BYOK (Bring Your Own Key)

BYOK (Bring Your Own Key) is one of VizChat's core differentiators. You can plug your own third-party LLM API key into Settings, and VizChat will call models as your account — meaning BYOK usage does not consume VizChat platform credits, and you retain full control over cost, model version, and data flow.

Why BYOK

• No platform credits consumed: BYOK usage is billed to your provider account; VizChat adds zero markup;
• Access cutting-edge models: freshly released models (e.g., a brand-new reasoning model) may not yet be on the platform's hosted roster, but your own key works immediately;
• Compliance / data boundary: enterprise customers who require conversation data to flow only to their signed providers are natively supported;
• Cost control: enterprise discounts, monthly quotas, and prompt caching you negotiated apply directly.

Plan support

Starter is designed specifically for BYOK users — even though its monthly credits match Free, unlocking BYOK lets you run unlimited calls against your own key.

Supported providers (15)

VizChat currently supports these 15 providers:

Provider IDs follow the OpenRouter prefix convention where possible, for easy multi-provider switching.

The last three are inference providers — they run other people's open models, so there is no "their own models" axis and you type the model name yourself (copy it from that platform's docs). OpenRouter is a gateway: one key routes the whole catalog.

Setup steps

1. Go to Settings → Custom Model;
2. Find the provider you want, click "Add key";
3. Paste your API key — the platform encrypts it immediately, and the plaintext key never leaves the server;
4. After saving, the key shows in masked form (e.g., sk-****...abcd); you can update or delete at any time;
5. In the conversation's model selector, pick a model from that provider — it auto-uses your key.

Whether a model can read images

When the AI checks a widget or a board it takes screenshots, and those are only attached for models that can read images. For most providers the platform works this out from public model listings. Groq, Together AI, Fireworks AI and Other (OpenAI-compatible) take a model name you type, which the platform can't look up — tick Models here can read images when you set up the key. A wrong tick does no harm: if a model turns out not to accept images, the platform falls back to the text report for it automatically.

Security & encryption

• All keys are encrypted at rest with AES-256-GCM;
• The key is decrypted server-side only when it's actually used — to send a request through your provider, or to compute the masked preview shown in Settings — and only for that instant;
• The list endpoint (/api/user/api-keys) returns masked values only — plaintext is never exposed;
• Deletion is a hard delete and takes effect immediately.

Billing transparency

When you use BYOK to call a model:

• VizChat side: zero platform credits deducted;
• Provider side: billed per your provider's rates (OpenAI token prices, Anthropic prompt cache discounts, etc. all apply);
• Mixed usage: within one account, you can mix BYOK models and platform-hosted models freely; each call is billed independently.

Next step

• Plans and credits
• Supported models