Label in the app | Where your data goes |
|---|---|---|
| Cloud | Runs anywhere — phone, tablet, browser | To the cloud provider for that part. |
| Hybrid (thinking only) | Easy turns here, hard ones in the cloud | Light turns stay on your computer. Harder turns are escalated to a cloud model and leave the machine. |
| Local | Stays on this machine — nothing leaves it | Nowhere. It runs on your computer. |
Each part is set independently. For example, you can think locally, hear locally and still speak with a cloud voice.
Local choices in the Sheet show This device only — needs the desktop app and can only be set up from the desktop app. When a model isn't installed yet, you'll see Open this Sidekick in the desktop app to download the model this needs.
What can run locally
| Part | Local option | Notes |
|---|---|---|
| Thinking | Local Gemma 4 and Local Qwen 3.8 Mind Modules, plus Hybrid Gemma 4 → Qwen 3.8 | Downloaded by the app the first time. Bigger models need more graphics memory. |
| Speech | Local Chatterbox | Speaks with a voice cloned from a short sample. Voice cloning happens in the desktop app. Needs Docker with the NVIDIA runtime. |
| Hearing | Local Whisper | Transcribes on your computer and detects the spoken language by itself. |
| Knowledge | Knowledge mirror | A local copy of the knowledge base, so document search can run on your computer too. |
Hardware and the "not ready" screen
Local models need a capable graphics card. A recent NVIDIA GPU is strongly recommended. Before a local conversation starts, the app checks that everything is downloaded and that there's enough free graphics memory. If not, the chat shows Not ready to run here yet with these options:
- Download: fetch the missing model.
- Run in the cloud instead: use the sidekick's cloud setup for this conversation. This is the quickest way to keep going.
- Try again: check again, for example after you've closed a game or video editor.
- Run here anyway (it will be very slow): run it locally anyway.
If your graphics card is busy, the message says how much memory the model needs, how much is free, and which other programs are using it.
Switching minds in chat
On the Mind tab, star the Mind Modules you use most. Your favourites appear in a quick switcher inside the chat in the desktop app, so you can move a conversation between local and cloud without opening the Sheet.
"This computer" for other abilities
Some abilities can use the computer the desktop app runs on:
- Browser Operator can use a private browser on a computer you've paired. See Abilities.
- Code Catalyst can work in Folders on this computer, not just on remote servers.
Both are admin-only abilities and ask before risky steps.