OpenClaw and Hermes Agent take different approaches to persistent AI coding assistants. One prioritizes ecosystem reach, the other deep learning over time.

The New Stack is a publication covering trends and technologies in cloud-native development, DevOps, and software delivery. Developers can learn about containerization, Kubernetes, and cloud computing, as well as explore topics such as microservices architecture, serverless computing, and continuous integration/continuous delivery (CI/CD) pipelines.

The New Stack

Two open-source persistent AI agent projects — OpenClaw and Hermes Agent — are compared as they tackle the problem of context loss between AI coding assistant sessions. OpenClaw (345K+ GitHub stars) prioritizes ecosystem breadth with 50+ messaging integrations, model-agnostic support, and a large community skill marketplace (ClawHub), but has faced serious supply chain security incidents. Hermes Agent (22K stars, from Nous Research) focuses on deep learning over time via FTS5-backed persistent memory, autonomous skill creation, and a self-training loop using the Atropos RL framework. A comparison table helps developers choose based on their needs. Both projects represent an architectural shift from session-bound AI tools toward always-on, stateful agent runtimes, raising open questions about ownership of accumulated agent knowledge.

OpenClaw vs. Hermes Agent: The race to build AI assistants that never forget

The always-on agent is a new software category