Local AI
The Context Graph captures your browsing structure. The Local AI is what makes it intelligent. Quira ships with a local LLM that runs entirely on-device. It powers page summarization, entity extraction, embedding generation, and Emost importantly Enatural language queries against your Context Graph. No data ever leaves your device by default (see AI Privacy Levels).
Model options
| Config | Model | Parameters | Min RAM | Use Case |
|---|---|---|---|---|
| Default | Phi-3.5 Mini | 3.8B | 8 GB | Standard summarization, entity extraction, NL Query |
| Optional | Llama 3.1 8B Q4 | 8B (4-bit) | 16 GB | Higher quality responses, deeper reasoning |
Runtime by platform
| Platform | Runtime | Acceleration |
|---|---|---|
| macOS (Apple Silicon) | MLX | Metal GPU |
| macOS (Intel) | llama.cpp | CPU |
| Linux | llama.cpp | CUDA / Vulkan / CPU |
AI Shield privacy indicator
- Green Shield (LOCAL) EAll AI processing on-device. This is the default.
- Purple Shield (CLOUD) ECloud AI is active (opt-in only). Heavy tasks offloaded to cloud.
First-run model download
- Launch Quira for the first time EA model download prompt appears with estimated size (~2.3 GB) and time.
- Background download EProgress bar with percentage and ETA. You can browse normally during the download.
- Ready notification E"Your local AI is ready" Eall AI features are now active.
Was this page helpful?