Home
DESIGN.mdStar on GitHub

AI is hardeasy to productionize

The AI developer platform to build AI agents, applications, and models with confidence.

W&B Weave: Build agentic AI applications
1import weave2weave.init("quickstart")3@weave.op()4def llm_app(prompt):5    pass  # Track LLM calls, retrieval, agent steps
W&B Models: Build AI models
1import wandb2run = wandb.init(project="my-training-project")3run.config = {"epochs": 1337, "lr": 3e-4}4run.log({"metric": 42})5artifact = run.log_artifact("./model.pt")
Color
Brand
Gold
#FFCC33
Mark, signature
Teal
#00AFC2
Secondary, links
Magenta
#FF0E65
Accent, crashed
Gold Orange
#FFAD33
Warm accent
Run spectrum
run-1
#FFCC33
run-2
#00AFC2
run-3
#FF0E65
run-4
#FFAD33
run-5
#5EEAD4
lavender
#EDE8FF
Surfaces
Base
#1A1C1F
Dark page
Surface
#212429
Panels
Elevated
#2B3038
Slate, raised
Light base
#FBFBFC
Light page
Typography
display
The AI developer platform
Source Sans 3700
serif
Trusted by leading AI teams
Source Serif 4
body
Track experiments, visualize metrics, and ship models.
15px400
mono
wandb.log({"train/loss": 0.214})
Inconsolata
Components
Buttons
Run state
running finished crashed queued
Signature · Experiment Runs
llm-finetune / workspace
train/loss — step 0 → 100
Runs · shown

Toggle a run to show/hide its curve · switch the metric to reslice every run.

Unofficial design reference · built from live wandb.ai · not affiliated with Weights & Biases