The hermes herald release is the point where Hermes stops feeling like a chat window and starts feeling like a colleague. Herald is Hermes Agent V0.20, shipped with 647 contributors behind it, and while the patch notes run long, three capabilities do almost all of the heavy lifting: real voice conversation, agent-to-agent communication, and a proper home for everything your agent creates.
Quick grounding first. Hermes is the open-source AI agent that lives on your own computer, and it is model-agnostic — you pick the brain, Hermes supplies the hands. You update to Herald from the Manage section of your Agent OS, and everything below is available the moment the update lands.
This page goes deep on the big three and what they unlock. If you want every line item — every fix and minor toggle — that lives in the full V0.20 changelog. Here, we go deep rather than wide.
What The Hermes Herald Release Actually Changes
Most agent updates add features. Herald changes interfaces — the routes between you and your agent, and between your agent and every other agent you run.
Before Herald, the loop was simple: you typed, Hermes typed back. After Herald, you can talk to Hermes from across the room, Hermes can talk to your other agents without you playing messenger, and everything it produces lands in one reviewable panel instead of being scattered across platforms.
Everything in this breakdown is drawn from Julian Goldie's hands-on testing of the release — the routine he runs on every agent update under the Goldie Bench banner. That is his personal testing process, not an objective industry benchmark: one operator's field notes.
The big three at a glance:
| Capability | What it unlocks |
|---|---|
| Talk to Hermes — real-time voice | Phone-call conversation: the voice speaks while the response generates, you can cut it off mid-sentence, and a wake word gives you hands-free control from across the room |
| A2A — the agent-to-agent protocol | Hermes can discover other agents, talk to them and be driven by them — so every workflow, skill and API across your agents can be orchestrated from one conversation |
| Artifacts on Hermes Desktop | One home for everything you have created with Hermes across all platforms, with sandboxed live previews for running and testing HTML apps safely |
Headline One: Talk To Hermes Like It Is A Phone Call
Voice features in AI tools usually follow a stilted pattern: you speak, you wait, a canned voice reads out a finished answer. Herald breaks that pattern in one specific way — the voice speaks while the response is still generating. There is no dead air between your question and the reply.
The part that seals it is interruption. Cut Hermes off mid-sentence just by speaking, and it stops, listens and adjusts — like a phone call. You do not press a button or wait politely for a paragraph it started before your correction. You talk over it, and the conversation bends to follow you.
Hermes Apollo: voice plus a visible transcript
Julian's preferred way to use this is Hermes Apollo inside his Agent OS. He speaks in real time, Hermes responds with a voice, and the full transcript stays visible on screen. You get the speed of speaking with the accountability of text — you can always scroll back and check what was asked and answered.
The wake word: hands-free from across the room
Herald also ships a wake word. Say "Hey Hermes" — or a custom phrase you choose — from across the room, and the agent wakes and responds. Pair that with Hermes Desktop, or the Agent OS on another monitor, and you get genuinely hands-free control while you work on something else. When you want it running, the wake word setup walks through it step by step.
Voice notes on every chat platform
The third voice piece is quieter but useful daily. Voice notes sent to Hermes on any connected chat platform — WhatsApp, DingTalk, LINE and the rest — now get automatic text-to-speech replies. Send a rambling voice note from the car; a spoken answer comes back in the same thread.
📺 Watch: Hermes Agent V0.20 Just Changed AI Agents Forever!
Headline Two: A2A Is The One To Watch
The second headline is a new plugin implementing the agent-to-agent protocol, shortened to A2A. It gives Hermes three new verbs: it can discover other agents, it can talk to them, and it can be driven by them.
Read that last verb again, because it cuts both ways. Hermes can orchestrate your other agents — and another agent can orchestrate Hermes. Neither side needs you copying and pasting between windows.
Julian's real-world use is the agent mastermind inside his Agent OS. He posts one message, and all of his agents — Hermes included — reply, bounce ideas off each other, and push the good ones into his idea pipeline. One prompt in, a panel discussion out, and the useful output lands somewhere actionable instead of dying in a chat log.
The strategic point is bigger than the demo. Every workflow, skill and API you have plugged into each individual agent becomes reachable through whichever agent you happen to be talking to. Your research agent's tools, your content agent's publishing hooks, your Hermes automations — A2A means you stop being the human router standing between them.
Julian's verdict from his testing: this is the thing to watch in the entire release. Voice is the feature you notice on day one, but if A2A keeps improving, controlling all of your agents becomes far easier — one conversation steering a whole stack.
If you want the Herald-era Hermes making you money, check out the AI Profit Boardroom — the Agent OS with Apollo, Oracle and the agent mastermind already wired in, so the big three are working on your business from the day you switch it on. → Get the Agent OS running your agents
📺 Watch: Run Hermes Agent Free Forever, Here's how...!
Headline Three: Artifacts Get A Proper Home
The third headline fixes a quiet annoyance. Everything you create with Hermes — pages, images, small apps — used to live wherever it happened to be made. Herald gives it all one home: the artifacts view on Hermes Desktop.
It gathers your creations from every platform you use Hermes on. The images Julian generated through the Higgsfield MCP, for example, show up right there in the panel — no digging through folders or scrolling back through old chats to find them.
The clever part is sandboxed live previews. HTML apps you have built with Hermes can be run and tested safely inside the artifacts panel itself. That turns Hermes Desktop into a small workshop: build a tool in conversation, open artifacts, try it immediately — without handing untested code the run of your machine.
📺 Watch: Higgsfield MCP Makes Hermes Agent Unstoppable
The Supporting Cast, Briefly
The big three are the story, but Herald's smaller improvements deserve a quick nod. The changelog linked above covers each one in full:
- Grounded citations — research answers now carry their sources on every response.
- Mid-turn corrections — type while Hermes is working and the turn redirects; no /stop needed.
- Smarter approvals — risky actions get more sensible approval prompts.
- Better compression — long sessions keep the recent conversation instead of losing it.
- Speed — first token in 0.9 seconds versus 4.3 before, and a telemetry gate running 54 times faster.
- Security fixes — the unglamorous work that keeps a local agent trustworthy.
Hermes Herald FAQ
What is the Hermes Herald release?
Herald is the codename for Hermes Agent V0.20, an update shipped with 647 contributors. Its three headline capabilities are real-time voice conversation, the A2A agent-to-agent plugin, and an artifacts home on Hermes Desktop.
How do I update Hermes to the Herald release?
Open the Manage section of your Agent OS and run the update from there. The new capabilities are available as soon as the update completes.
Can I interrupt Hermes while it is speaking?
Yes. The voice speaks while the response is still generating, and if you talk over it mid-sentence it stops, listens and adjusts — the same rhythm as a phone call.
What does A2A actually let Hermes do?
Three things: discover other agents, talk to them, and be driven by them. In practice, Hermes can orchestrate the workflows, skills and APIs attached to your other agents — or act as one voice inside a bigger multi-agent setup like Julian Goldie's agent mastermind.
Do artifacts only show work made on Hermes Desktop?
No. The artifacts view lives on Hermes Desktop, but it collects everything you have created with Hermes across all platforms — and its sandboxed previews let you run HTML apps safely inside the panel.
The Bottom Line
The Herald release earns its codename: this is Hermes announcing itself out loud. Voice makes the agent conversational, artifacts make its output visible and testable in one place, and A2A — the sleeper of the three — positions Hermes as the conductor of every other agent you run.
The move now is simple. Update from the Manage section, say "Hey Hermes" to feel the voice difference, then connect a second agent over A2A and watch the orchestration happen. The operators who learn to run fleets of agents from one conversation are going to look very smart when this protocol matures.
