Needs a WebGPU browser — Chrome/Edge, or Safari 26.
For custom models. Optional.
1B for phones. Bigger needs a desktop GPU.
Your own MLC model — overrides the picker.
No download in progress.
~1–2 GB download. Desktop only.
Embeddings
Your local server handles embeddings on the same endpoint, with no API key — it's automatic. Just make sure an embedding model is loaded and named below.
Where documents & memories are embedded for retrieval — separate from chat, so you can chat on a provider with no embeddings API (e.g. OpenRouter) and embed here. Leave the endpoint blank to reuse the chat connection.
Models
Generation
8,192 tokens
4K8K16K32K64K128K
0.8
precisecreative
Notes
How Weavify responds to diary entries in Notes mode.
Read aloud
Voice needs WebGPU or Web Speech — first use downloads a small model.
On-device stays private & offline. The natural voice is clearer but sends the reply text to a cloud voice service.
Stored only in this browser; used only for the natural voice. get a key
Text-to-speech — nothing leaves this device.
ADVANCED — RETRIEVAL
The embedding model & endpoint live in the Embeddings section above. These two knobs control how retrieved snippets are selected each turn.
Appearance
Data
Helps Weavify recognize when a note or message is about you. Stays in your local config — never exported with chats.
Everything lives only in this browser (IndexedDB) — export a backup before clearing site data, switching browsers, or moving to a new device. Import restores from a file made here.
Danger zoneErase every conversation & its memory on this device.