Get started

Quick start

From an empty machine to a reply, a conversation in the browser, and a picture, in about five minutes.

This assumes you have installed Kvad and said yes to the background service. Without the service every command below still works: it runs in the process instead of going to a server.

#1. Ask something

kvad run --prompt "Why is the sky blue?"

With no model chosen this downloads HuggingFaceTB/SmolLM2-135M-Instruct, a 270 MB model that is small enough to arrive in seconds, and answers. The first line of output says where the command ran:

kvad: http://127.0.0.1:5823 (the kvad service; --local to run in this process instead)

#2. Get a better model

search asks Hugging Face and says which results this build can run, and at which precision they fit in this machine's memory.

kvad search qwen2.5
kvad pull Qwen/Qwen2.5-1.5B-Instruct
kvad use Qwen/Qwen2.5-1.5B-Instruct

use makes it the default for every command that names no model.

#3. Have a conversation

kvad chat

Or in the terminal app, which also browses and downloads models:

kvad-tui

#4. Open the web UI

The service is already listening. Open http://127.0.0.1:5823.

If you did not install the service, start a server by hand:

kvad serve

Chat, the playground, images, videos, training, evals and benchmarks are all pages there. See The web UI.

#5. Make a picture

On a Mac:

kvad pull stabilityai/stable-diffusion-xl-base-1.0
kvad images make "a lighthouse on a cliff at dusk, oil painting" \
    --model stabilityai/stable-diffusion-xl-base-1.0 --out lighthouse.png

SDXL is a 6.3 GB download. On an M5 Pro a 1024² picture at 30 steps takes about a minute and a half.

#6. Point a client at it

Anything that speaks OpenAI's API works against http://127.0.0.1:5823/v1:

curl http://127.0.0.1:5823/v1/chat/completions \
  -H 'content-type: application/json' \
  -d '{
    "model": "Qwen/Qwen2.5-1.5B-Instruct",
    "messages": [{"role": "user", "content": "Say hello in Norwegian."}]
  }'

#Where next

esc

Try , or .