šŸ¤–HermesBlog
Hermes Version History Ā· Part 18/9/2026

Hermes v2026.8.3 Update: Hermes Agent v0.20.0 (2026.8.3)

Hermes v2026.8.3 (2026-08-03) update: Hermes Agent v0.20.0 (2026.8.3)

This release turns Hermes from a helpful tool into a true conversational partner—it can now talk back in real time, listen for your voice from across the room, and even carry messages to other agents. Think of it as upgrading from sending letters to having a live phone call.

What’s New: The Voice Revolution

The biggest change is that Hermes finally speaks like a human, not a robot reading a script. In the past, voice mode meant you asked a question, waited for the full answer to generate, and then listened to one long audio file. Now, Hermes talks clause-by-clause as it thinks—like a friend who thinks out loud. You can interrupt mid-sentence, and it stops, listens, and adjusts. It even knows when you’re busy and won’t talk over you. This works in the terminal, on the desktop app, and through messaging platforms.

Even better, you can set a custom wake word (like ā€œHey Hermesā€ or anything you pick). Detection happens on your device, so nothing is sent to the cloud while it’s just waiting. You can even assign different wake words to different profiles—say ā€œWorkā€ to get your coding assistant, ā€œHomeā€ for your personal planner. And saying ā€œstopā€ ends the conversation on any device, no keyboard needed.

Messaging platforms get a voice too. Send a voice note on WhatsApp, Feishu, DingTalk, LINE, QQ, or Weixin, and Hermes transcribes it, answers, and replies with a spoken message. It’s like having a multilingual assistant who never sleeps.

Carrying Word to Other Agents

Hermes now supports A2A v1.0 (Agent-to-Agent). Imagine two colleagues passing notes—Hermes can now formally hand off tasks, share context, and receive results from other AI agents. It’s no longer an island; it’s part of a team.

Smarter and More Reliable

  • Grounded research: Hermes now cites its sources with verifiable links and does fact-checking before answering. No more guessing where information came from.
  • Self-healing tools: If a tool fails (like a web search timeout), Hermes retries with a different approach instead of making the model guess. It’s like a chef who checks the oven before blaming the recipe.
  • Smarter memory: Compression is gentler—it keeps important details longer and forgets trivial ones, so long conversations feel less like a game of telephone.

Desktop and CLI Upgrades

The desktop app is now a full platform: you get live previews for artifacts (like charts or code), a plugin SDK for custom extensions, and quick-entry from anywhere with multiple windows. The CLI got power commands like ! for shell mode, /init to start projects, /diff to see changes, and /context to inspect what Hermes remembers.

What This Means for You

If you’ve ever wanted to just talk to your computer while cooking, driving, or pacing around your office—this is it. Hermes is no longer a tool you type at; it’s a colleague you converse with. The research citations mean you can trust its answers for work. And the self-healing tools mean fewer ā€œI’m sorry, I couldn’t do thatā€ moments.

This is the release where Hermes stops being a clever script and starts being a presence in your workflow. Try the voice mode first—it’s the closest thing to magic we’ve shipped yet.

šŸ“Ž Official release notes: https://github.com/NousResearch/hermes-agent/releases

šŸ“– Official Docs

This article is based on the official Hermes Agent documentation:GitHub ›/releases/tag/v2026.8.3