Here’s the full picture. Hermes V0.20 — the Herald release — just dropped, and it’s a big one: real-time voice conversations, cited research, agent-to-agent protocol, sandboxed artifacts and serious speed gains, shipped by 647 contributors.
I updated the day it landed. Here’s the full changelog walkthrough and how to get it.
Last updated: August 2026.
Key takeaways
- Talk to Hermes: real phone-call-style voice — it speaks while generating and you can interrupt mid-sentence.
- A2A plugin: Hermes now speaks the agent-to-agent protocol — discovering, talking to and driving other agents.
- Grounded citations: research with sources on every claim, Perplexity-style.
- Speed: first token ~0.9s (was 4.3s); telemetry gate 54x faster; desktop much snappier.
- Update via Manage → Update in the Agent OS — get it pre-configured in the AI Profit Boardroom.
What Is Hermes V0.20?
Hermes is the free, open-source agent, and V0.20 — codenamed Herald — is its newest release. Where V0.19 Quicksilver was about raw speed, Herald is about conversation and connection: your agent now talks like a phone call, backs its research with sources, and speaks to other agents directly.
Updating takes one click: Manage → Update Hermes in the Agent OS, or hermes update in the terminal.
The Full Changelog, Feature by Feature
Talk to Hermes
The headline. Hermes answers while it thinks, speaking as the response generates — and you can cut it off mid-sentence just by talking, like interrupting a call. It stops, listens and adjusts. Paired with the wake word (“Hey Hermes” from across the room), it’s genuinely hands-free.
Voice on every platform
Send Hermes a voice note on WhatsApp, DingTalk, Line and more — it auto-replies with text-to-speech, platform-aware.
Grounded citation skill
Deep-research answers now cite sources for everything — the anti-hallucination upgrade. Ask for research and you get receipts, not vibes.
Desktop artifacts
Hermes Desktop now shows everything you’ve created across platforms — my Higgsfield images appeared automatically — with sandboxed live previews for HTML apps, which is both convenient and safer.
A2A: agent-to-agent
A new plugin implements the agent-to-agent protocol: Hermes can discover, talk to and be driven by other A2A-compatible agents. In my agent mastermind, one message gets all my agents replying to each other and adding ideas to the pipeline.
Mid-turn corrections
See it heading the wrong way? Just type the correction — no /stop and re-explain. The active turn redirects in real time.
Compression, approvals, speed
Smarter conversation compression (recent context always survives), smart approvals on risky actions like file deletion, first token in ~0.9 seconds, a 54x faster telemetry gate, plus security fixes — from 647 contributors.
The Speed Story in Numbers
| Metric | Before | V0.20 |
|---|---|---|
| First token | ~4.3s | ~0.9s |
| Telemetry gate | Baseline | 54x faster |
| Desktop feel | Sluggish | Much snappier |
Pair it with a fast new model like DeepSeek V4 Flash and things that took minutes — like /learn ingesting a guide into a skill — now take seconds.
What Matters Most
If you only track one thing from V0.20, make it A2A. Agents that can discover and drive each other is the foundation for everything multi-agent — orchestration without glue code. The voice features are the fun part; the protocol is the future. Full context in my Herald release deep-dive and agentic OS for Hermes.
Get It Pre-Configured
The Agent OS with V0.20, Apollo voice, Oracle, the outreach tool, studio and memory system all wired in is inside the AI Profit Boardroom.
New here? Start free with my AI Money Lab community (free AI course + 1,000+ AI agents), or grab a free strategy session.
FAQ
What is Hermes V0.20?
The Herald release of the free, open-source Hermes agent — real-time voice conversation, grounded citations, agent-to-agent protocol, desktop artifacts, mid-turn corrections and big speed gains.
How do I update to V0.20?
One click: Manage → Update in the Agent OS, or run hermes update in the terminal.
What is Talk to Hermes?
Phone-call-style voice: Hermes speaks while generating and you can interrupt mid-sentence by talking — it stops, listens and adjusts.
What is the A2A plugin?
An implementation of the agent-to-agent protocol — Hermes can discover, talk to and be driven by other compatible agents, enabling real multi-agent orchestration.
How much faster is V0.20?
First token dropped to ~0.9 seconds (from 4.3s), the telemetry gate is 54x faster, and the desktop app is noticeably snappier.
The Bottom Line
Hermes V0.20 makes your agent conversational, cited, connected and fast. Update in one click — or get it pre-configured in the AI Profit Boardroom.
