Hermes Desktop Is Here: v0.16.0 'Surface Release' Ships Native App, Admin Panel, and 简体中文

TLDR: Hermes Agent shipped v0.16.0 “The Surface Release” on June 5 — a native desktop app (Electron), full web admin panel, Simplified Chinese UI, leaner skills, and NVIDIA/skills as a built-in trusted tap. This caps a sprint that spanned four releases in nine days, from the v0.15.0 refactor to Hermes Desktop going public.

Hermes Desktop: Native App Goes Public

First teased during Jensen Huang’s GTC 2026 keynote and covered by MarkTechPost, Hermes Desktop is now in public preview with v0.16.0. It’s a native Electron application for macOS, Windows, and Linux that wraps the Hermes Agent core in a full graphical interface.

What’s in the box:

• One-click installers — macOS and Windows get direct downloads; Linux uses --include-desktop during terminal install
• In-app self-update — no manual pip upgrades needed
• Drag-and-drop files + clipboard image paste — attach content without terminal wrangling
• Cmd+K command palette — quick access to any feature
• Inline model picker in the status bar — swap models mid-conversation
• Concurrent multi-profile sessions — run separate agent instances side by side
• Streaming tool output with live activity indicators and a side-by-side preview pane

The critical insight: sessions are shared across surfaces. Start a task in the desktop app, resume it in the CLI or TUI, switch back to desktop — same context, same memory, same skills. It’s one agent with multiple frontends, not a fork.

The desktop can also connect to a remote Hermes gateway over secure WebSocket with OAuth or username/password authentication, and cross-profile @session links let you jump between agent instances.

The Web Dashboard Becomes an Admin Panel

What was a basic web dashboard is now a full administrative control panel:

• Channels page — manage all connected messaging platforms from one view
• MCP catalog — browse, enable, and disable Model Context Protocol servers
• Credential management — rotate API keys and OAuth tokens through a GUI
• Webhooks — create, test, and monitor webhook subscriptions
• Memory configuration — tune recall settings, curation intervals, and storage backends
• One-click Debug Share — generate a shareable debug snapshot for troubleshooting

Simplified Chinese (简体中文) is In

v0.16.0 ships full i18n for Simplified Chinese across all desktop UI surfaces — chat, sidebar, settings, profiles, and skills. The implementation uses a typed i18n layer designed for extensibility, so additional languages can follow the same pattern without code changes.

Leaner Skills, NVIDIA Trusted Tap

The skill system got two significant upgrades:

1. Leaner default set — Redundant and dead skills (spotify, linear, kanban-codex-lane) are removed. Heavier and niche skills moved to optional installs. A new environments: relevance gate limits skills to context-appropriate scenarios.

2. NVIDIA/skills trusted tap — NVIDIA’s skill catalog is now a built-in trusted tap alongside OpenAI, Anthropic, and HuggingFace. Skills are discoverable, searchable, and auto-updating from NVIDIA’s registry.

Quick Setup via Nous Portal

Two new setup paths eliminate the configuration barrier:

• Quick Setup — sign in with Nous Portal, immediately start chatting
• Full Setup wizard — guided walkthrough for self-hosters

A new hermes portal alias streamlines authentication for existing users.

Fuzzy Model Picker Everywhere

The model picker — available in desktop, web, TUI, and CLI — now supports fuzzy matching: type v4fl to find deepseek-v4-flash. Multi-endpoint providers are grouped, and the catalog refreshes hourly. New models added: deepseek-v4-flash, MiniMax-M3 (1M context), and qwen3.7-plus.

/undo [N] — Rollback Last N Turns

A long-standing request (#21910) is finally closed. /undo [N] rolls back the last N conversation turns across CLI, TUI, Telegram, and Discord. Works on both user messages and agent responses.

Quick Bites from the Release

• 874 commits, 542 merged PRs, 399 issues closed since v0.15.2
• 170 community contributors across the v0.16.0 cycle
• Default interface choice — configure cli or tui as default, with --cli override
• CVE-2026-48710 patched — Starlette pin, SSRF off-loop hardening, subprocess credential stripping
• NVIDIA’s skills tap joins OpenAI, Anthropic, and HuggingFace as trusted skill sources

The Full May-June Sprint

This caps nine days of shipping that started with v0.15.0 “The Velocity Release” on May 28 — the 76% core refactor (16K → 3.8K lines) and 4,500× faster session search — through v0.15.1 (dashboard infinite-reload fix), v0.15.2 (plugin manifest packaging), and now v0.16.0 (desktop + admin panel).

How to Update

pip install --upgrade hermes-agent

For Hermes Desktop:

# macOS/Windows: download from hermes-agent.nousresearch.com/desktop
# Linux:
pip install hermes-agent --include-desktop

Verify:

hermes --version
# Should show: v0.16.0 (v2026.6.5)

What’s Next

With the desktop surface and admin panel live, the team is expected to focus on deeper multi-profile workflows, plugin-based agent loop customization, and maturing the Kanban Swarm platform. The i18n layer opens the door for more languages, and the web admin panel provides the foundation for team-managed Hermes deployments.

Track development on GitHub and join the community.

---

Hermes Tutorials is an independent community resource. Not affiliated with Nous Research.

📖 Related Reads

• ToolBrain — tool reviews, LLM comparisons, and AI workflow guides
• NoCode Insider — AI workflow automation with no-code tools, agents, and APIs

Cross-links automatically generated from Hermes Tutorials.