What if your AI agent could work across your terminal, desktop, browser, and messaging apps without constantly losing context or duplicating tasks? And what if multiple agents could collaborate in the same conversation while you controlled exactly how much they were allowed to spend?
Meet Hermes Agent v0.23.0, codenamed The Hearth Release, from Nous Research.
Here's what's new.
1. Your entire AI workspace now runs in a browser
The Hermes Desktop experience is now available through a web interface.
Same conversations. Same agents. Same plugins and settings.
You can access your workspace from another computer, over your local network, or through an authenticated public URL.
No need to install the desktop application on every device.
The web interface is disabled by default, so you need to enable it in your settings.
2. One gateway. One conversation. No duplicate agents.
Previously, different Hermes interfaces could launch separate agent processes against the same conversation.
That could lead to duplicated actions, inconsistent histories, and unnecessary resource consumption.
Now a single gateway manages each conversation on its host.
The CLI, terminal UI, desktop, browser, editors, bots, and scheduled tasks connect to the same session.
Your conversation stays consistent, no matter where you access it.
3. /Rewind: Go back without destroying your work
Ever wish you could return to an earlier message, change your instructions, and try a completely different approach? Now you can.
Use: /rewind
Or press Escape twice in the terminal interface.
Choose an earlier message, edit it, and start a new branch of the conversation.
The original session remains intact.
4. Group Chats with automatic agent routing
This might be one of my favorite improvements.
Hermes Group Chats can now decide which agent should handle a message.
Instead of manually addressing every specialist, a lead agent or automatic router can select the appropriate participant.
The system can route messages, maintain shared context, and prevent agents from getting stuck replying to one another forever.
Because apparently, even AI agents need someone to stop an unnecessary meeting.
5. Spending limits for paid tools
This is a particularly welcome improvement.
Paid MCP tools can now expose their costs before execution.
Hermes shows the price of a tool call, tracks the conversation's spending, and asks for approval when a configured budget is exceeded.
Even in YOLO mode. So your AI agent can't casually spend your entire API budget generating 3D models and cinematic videos without asking.
6. A much larger plugin ecosystem
The Hermes plugin catalog has grown from approximately 600 to more than 1,050 integrations.
New official plugins include:
Meshy for AI-powered 3D generation
ElevenLabs for music and sound effects
X Chat integration
Reverse-engineering tools
Additional sandboxing and automation capabilities
Plugins can also be updated together, with optional automatic updates.
7. Smaller, faster, and more efficient
Nous Research has also made substantial improvements under the hood.
Approximately 1,570 fewer prompt tokens per one-shot call
Shared prompt caching for recurring scheduled jobs
Faster file operations
Improved memory management for long-running gateways
A smaller built-in skill set, reduced from 58 to 44
Better performance for large conversation histories
These optimizations are particularly useful if you run Hermes continuously rather than launching it for occasional tasks. And here's the bigger picture.
We're gradually moving beyond AI agents that live inside one chat window.
The next generation of agent platforms needs shared sessions, persistent context, coordinated specialists, cost controls, and interfaces that work across multiple devices.
Hermes v0.23.0 is moving in exactly that direction.
Not just a smarter chatbot.
A more complete environment for running AI agents as an ongoing system.
Of course, this is a major release, so existing users should check the migration notes. Some browser tools have changed, and several built-in capabilities have moved into plugins.
But I particularly like the philosophy behind this update.
The future of personal AI might not be one giant agent doing everything.
It might be a small team of specialized agents working together, remembering what they've done, and knowing when to ask for your approval.
And ideally, not setting your API budget on fire while you're sleeping.
Hermes Agent v0.23.0: Full Release Notes
https://github.com/NousResearch/hermes-agent/releases/tag/v0.23.0
Hermes Agent on GitHub:
https://github.com/NousResearch/hermes-agent
#HermesAgent #NousResearch #AIAgents #AgenticAI #OpenSourceAI #AIAutomation #ClaudeCode
YouTube / ABVX