Models, keys, and limits
Keys go in Vault. Big models go where they fit.
The model policy is simple: store secrets safely, use hosted providers for heavyweight routes, use OpenZero for local ownership, and keep local model files under 15 GB on current target machines.
Key And Provider Matrix
| Provider | Use | Boundary |
|---|---|---|
| Groq | Fast default/reasoning routes. | Server route or user key depending account state. |
| OpenAI | Premium code, writing, long analysis, agent planning. | Pro or BYO key. |
| Gemini / xAI / NVIDIA | Alternative hosted providers. | BYO key in Vault. |
| Featherless | Hosted large models such as GLM-family and talktoaiZERO style routes. | Usually paid/BYO key; external billing and terms apply. |
| OpenZero Node | User-owned local AI node and OpenAI-compatible API key. | Model file under 15 GB unless hardware changes. |
| Serper | Web search and source ledger support. | Use source title, URL, snippet, timestamp, and evidence notes. |
| IonQ / IBM Quantum | Quantum context, telemetry, Qiskit packets. | Classical LLM reasoning over quantum evidence. |
| ZeroThink API | Admin/pro integrations. | Server-issued, redacted, revocable, and account-scoped. |
GLM And Chinese Model Route
GLM-family routes are useful for Chinese/English reasoning and technical workflows. Use hosted access for models too large for OpenZero. For local tests, choose quantized GGUF sizes under 15 GB, with Q5 or Q6 preferred when hardware permits and Q3/Q4 when memory is tight.
Secret Handling
Never paste API keys into chats, screenshots, GitHub issues, or public docs. Vault display should be redacted, revocation should be immediate, and admin/master actions should be audited.
Questions And Answers
What is ZeroThink Pro for?
Pro is for heavier routes such as advanced providers, Quantum lanes, hosted model access, OpenZero bridge trials, and higher-limit workflows where configured.
Is Featherless local?
No. Featherless is a hosted model provider and generally requires a paid or bring-your-own Featherless API key.
What is the local model size rule?
For current OpenZero target machines, local model files should stay under 15 GB. Use hosted providers for larger models.
For business, research, hosting, security, AI, SEO, and product scoping: use the contact form. No login, signup, or public email address is needed on this docs page.