Hermes Agent already had most of the primitives we needed to build agents that can operate inside a company: tool use, skills, memory, planning, and multi-step execution. It is open source, which gives us more control over how the runtime is deployed, extended, and connected to company systems.
For Iris, the goal was not to start with a model API and rebuild the agent layer ourselves. We wanted to focus on the product around it: integrations, company context, permissions, workflow execution, cross-team coordination, and making agents reliable enough to take on recurring operational work.
