Sabaki on Kifubara's GPUs

beta

Love Sabaki but your laptop struggles with KataGo? Use Kifubara as a cloud KataGo — every analysis runs on our GPUs. Your laptop stays cool, your battery thanks you, and you keep the Sabaki UX you already know.

Who this is for

You play and study in Sabaki on a laptop that doesn't have a great GPU (or any GPU).
You'd rather not babysit a KataGo install, model downloads, or driver updates.
You want analysis fast enough that Sabaki's streaming side-panel actually feels live.

How it works

You add an engine in Sabaki's Preferences with three short fields — Name, Path, and a one-line Args string we generate for you.
Each analysis request travels over a secure connection to our GPU pool; responses stream back to Sabaki the same way a local KataGo would.

No persistent install on your computer. No background process. KataGo positions are sent to our GPU and discarded once the analysis is delivered.

What the snippet looks like

A peek at the Args field you'd paste into Sabaki. Sign in to get yours with a personal token in place of the placeholder.

python3 -c "<bootstrap line generated for you on sign-in>" \
  https://app.kifubara.app/static/engine/launcher.py \
  wss://app.kifubara.app/ws/engine/sabaki/<your-personal-token>

Goes into Preferences → Engines → Add → Args. Path is just python3. Save, attach the engine, F4 to toggle analysis.

What you get

Up to 2500 visits per position — solid analysis depth.
Sabaki's analysis side-panel works exactly as designed, with live streaming as positions are explored.
Engine-vs-you (kata-genmove_analyze) works for blitz or self-study.
Standard board / komi / handicap all handled.
Need higher caps for a specific use case? Get in touch — we'll see what we can do.