Version Highlights
Hermes v2026.8.3 update: easy-to-understand summary of new features
Version Highlights
| Item | Details |
|---|---|
| Version | v0.20.2 (tag: v2026.8.16) |
| Release Date | August 16, 2026 |
| Official Classification | Patch Release — consolidating ~397 merged improvements since v0.20.1 to provide a stable build for downstream users |
In plain terms: this is a “best-of-everything” stable release that packages nearly a thousand code changes from the past three days into one tidy update, so you can use it with less hassle.
What’s New, Explained Simply
Your Little Assistant No Longer Drops Messages Mid-Errand
Previously, when you asked Hermes to execute a complex multi-step task, the messages between intermediate sub-tasks would occasionally get “lost” — like sending a messenger on an errand and having them forget halfway. This update specifically fixes that. Now, message passing between sub-tasks is far more reliable, and your instructions reach every step intact, without dropping the ball along the way.
Your Workspace Auto-Resumes After Your Computer Wakes Up
Ever close your laptop, open it back up later, and find your task cards stuck or frozen? This update lets Hermes automatically restore your in-progress sessions after your machine wakes from sleep — just like a bookmark that picks up exactly where you left off, no manual searching required.
Better Memory: The AI Remembers Where You Left Off
The “memory” feature has been improved. Previously, when switching between different projects or contexts, the AI might forget preferences you’d set earlier. Now your settings are saved under separate “personal profiles,” so switching contexts won’t mix things up — like everyone having their own notebook instead of sharing one.
It Can Read Your Handwriting: Photos Become Fonts
Here’s a fun new trick: snap a photo of your handwritten text, and Hermes will generate an installable font file for you. You can then type on your computer using your own handwriting — perfect for journaling, designing greeting cards, or just adding a personal touch to your documents.
Smarter Updates: Your Custom Tweaks Won’t Be Overwritten
If you’ve manually modified any skill files, this update will automatically skip them rather than forcibly overwriting your customizations. Unless you explicitly choose “force update,” your personalized settings stay safe and sound.
Safer Command Execution
When you ask Hermes to run commands containing special symbols in chat, it can now more accurately determine which ones are safe — no false alarms, and no letting dangerous operations slip through. Think of it as a smarter doorman who can tell the difference between a delivery person and a suspicious character.
New “Worktree” Feature: Branch Out While You Code
Inspired by Copilot CLI, you can now create independent “worktrees” (git worktrees) directly from your conversation — essentially opening parallel, isolated spaces within the same project. Great for experimenting with multiple approaches at once, without worrying about breaking your main version.
Summary
This update doesn’t come with flashy new interfaces, but every change addresses a real “minor annoyance” in daily use: no lost messages, wake-from-sleep recovery, memory that doesn’t get mixed up, updates that don’t overwrite your tweaks, and safer command execution. For everyday users, this means Hermes Agent is more reliable, more attuned to you, and more hassle-free. It’s like having a thoughtful butler who quietly tidies up all the little inconveniences in your life.
📎 Official Release Notes: https://github.com/NousResearch/hermes-agent/releases