AI gateway and provider failover
AI router
Source repository: OmniKey AI
Plan provider selection, retry budgets, fallback order, and credential boundaries without accepting secrets.
Live provider adapter
Discover available models
The server calls the provider model endpoint and normalizes the response. Your key is never persisted.
Supported now: OpenAI, Gemini, Groq, OpenRouter, and OpenAI-compatible endpoints such as Ollama, LM Studio, vLLM, and custom gateways.
Stacks availability
Provider key stacks
Loading stacks…
Each key carries its own free-tier quota — 20 keys on one model means roughly 20× the free calls. Counts refresh every 30 seconds.
Omni key generator
Mint a call key, then call the stacks
One consumer key rides the whole keystack — every stacked provider key backs your calls. Master key never leaves this tab and is never stored.
Inputs
- • Provider
- • Model
- • Endpoint
- • Masked key placeholder
- • Fallback order
- • Omni call key
Boundary
The key is used only for this discovery request and is not stored. No chat request is routed and no latency is measured.