Oumi Extends Frontier AI Engineer to Build and Evaluate AI Agents

Oumi has expanded its Frontier AI Engineer (FAiE) to help companies develop AI agents and train the specialized models behind them, extending its model-development workflow to agent tasks.

The company announced the update Wednesday, Nov. 4. FAiE already handles evaluation, data curation, training, inference and continuous improvement for specialized AI. The update extends that work to agents and the models that power them.

Oumi says teams can use FAiE to define an agent’s task, tools and environment, then evaluate its behavior, diagnose failures, prepare data and run training experiments. The company also says production interactions can be turned into evaluation sets and training data for future runs.

With Bring Your Own Harness, companies can connect an existing agent setup while keeping its orchestration, tools, memory, environment and application logic in place. Oumi’s system can then be used for evaluation, data curation and model training around that setup.

The update adds direct preference optimization (DPO) alongside supervised fine-tuning (SFT). Oumi says teams can chain different training methods in one run. Group Relative Policy Optimization (GRPO), which Oumi describes as reinforcement learning based on task outcomes, remains in private preview. The company did not give a general-availability date.

Oumi co-founder and CEO Manos Koukoumidis said agents will improve by “training the weights of the models powering them, learning from their own work.” Oumi presents specialized, customer-owned models as a way to improve agent performance and control costs, but the announcement included no benchmarks, named customer deployments or cost comparisons to quantify those benefits.