COREWEAVE ARIA
AI Research and Iteration Agent
CoreWeave ARIA is an AI research agent built directly into Weights & Biases to analyze experiment data and production traces at scale. It uncovers hidden patterns, identifies performance drivers, generates persistent visualizations, and tracks trends across runs over time. By analyzing thousands of runs and tens of thousands of metrics in minutes, ARIA closes the autoresearch loop between analysis and action, helping teams iterate faster, continuously improve models, and build more reliable AI agents.
Continuous improvement on autopilot
Describe what you want to learn, and ARIA forms hypotheses, designs experiments, launches runs through W&B Launch, evaluates results, and recommends the next iteration. It compares outcomes against baselines, updates workspaces with new findings, and drafts reports that capture what changed and why. When results are inconclusive, ARIA proposes follow-up experiments to refine the search. Researchers stay in control by approving launches, while ARIA handles the repetitive work between iterations. The result is a faster research loop where models and agents continuously improve, and teams spend less time managing experiments and more time solving hard problems.
Live, persistent dashboards to backup findings
When ARIA uncovers an insight, it automatically creates W&B workspaces, panels, and reports that make its reasoning transparent and actionable. ARIA selects the right visualization for the problem, whether that is a heat map for parameter sweeps, a parallel coordinates plot for hyperparameter interactions, or a bar chart for comparing configurations, and organizes the most relevant runs into a shareable workspace. These artifacts persist beyond the conversation, update automatically as new runs arrive, and remain fully customizable. Every recommendation is backed by live visual evidence, making it easy to validate findings and move from insight to action with confidence.
Full experiment context, already loaded
ARIA understands your project, runs, filters, metrics, artifacts, checkpoints, and training code behind the page you’re viewing. It can pull from experiment logs, loss curves, and evaluation results, then connect those findings to related work across projects and teams. By analyzing hundreds of thousands of logged metrics in context, ARIA surfaces relationships and performance drivers that are difficult to discover manually. Instead of spending time gathering data and explaining your setup, you can jump straight to the question and get answers grounded in the full history of your research.
Available on the go
ARIA is in the
Weights & Biases mobile app
You can monitor runs, investigate results, and interact with ARIA from anywhere.
“ARIA has become a valuable part of my daily workflow. It helps me quickly generate reports, create sweep configurations from natural language, and automate tasks that would otherwise require a lot of manual setup. What excites me most is the potential to connect workflows end-to-end, from launching experiments and configuring automations to generating insights and reports automatically. Even today, the agent saves time on repetitive work and makes interacting with W&B much more intuitive.”
— Praneeth Gangavarapu, PhD Candidate, Scripps Research
Get started with ARIA
The Weights & Biases end-to-end AI developer platform
Weave
Models
The Weights & Biases platform helps you streamline your workflow from end to end
Models
Experiments
Track and visualize your ML experiments
Sweeps
Optimize your hyperparameters
Registry
Publish and share your ML models and datasets
Automations
Trigger workflows automatically
Weave
Traces
Explore and
debug LLMs
Evaluations
Rigorous evaluations of GenAI applications