Osmosis
About
Osmosis is a reinforcement fine-tuning platform for task-specific models and AI agents. It combines hands-on post-training, reward-function design, multi-turn tool training, managed training and serving infrastructure, and continuous retraining, while also publishing open models such as Osmosis-MCP-4B.
What They Offer
Commercial or operational services this company provides.
Rubrics / verifiers
Managed infrastructure
Products & Public Artifacts
Osmosis-MCP-4B
Model · Open source
Open model trained with reinforcement learning for multi-step MCP-style tool use.
Visit artifact →Osmosis RL Platform
Tool · Commercial
End-to-end reinforcement fine-tuning platform for task- and domain-specific models, handling compute orchestration, reward modeling, and training-run observability.
Visit artifact →Technical Capabilities
Tool / API / MCP UseLong-Horizon TasksProgrammatic Verifiers
Only publicly documented capabilities are shown. Absence does not imply the capability is unavailable.
Focus Areas
RLHF / Post-trainingAgent Infrastructure
Domains
Software EngineeringEnterprise
Leadership
Kasey Zhang
Founder & CEO
Andy Lyu
Founder & CTO