Oumi PBC has updated its Frontier AI Engineer platform, enabling users to build AI agents and retrain the underlying models. The tool allows teams to define agent objectives in plain language, while the system automatically identifies necessary tools and environments for subsequent training and evaluation.
The platform now supports direct preference optimization and group relative policy optimization, which is currently available in private preview. Customers can also utilize the Bring Your Own Harness feature to connect existing orchestration and memory systems to Oumi for model tuning without migrating their infrastructure.
Oumi advocates for model ownership, arguing that specialized, customer-owned models perform better and cost less than generic alternatives. The startup, which launched its commercial platform in March, focuses on training models based on task outcomes and user preferences to improve reliability in production environments.
Based in Seattle, Oumi is a venture-backed startup founded by former engineering leaders from Google’s Gemini group. In January 2025, the company secured a $10 million seed funding round, which was led by Venrock Management and Obvious Management Services.
