Skip to content

Managed Kimi inference

Nestor exposes Kimi K3 through an OpenAI-compatible API. Your Nestor key is the only credential required; upstream provider credentials are never shared.

Python quick start

from openai import OpenAI

client = OpenAI(
    api_key="nsk_...",
    base_url="https://api.serve.nestor.software/v1",
)

stream = client.chat.completions.create(
    model="kimi-k3",
    messages=[{"role": "user", "content": "Hello"}],
    max_tokens=4096,
    stream=True,
)

for event in stream:
    delta = event.choices[0].delta if event.choices else None
    if delta and delta.content:
        print(delta.content, end="", flush=True)

Kimi may return reasoning_content separately from final content. Reasoning tokens count toward the output limit, so leave enough output capacity for both.

Models and usage

curl https://api.serve.nestor.software/v1/models \
  -H "Authorization: Bearer $NESTOR_API_KEY"

curl "https://api.serve.nestor.software/v1/usage?days=30" \
  -H "Authorization: Bearer $NESTOR_API_KEY"

Responses include request, RPM, and TPM headers. Save x-nestor-request-id when contacting support.

Data handling

Nestor records request metadata, timing, token usage, errors, and charges. Prompt and response content are not stored or written to application logs.

Limits and errors

  • 401: missing, invalid, or revoked Nestor key
  • 403: Kimi is not enabled for the account
  • 429: concurrency, RPM, TPM, or daily spend limit reached
  • 502 or 503: upstream service unavailable
  • 504: upstream timeout

Streaming requests are never restarted after the first response chunk. Clients may retry a failed request when no response content was received.