OpenAI SDK

The gateway speaks the OpenAI Chat Completions API. Point any OpenAI SDK at it.

Node / TypeScript

import OpenAI from "openai"

const client = new OpenAI({
  baseURL: "https://inference.nikcli-ai.dev/v1",
  apiKey: process.env.NIKCLI_API_KEY!,
})

const completion = await client.chat.completions.create({
  model: "nikseek",
  messages: [{ role: "user", content: "Explain QUIC vs HTTP/2 briefly" }],
})

console.log(completion.choices[0].message.content)

Python

from openai import OpenAI

client = OpenAI(
    base_url="https://inference.nikcli-ai.dev/v1",
    api_key=os.environ["NIKCLI_API_KEY"],
)

resp = client.chat.completions.create(
    model="nikcli-coder",
    messages=[{"role": "user", "content": "Write me a Rust hello world"}],
)
print(resp.choices[0].message.content)

Streaming

const stream = await client.chat.completions.create({
  model: "nikseek",
  stream: true,
  messages: [{ role: "user", content: "Write a haiku about caches" }],
})
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "")
}

Tool calls

OpenAI tool calling works as-is — the gateway forwards tools and tool_choice to the upstream provider. Best with models that have toolcall: true in /v1/models.

Cache & reasoning hints

Pass nikcli-specific options in body.nikcli:

const completion = await client.chat.completions.create({
  model: "kimi-k2.6:thinking",
  temperature: 0,
  messages: [...],
  // @ts-expect-error — extra field accepted by the gateway
  nikcli: {
    cache: true,             // opt-in caching even for non-deterministic calls
    cacheTtlSeconds: 3600,
    preferProvider: "openrouter",  // pin a provider
  },
})