NousResearch Releases Hermes Agent v0.15.0, Cuts Code 76% and Achieves 4,500x Search Speedup

According to NousResearch, the team released Hermes Agent v0.15.0 in late May, reducing the core run_agent.py code by 76% from 16,083 to 3,821 lines. The update delivers significant performance gains: cold-start time compressed to 0.8 seconds, and session search optimized to 20 milliseconds—a 4,500-fold speedup achieved by replacing large language model calls with local filtering.

The kanban tool evolved into a multi-agent collaboration platform enabling developers to initialize Swarm workflows with coordinators and parallel workers via a single command. New security defenses address Brainworm-style attacks through traffic filtering across tool outputs, memory recall, and skill loading. Ecosystem integration expanded to include Bitwarden-managed API keys, xAI Web Search plugins, and Grok execution guidance. A v0.15.1 patch released hours later resolved a loopback 401 reload loop and Docker path resolution failures.