nikcli Inference

A single OpenAI-compatible HTTP endpoint that:

• Routes each request to the cheapest healthy provider for the chosen model.
• Caches deterministic responses (temperature=0 or seed set) across all your traffic, cross-region.
• Coalesces concurrent identical requests into one upstream call.
• Supports reasoning natively (R1, QwQ) and via :thinking variants on hybrid models.

Behind the scenes we currently route across 14 providers: Together, Fireworks, DeepInfra, Groq, Cerebras, SambaNova, Hyperbolic, Nebius, OpenRouter, DeepSeek, Mistral, Moonshot, Zhipu, and a self-hosted vLLM fallback.

Base URL

https://inference.nikcli-ai.dev/v1

Authentication

Authorization: Bearer <your-key> on every request. Create keys at /dashboard/keys.

Keys start with nik_live_. The full plaintext is shown only once at creation — copy it immediately.

Next steps

• Quickstart — your first request
• Cursor setup — drop-in for Cursor
• Models — full catalog + aliases