BYOK model providers

BYOK (Bring Your Own Key) lets you plug your own provider API key into VizChat. Inference and generation requests are then billed to your account with that provider, and do not consume VizChat platform credits. For heavy users, this is the cheapest path.

Prerequisites

• Subscription tier ≥ Starter (Free does not support BYOK). New users get an auto-activated 30-day Starter trial on signup, no credit card required.
• Paste the key in Settings → Custom Model. VizChat AES-encrypts it at rest and only displays a masked preview.
• Deleting a key does not fall back automatically. If the model selector is still pointed at a custom model from that provider, requests will fail until you add a new key or switch to an official model yourself.

Supported providers (15)

Provider IDs align with OpenRouter's routing prefixes where possible; the rest use common names.

The three modes in the last column follow from where each vendor sits in the ecosystem:

• Dropdown — these vendors both build models and sell API access, so their models are listed for you.
• Dropdown (full catalog) — OpenRouter is a gateway: one key reaches the entire catalog, including authors like Llama and Nvidia that don't sell API access directly.
• Type the model name — Groq / Together / Fireworks are inference providers running other people's models, so there is no "their own models" axis. Copy the model name from that platform's docs.
• Type URL + model name — custom is the catch-all: anything exposing an OpenAI-compatible API (self-hosted vLLM, an internal gateway, a vendor not yet listed here) can be plugged in. See the address rules below.

Address rules for Other (OpenAI-compatible, custom)

• Only http / https are accepted. Enter the OpenAI-compatible API address from that service's docs (it usually ends in /v1).
• Services on your own machine or local network (a local Ollama or LM Studio, say) work directly in the VizChat desktop app, even offline. On the web, such an address shows a connection error — to use it there, expose the service at a public address first.
• Whether the address can be reached is checked before each conversation, not only when you save it.
• Type the model name after saving: open the chat model selector, find this provider, and enter it there.

For dropdown providers, the models on offer change over time. To see what's currently available, open that provider's dropdown on the Custom Model settings tab.

Model routing

VizChat's frontend splits models into two slot types:

• Official: platform-hosted Smart / Fast slots, billed in credits.
• Custom: BYOK-injected models, grouped by provider in the dropdown.

The Agent defaults to the Official Fast slot; switch to a custom model and the whole conversation runs on BYOK. BYOK requests are gated only by subscription tier (Starter and above); no platform credits are debited, and VizChat applies no rate limit of its own — your provider's own limits still apply.

FAQ

• My key leaked — now what? Delete it in VizChat immediately, then revoke it in the provider's console. VizChat's ciphertext uses a project-level encryption key; after deletion it cannot be recovered.
• Can I configure multiple providers? Yes. Each provider holds at most one key, and they're independent.
• Does OpenRouter work? Yes — add an OpenRouter key and it routes the whole catalog. VizChat also uses OpenRouter's catalog to list models and pricing for other providers, but when you supply their key the call goes straight to that provider's own endpoint, not through OpenRouter.
• Can I connect a service that isn't listed? Yes — use Other (OpenAI-compatible) (see the address rules above). The other 14 have preset base URLs that aren't user-editable: those addresses are known, so letting users type them would only add a way to get it wrong or abuse it.

Next step

• BYOK feature guide
• Plan limits comparison