LLM & reasoning architecture

Mystilink supports two LLM access modes that can coexist:

Mode	Description
Newton LLM	Mystilink’s in-house LLM platform: sign in, select Newton models, inference on the Newton platform, prepaid balance billing
BYOK	Bring-your-own vendor API keys: keys stay local; requests go directly from your device to OpenAI, Anthropic, etc.

Setup: Newton → Newton LLM; BYOK → LLM providers. Step-by-step: Official API keys, OpenAI-compatible providers, Local LLM deployment. Connection issues: Model connection.

Two layers with metaphysics theories

Your question
    ↓
Metaphysics theory (BaZi / Tarot / Auto …)  →  charting, inquiry, draw/cast
    ↓
LLM (Newton / GPT-4o / Claude / …)         →  understand, plan, write reading

Where	Controls
Theory selector near the input	Metaphysics workflow
Settings → Models	Newton or BYOK provider, model, primary, fallback

Newton path (in-house platform)

Desktop (signed in)
    → Mystilink Platform (account / wallet / usage)
    → Newton inference service

• Auth is tied to Mystilink account sign-in (customer mnk_ API keys are not used in this phase).
• Insufficient balance may refuse requests (e.g. 402); see Newton billing.

BYOK multi-provider routing

• Configure multiple providers (OpenAI, Anthropic, Google, DeepSeek, Moonshot, MiniMax, etc.).
• Set a primary model as the default target.
• Optional fallback when the primary is unavailable.
• API keys are not uploaded to Mystilink; this path does not use the Newton platform.

Test connection when adding a model so you do not save a config that cannot call.

Privacy

• BYOK: Keys stay on device; content goes directly to your chosen vendor.
• Newton: That inference request is processed by Mystilink / the Newton platform; session files can still live under local .mystilink/.
• Local usage aggregates: Settings → Usage.

See Design principles and FAQ: LLMs & privacy.

Troubleshooting

Invalid keys, model IDs, Base URL, proxy, Newton sign-in / 402: Model connection issues.