Using Xome

Models & brains.

Xome runs on the model you choose, a cloud API with your own key, or one running entirely in your browser.

Cloud models (bring your own key)

Add a key in Settings → Active brain (or during onboarding). Supported providers: Claude (console.anthropic.com), GPT (platform.openai.com), and Gemini (aistudio.google.com). The key is stored in your browser's IndexedDB and attached per-request through a stateless proxy, it never lands in a server database or log.

Switch the active model anytime from the model menu above the composer; each conversation can also override the provider/model just for itself.

On-device models (WebGPU)

Private, free, and offline-capable, the model runs in your browser via WebGPU (Chrome/Edge 113+). The download is cached after the first load. Smaller models are chat-only; the catalog marks which ones handle tools:

Model	Size	Tools
`Qwen2.5 3B Instruct`	~1.9 GB	Yes, recommended default
`Hermes 3 (Llama 3.2 3B)`	~2.0 GB	Yes, best tool reliability
`Qwen2.5 7B Instruct`	~4.7 GB	Yes, most capable, needs a good GPU
`Llama 3.2 1B / Gemma 2 2B`	~1–1.4 GB	Chat only

Small local models get a lexically pre-filtered subset of tools per message (instead of the full registry) to keep prompts fast. Long conversations on small contexts benefit from /compact, see Chat features.

Token usage

Cloud usage is tracked per conversation and shown in History (e.g. “· 12.4k tokens”).
Billing is entirely between you and your model provider, Xome never proxies your account.