Descriptions:
News Research released Hermes Agent version 0.9—the ‘everywhere release’—and Wes Roth provides one of the first detailed setup walkthroughs of the framework, which positions itself as an open-source alternative to OpenClaw with a fundamentally different architecture: persistent memory and a continuous self-improvement loop. Rather than executing one-off tasks, Hermes Agent forms hypotheses, tests approaches, and codifies learned solutions as persistent skills that survive across sessions and grow more capable over time.
The setup demonstration covers Docker-based deployment on a Hostinger VPS, including docker compose commands, container entry, virtual environment activation, and initial configuration. A fresh v0.9 install ships with 74 pre-loaded skills spanning academic paper research, image and video creation, GitHub automation, local model quantization, fine-tuning workflows, and integrations with both Claude Code and CodeX. The framework also includes a ‘god mode’ skill for jailbreaking API-served language models. News Research—the distributed, open-source lab behind models like World Sim—built Hermes on the principle that AI behavior and ethics should be controlled by users, not by large labs.
Version 0.9 also introduces support for agentic reinforcement learning pipelines, enabling large-scale asynchronous LLM training using Hermes primitives. Upcoming local deployments pairing Hermes with Gemma 4 on consumer MacBook hardware are previewed, and the framework already integrates Andrej Karpathy’s LM wiki. For developers tracking open-source agent alternatives to cloud-dependent platforms, Hermes Agent represents a notable new entry point.
📺 Source: Wes Roth · Published April 15, 2026
🏷️ Format: Hands On Build
