Skip to content

PAIS — Platform for AI Services

University of Auckland's self-hosted AI compute platform for research.

PAIS gives researchers and research software engineers (RSEs) direct access to GPU-accelerated AI infrastructure — without the complexity of cloud billing, data sovereignty concerns, or per-token costs. Everything runs on university-controlled hardware inside the New Zealand research network.

What can I do with PAIS?

PAIS exposes an OpenAI-compatible API — point any Python script, notebook, or agentic framework at https://api.pais.auckland.ac.nz/v1 and use the same code you'd write for ChatGPT. Works with the OpenAI SDK, LangChain, LlamaIndex, and any other OpenAI-compatible tool.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.pais.auckland.ac.nz/v1",
    api_key="your-pais-api-key",
)
response = client.chat.completions.create(
    model="llama-3-1-8b",
    messages=[{"role": "user", "content": "Summarise this research abstract..."}],
)

Kubeflow Notebooks gives every research group isolated Jupyter environments with direct GPU access. No local CUDA setup, no fighting over lab workstations — spin up a notebook and start training.

Define reusable training, evaluation, and deployment workflows using Kubeflow Pipelines. Run hyperparameter sweeps with Katib. Track every experiment with MLflow.

Open WebUI provides a private ChatGPT-like interface running entirely on university hardware. Log in with your University of Auckland account via Tuakiri SSO.

Platform at a glance

What Detail
Inference models Llama 3.1 8B (chat), Qwen3-VL-Embedding-8B (embeddings)
API compatibility OpenAI-compatible (Chat Completions, Embeddings)
GPU NVIDIA L40S 48 GB, time-sliced across workloads
Authentication Tuakiri SSO (institutional login) + API keys via Portal
Storage VAST network-attached storage, shared per research group
Access model Research group membership — see Requesting Access

Where to start

  • New to PAIS?

    Follow the Getting Started guide — request access, get an API key, and make your first call in under 15 minutes.

  • Calling the API?

    Jump to Inference API for endpoint reference, authentication, and model details.

  • Building an agentic system?

    See Agentic Development for framework-specific setup (LangChain, LlamaIndex, Claude Code).

  • Looking for examples?

    The Worked Examples section has full, runnable code for common research AI patterns.

Research groups

PAIS uses a condominium model — research groups contribute GPU nodes and receive priority access. Idle capacity is shared across all groups. Current groups on the platform:

  • rg-compsci — Computer Science
  • rg-abi — Auckland Bioengineering Institute
  • rg-general — General research use (shared capacity)

Contact your group's PI or the PAIS team to join a group.

Getting help

  • Access requests: Email the PAIS team or contact your research group PI
  • Documentation improvements: Use the edit button on any page