New FREE Hermes AI Agent Update is INSANE!
The video covers the major new Hermes AI Agent V0.3.0 update from NUS Research, highlighting features like streaming responses, voice mode, plugin architecture, and IDE integration. Hermes is positioned as a free, open-source AI agent that learns and remembers users over time, unlike traditional AI tools that reset each session.
Summary
Julian introduces the Hermes AI Agent V0.3.0 update, released March 17th, 2026, from NUS Research. He emphasizes how most AI tools require users to start fresh each session, while Hermes builds a persistent model of users and their workflows. The agent can run continuously on cloud servers, allowing users to interact via mobile while tasks execute remotely. The update introduces unified streaming for real-time token responses across all platforms, replacing previous block text outputs. A new plugin architecture allows developers to simply drop Python files into a plugins folder for automatic integration without complex setup. The update adds native Anthropic provider support with prompt caching, eliminating the need for OpenRouter as a middleman for Claude models. Voice mode capabilities include push-to-talk CLI, voice notes via Telegram/Discord, and Discord voice channel participation, all using local Faster Whisper transcription for privacy. Other improvements include concurrent tool execution using threads, smart approval systems that remember user preferences, IDE integration through Agent Communication Protocol for VS Code, Zed, and JetBrains, and persistent shell mode maintaining state across tool calls. The memory system now prioritizes personal preferences, includes PII redaction for sensitive data, and offers better multi-user isolation. New skills were added for Linear, Twitter, Twilio, 1Password, and Blender, accessible through the skills hub. Installation requires a single curl command, and the tool supports over 200 models through various providers.
Key Insights
- Julian argues that most AI tools have a fundamental workflow problem where they forget everything when you close the tab, forcing users to teach the AI assistant who they are from scratch repeatedly
- Julian claims that Hermes can run on a $5 VPS in the cloud and continue working while users sleep, allowing them to message it from Telegram on their phone while it executes tasks on a server they never need to access
- Julian explains that the new plugin architecture allows developers to simply drop a Python file into the Hermes/plugins folder for automatic pickup, requiring no forking of the codebase or complicated setup
Topics
Full transcript available for MurmurCast members
Sign Up to Access