CoreWeave Inc., widely recognized as a specialized cloud provider tailored for artificial intelligence workloads, has announced the launch of CoreWeave Forge, a new development layer designed specifically for engineering teams tasked with building, refining, and scaling advanced machine learning models and intelligent agents. The announcement was made during Fully Connected, CoreWeave’s flagship annual AI cloud conference, which gathered more than 4,500 industry customers, technology partners, software developers, and artificial intelligence leaders to discuss the practical realities of building and running production-grade AI systems.
According to the company, CoreWeave Forge is engineered to run the entire continuous AI development loop—encompassing running, observing, curating, improving, evaluating, and repeating—within a single, unified environment. Crucially, the platform has been built to remain entirely open, allowing developers to integrate it across the diverse array of external models, software frameworks, and alternative cloud infrastructures that modern engineering teams routinely rely on. Early enterprise adopters, including prominent consumer platforms such as MasterClass and Canva, are already leveraging the new development layer to accelerate their internal AI initiatives.
The launch of Forge addresses a longstanding and costly friction point within the artificial intelligence engineering lifecycle. Historically, the process of developing, fine-tuning, and maintaining sophisticated models and autonomous agents has been fragmented across a sprawling ecosystem of tools provided by separate vendors, many of which were never designed to interoperate smoothly. For researchers, data scientists, and systems engineers guiding a model from its initial conception through its complete production lifecycle, this fragmentation carries a heavy operational and financial cost.
In a typical fragmented workflow, operational telemetry and production traces generated by a deployed model do not automatically feed back into the next training run. Similarly, the insights gained from a completed experimental iteration often fail to inform subsequent evaluation metrics. Every time an artifact or process is handed off between disparate tools and teams, valuable signal data is inevitably lost, and additional engineering hours—and financial capital—are wasted. CoreWeave’s leadership identified this systemic inefficiency as a primary bottleneck slowing down the pace of AI innovation across the enterprise landscape.
To solve this challenge, CoreWeave Forge brings together several key components and partnerships into a cohesive platform built for continuous operational improvement. The new development layer integrates Weights & Biases Models for experiment tracking and visualization, specialized post-training expertise and tooling from OpenPipe, and the open-source marimo interactive notebook project. These capabilities are deeply coupled with CoreWeave’s own high-performance infrastructure services, creating an interconnected ecosystem where training runs, experiment tracking metrics, rigorous evaluations, and live agent execution traces all live directly where the model or agent actually operates.
Furthermore, the platform incorporates CoreWeave Registry to ensure that every individual checkpoint, model artifact, and agent configuration file remains meticulously versioned throughout its lifecycle. By housing these operations within a unified environment, Forge aims to close the continuous AI loop, assisting development customers in systematically improving their models and autonomous agents regardless of their internal team’s specialized expertise. At the same time, the system maintains strict architectural neutrality, ensuring that development teams are never locked into proprietary ecosystems and remain completely free to build using whichever third-party models, development frameworks, or external clouds best suit their technical requirements. This openness ensures that performance improvements can compound natively with every successive model version.
The unveiling of Forge at the Fully Connected conference underscores CoreWeave’s broader strategic evolution from a pure-play infrastructure provider offering high-density graphics processing unit compute capacity into an integrated platform provider that addresses the entire software development lifecycle for artificial intelligence. As enterprise adoption of generative AI and autonomous agent architectures moves rapidly past the initial proof-of-concept phase and into heavy production environments, the demand for streamlined, interoperable tooling has grown increasingly acute. By bridging the traditional gap between infrastructure management, model training, and continuous evaluation, CoreWeave aims to position its cloud platform as a central hub for the next generation of AI engineering workflows.
