In-app Local AI assistant
Glitr can help you on this device: answer questions from the product docs, propose a navigation or settings change for you to confirm, and walk you through pairing, remotes, and first messages.
Local AI is available in the browser and desktop apps after you load a model. It is not a cloud chatbot: turns stay in-process. Model weights and the docs index may still download over the network the first time.
Open the assistant
- Tap the sparkles button at the top right.
- If no model is loaded, pick one and wait until it finishes downloading.
- Ask a question, or use a Get started chip (Pair with a peer, Connect remote, Set up Local AI, and similar).
You can also open the full-screen Local AI page from the nav / speed dial (/ai). Both surfaces share the same conversation.
A sky dot on the sparkles button means a proactive tip is waiting. An amber dot means an action needs Confirm or Dismiss.
Ask about Glitr
Ask in plain language (“How do I pair?”, “Where do I attach a git remote?”). Answers can cite docs as numbered chips — tap a chip to open that page on glitr.io.
When the assistant proposes a change (open Sync settings, start a guide, open a chat), you get a Confirm / Dismiss card. Nothing runs until you confirm.
Guided journeys
A journey is a short checklist the app drives for you:
| Chip / journey | What it helps with |
|---|---|
| Pair with a peer | Talk now — Add contact → Pair now → share QR or paste |
| Add git contact | Add a contact by peer git URL |
| Connect remote | Connect repository / Git remotes |
| First message | Open a contact chat and send |
| Pair a device | Another device |
| Troubleshoot | Troubleshooting |
| Set up Local AI | Load an on-device model |
After you confirm Guide me (or Confirm on a start_journey card), the app can navigate, open dialogs, and generate a pair invite on its own. You still type codes, enter credentials, and tap Send. The assistant opens when it is your turn, and a tip can point at the next control.
Use Skip / Stop on the guide card to leave the journey. You can always follow the manual how-tos linked above instead.
Privacy
- Generation runs on this device after the model loads — not via a remote LLM API.
- To guide you, the assistant may use app context on this device (where you are in the UI, contacts/chats summaries, and recent messages in an open thread). That context is not uploaded to an LLM provider.
- The Local AI thread is not synced through your git mailbox.
- First-time model download and refreshing the docs index still need the network.
Technical wiring: Local AI.
Next
- Talk now — live Pair now without a git URL
- Your first conversation — choose a track
- Troubleshooting — when delivery or pairing stalls