Companies can keep their existing AI agent systems while using Oumi's expanded Frontier AI Engineer to assess agent behavior and develop specialized models, the company said in an announcement Nov. 4.
Through Bring Your Own Harness, a company's existing setup can retain its orchestration and tools, along with its memory, environment and application logic. Oumi's platform can support evaluation, data curation and model work around those components. Teams can specify an agent's task and environment, review its behavior, investigate failures and test training runs.
Oumi says production interactions can feed into future evaluation sets and training data.
Direct preference optimization (DPO) now joins supervised fine-tuning (SFT) in the platform. Oumi says teams can sequence multiple training methods within one run. Group Relative Policy Optimization (GRPO), a reinforcement-learning method based on task outcomes, remains in private preview.
Oumi presents customer-owned specialized models as a potential way to improve agent performance and manage costs. The announcement gave no benchmarks, named customer deployments or cost comparisons.
