Demo on Kodiak-v0.4-1B

v0.3: Kodiak-v0.3-1B default

Pin the builder to gradio 6.29.0 (matches requirements.txt)

ZeroGPU: reserve 5 s per decision, 20 s per table (quota is charged the reserved time)

ZeroGPU: import spaces before CUDA initializes

Python 3.12 (kodiak-s1 needs >= 3.12; ZeroGPU defaults to 3.10)

v0.2: Kodiak-v0.2-1B default; ZeroGPU (@spaces.GPU)

Kodiak XL as default; examples that show real strengths (tool list, stated facts, honest can't-tell)

New look: Kodiak header, answer cards with confidence bars, styled game; Gradio pinned

Lighter demo: drop the duplicate single large model; lists always run on small

Pin kodiak-s1 to the current commit (the cached install lacked ensemble support)

Accuracy mode as the default model; turn off SSR (its proxy returned 502s for CSS)

Game: clear game-over screen after 5 customers

Game: result above the next customer, labeled; yes/no in records

Game tab: the Enchanted Returns Desk, you vs. the Bear

Feedback form: question list fills from the questions box; no Decide needed

Feedback form: 'Did Kodiak get something wrong?'

Reviews preset: positive / negative / mixed / neutral with clearer labels

Model picker: small (default) and large

Research preview banner

Kodiak demo app

initial commit