quite.spaceThe assistant in quite.space is not a button and not a panel but an ordinary space with one interlocutor who answers. It is local: it is not announced on the network, not published to a relay, not listed in any catalog, and not synced to other devices.
It appears in the navigator under AI once a provider is configured.
Setting it up
Settings → AI.
| provider | what is needed |
|---|---|
| Claude | a model and an API key |
| OpenAI | a model and an API key |
| Custom | a compatible address, a model, a key |
| Local | the address of a local server (Ollama, for example) — no key needed |
The Test connection button checks the link before you write anything.
The key is stored on this device and goes only to the provider you named.
What exactly leaves the machine
This is written right under the input line, not tucked away in the settings — because it is what you need to know before you start typing:
Quite AI sees the last 20 messages of this space and nothing more. Every question sends that window anew to
<provider> · <model>over the internet.
For a local model the phrase is different and just as precise: everything goes to the model on this machine; nothing leaves the device.
Two properties that do not change
AI does not write to the shared journal. The assistant is a separate participant with its own signing key, and its answers appear as the assistant’s answers, not as yours. It cannot pass itself off as you.
AI proposes — you publish. The model can prepare a post draft (posts) or a space’s appearance, but only a human publishes.
When something goes wrong
- No provider configured — the app refuses before sending, so that the journal is not left holding a question no one can answer.
- The provider returned an error — it is shown next to the question, locally, with an “ask again” button. Your provider’s error is a diagnostic of your device, not a space event, and it does not enter the journal.
- A restart with a question in flight — the question stays unanswered, and that is what it will say. That is the truth, not a failure.
What can be sent to the assistant
Anything that can be forwarded: a message or a post can be forwarded to the AI like to any other place (forwarding). The forwarding comment is in that case offered as an instruction: “What should the AI do with this?” — an ordinary human message, not a hidden prompt.