Open models for agentic workloads, run on-premise.

# the recommended model; a successful call returns a chat completion
curl https://inference.restech.unsw.edu.au/v1/chat/completions \
  -H "Authorization: Bearer sk-..." \
  -H "Content-Type: application/json" \
  -d '{"model": "qwen3.8-flash-next", "messages": [{"role": "user", "content": "Hello"}], "stream": true}'

# pip install openai
from openai import OpenAI

client = OpenAI(base_url="https://inference.restech.unsw.edu.au/v1", api_key="sk-...")

resp = client.chat.completions.create(
    model="qwen3.8-flash-next",
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

// install:  npm i -g opencode-ai
// config:   ~/.config/opencode/opencode.json
{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "restech": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "RTS Local Inference",
      "options": {
        "baseURL": "https://inference.restech.unsw.edu.au/v1",
        "apiKey": "sk-..."
      },
      "models": {
        "qwen3.8-flash-next": { "name": "Qwen3.8 Flash Next" },
        "glm-5.3-flash":      { "name": "GLM-5.3 Flash" }
      }
    }
  }
}
// run:  opencode  →  /models  →  RTS Local Inference

# install:  npm i -g @openai/codex
# config:   ~/.codex/config.toml
model = "qwen3.8-flash-next"
model_provider = "restech"

[model_providers.restech]
name     = "RTS Local Inference"
base_url = "https://inference.restech.unsw.edu.au/v1"
env_key  = "RESTECH_API_KEY"
wire_api = "responses"

# run:  export RESTECH_API_KEY=sk-...  &&  codex
# Codex speaks the Responses API only; this gateway serves /v1/responses.

# install:  npm i -g @qwen-code/qwen-code
# config:   ~/.qwen/.env   (or export these in your shell)
OPENAI_BASE_URL=https://inference.restech.unsw.edu.au/v1
OPENAI_API_KEY=sk-...
OPENAI_MODEL=qwen3.8-flash-next

# run:  qwen

# install:  curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash
# config:   ~/.kimi/config.toml
default_model = "rts-qwen"

[providers.restech]
type     = "openai_legacy"
base_url = "https://inference.restech.unsw.edu.au/v1"
api_key  = "sk-..."

[models.rts-qwen]
provider         = "restech"
model            = "qwen3.8-flash-next"
max_context_size = 1048576

# run:  kimi

// install:  npm i -g @mariozechner/pi-coding-agent
// config:   ~/.pi/agent/models.json
// note:     "models" replaces the provider's built-in list, so declare every model you want.
{
  "providers": {
    "restech": {
      "baseUrl": "https://inference.restech.unsw.edu.au/v1",
      "apiKey": "RESTECH_API_KEY",
      "api": "openai-completions",
      "models": [
        {
          "id": "qwen3.8-flash-next",
          "name": "Qwen3.8 Flash Next",
          "reasoning": true,
          "input": ["text", "image"],
          "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
          "contextWindow": 1048576,
          "maxTokens": 32768
        }
      ]
    }
  }
}
// apiKey takes a literal key or an environment variable NAME (no $).
// run:  export RESTECH_API_KEY=sk-...  &&  pi --model restech/qwen3.8-flash-next

# install:  pip install hermes-agent
# config:   ~/.hermes/config.yaml   (put the key in ~/.hermes/.env if you prefer)
model:
  default: qwen3.8-flash-next
  provider: custom
  base_url: https://inference.restech.unsw.edu.au/v1
  api_key: sk-...

# run:  hermes setup   →   hermes chat
# `provider: custom` is a first-class provider; with base_url set, Hermes calls
# this endpoint directly. `hermes model` configures the same thing interactively.
# config.yaml is the single source of truth. There is no LLM_MODEL env var.

Cursor is not supported here. That is by design, not by omission.

Cursor builds the final prompt on its own servers. Even when you supply your
own API key, every request travels:

    your editor  →  Cursor's backend  →  this gateway

Two consequences:

  1. It needs a *publicly reachable* HTTPS endpoint. This gateway is
     campus-internal and has no public address, so Cursor cannot reach it.

  2. Your prompts and source code would leave the campus network and pass
     through a third party. That is the precise thing an on-premise inference
     service exists to prevent.

Exposing the gateway publicly to satisfy point 1 would guarantee point 2.
So we do not.

Every agent in the other tabs sends its inference requests straight from your
machine to this gateway, so your prompts and code stay on the campus network.
(The tools themselves may still check for updates or phone home on their own.
That is between you and them; it is not this gateway.)