Hermes Agent v0.22.0 The Threshold Release คืออะไร? รุ่นใหญ่ที่สุดของ AI agent ที่โค้ดหดลง 34%

Hermes Agent v0.22.0, The Threshold Release, is the largest AI agent release yet, with a 34% reduction in code, according to Nokka (Nok-ga). Published on October 17, 2026, by AI using Hermes Agent under human monitoring, this version is designed for users who find lengthy settings overwhelming. Hermes Agent v0.22.0, named The Threshold Release or Threshold Door, is the biggest release by the Nous Research team.

It enables users familiar with AI tools to overcome intimidating settings screens. This release marks a significant upgrade from v0.21.0, with the team reporting a 2.5x increase in size. The release features 24,200 commits, 6,080 lines of code added, 15,600 lines modified, 2,045,000 characters added, and 1,140,000 characters removed.

It addresses 7,180 closed issues and involves 2,520 developers. Notably, 254,551 tokens, 55,325 modifications, and 49,672 open issues exist. The core idea of The Threshold Release is to help newcomers transition into AI agents. Unlike previous versions, this release introduces a simplified onboarding process with a short chat to gather essential information.

Users can now configure themes, dark mode settings, and add plugins without needing an API key. The agent runs on the free trial model nous/welcome, which can be used once per system. Users can restart the onboarding process anytime through Settings → Run onboarding again. The release enhances connectivity options, integrating with Linear and Notion platforms through the Nous tool gateway.

Users can sign up once and use the tool for one session. The release also introduces a Connectors menu for managing multiple accounts in a single workspace, allowing users to switch between work and home accounts seamlessly. The local model installation supports llama.cpp, enabling users to choose their preferred engine based on their hardware specifications, including CPU, GPU, CUDA, Metal, Vulkan, and ROCm.

The release includes Qwen3.8 27B, Qwen3.8 Flash Next with speculative MTP tokens, Qwen3.6 35B-A3B, and DeepSeek V4 Flash models, among others. The update process is streamlined, with optional server and gateway restarts and automatic environment updates. Users can create new profiles for each account, each with its own settings, permissions, tasks, and sessions.

The subagent feature allows users to view running subagents in the CLI, enhancing visibility during operation.