For the complete documentation index, see llms.txt. This page is also available as Markdown.

Serverless Inference

Run text, vision, image, video, embedding, and reranking models on managed OpenAI-compatible endpoints.

Serverless Inference gives you managed endpoints for state-of-the-art models — no GPUs to provision, no weights to load, no scaling to manage. Sign in, create a key, and call the API.

The endpoint is OpenAI-compatible and lives at https://inference.nebulablock.com/v1, so any OpenAI SDK or tool works by changing the base URL.

What you can do

Guide

Chat and text generation, including tool use and streaming

Send images to a model and ask about them

Generate and edit images

Generate video from text or an image

Turn text into vectors for search and RAG

Reorder retrieved documents by relevance

The Model Catalog lists every model currently served, grouped by what it does.

Prerequisites

Key features

  • OpenAI compatible. Point the OpenAI Python, Node, or any compatible client at the base URL above.

  • One key, every modality. Text, vision, image, video, embedding, and reranking models share the same authentication and the same base URL.

  • Pay as you go. Billed on usage, metered per token — or per image or per second for some media models. See the pricing page.

  • Try before you integrate. Every model has a playground in the console under Serverless.

  • Canadian-hosted options. Models marked 🇨🇦 in the catalog run in Canada for data-residency requirements.

Rate limits

Two limits apply to every call: your account-wide rate limit (RPM, TPM, RPD) and a per-model daily cap that varies by tier. Both are documented in Tiers and Rate Limits, and your account's current numbers are shown under Limits in the console.

See also

Last updated