Nous Research brings Hermes Agent to enterprise teams

AI startup Nous Research has released Hermes Business and Hermes Enterprise, two new tiers designed to bring its popular autonomous agent framework to corporate environments. The launch follows reports that the open-source research lab is negotiating a funding round at a $1.5 billion valuation. After its February release, the underlying Hermes Agent passed 175,000 GitHub stars within four months and has now reached roughly 214,000 stars and nearly 40,000 forks. The new offerings transition the technology from individual developer environments into collaborative and highly secure corporate setups.

The Business tier operates on the Nous Portal, which consolidates access to 248 distinct AI models and a hosted Tool Gateway under a single login. Instead of developers managing individual API keys and separate subscriptions, the team tier introduces a shared credit balance with administrative controls to set spending caps per member. This prevents runaway agent loops from exhausting company funds. Crucially, the platform allows teams to build a shared library of reusable agent skills—such as deployment checks, data-retrieval routines, and report templates—ensuring that automated workflows remain proprietary company assets even if individual employees leave.

For practitioners, these shared skills offer significant efficiency gains. Internal benchmarks from Nous Research show that agents equipped with more than 20 self-created skills completed recurring tasks about 40% faster and consumed roughly 40% fewer tokens compared to fresh agent instances. While these performance gains do not apply to one-off, cross-domain tasks, they provide immediate latency and cost benefits for stable, repetitive workflows like report generation.

For organizations with strict compliance requirements, Hermes Enterprise allows the entire agent, memory, and skill stack to run on-premises or within a private cloud. This self-hosted edition can operate without telemetry or tracking, enabling companies to pair the agent with open-weight models hosted on local GPUs. However, practitioners must ensure that connected search APIs or external tools also comply with local data policies, as requests sent to external providers will still exit the private environment.