- Nous Research just dropped Hermes Agent v0.11.0 — 1,556 commits, 761 merged PRs, a full React/Ink TUI rewrite, 5 new LLM providers, GPT-5.5 via Codex OAuth, and QQBot as the 17th messaging platform.
- The self-improving open-source agent doubles down on extensibility.
TL;DR
Hermes Agent v0.11.0 (tag v2026.4.23) is the largest release Nous Research has shipped since v0.9.0. The numbers: 1,556 commits, 761 merged PRs, 1,314 files changed, 224,174 insertions, 29 direct contributors (290 including co-authors). Three headlines — a brand-new React/Ink TUI, a pluggable transport layer with 5 new native LLM providers, and QQBot as the 17th messaging platform. The repo sits at 115k stars and stays MIT-licensed.
What's new
v0.10.0 only shipped the Nous Tool Gateway, so v0.11.0 folds in roughly two weeks of deferred work across the whole stack. The release has three pillars:
- TUI v2 — a full React/Ink rewrite of the interactive CLI, with a Python JSON-RPC backend (
tui_gateway). Around 310 commits went into the UI alone. - Pluggable transport layer — format conversion and HTTP transport extracted into
agent/transports/, enabling native support for Anthropic Messages, OpenAI-compatible, AWS Bedrock Converse, and OpenAI Responses APIs. - QQBot — Hermes now runs on 17 messaging platforms, with QQ Official API v2 integration matching the WeCom/Weixin feature set.
Why it matters
Most coding copilots (Cursor, Windsurf, Claude Code) are IDE-bound and stateless per project. Hermes Agent keeps betting on a different shape: a long-lived autonomous agent that lives on your server, builds persistent memory across sessions, and auto-creates skills from experience via its built-in learning loop. v0.11.0 hardens that bet — the plugin surface now supports slash commands, tool dispatch, execution blocking, result transformation, terminal output rewriting, custom image backends, and dashboard extensions. Hermes is graduating from CLI to extensible platform.
Technical facts
The new TUI v2 ships with a sticky composer, live streaming with OSC-52 clipboard support, stable picker keys, a status bar showing per-turn stopwatch and git branch, a /clear confirm dialog, a light-theme preset, and a subagent spawn observability overlay.
Five new native LLM inference paths landed, along with a batch of new model support:
| Category | What shipped |
|---|---|
| New providers | AWS Bedrock (Converse), NVIDIA NIM, Arcee AI, Google Gemini CLI OAuth, Vercel ai-gateway |
| New models | GPT-5.5 via Codex OAuth, Kimi K2.6, Xiaomi MiMo v2.5-pro, GLM-5V-Turbo |
| Image backends | FAL multi-model picker, Recraft V4 Pro, GPT Image 2, xAI Grok-Imagine |
| Messaging | QQBot (QR scan-to-configure, streaming cursor, emoji reactions, DM/group gating) |
| Dashboard | i18n (EN + ZH), live-switching theme system, plugin tabs/widgets, Vercel deploy |
The fix column alone racked up 482+ PRs, covering streaming cursor artifacts, Unicode encoding errors, memory tool on Windows, and compression-exhaustion loops.
Comparison with recent releases
| Aspect | v0.9.0 | v0.10.0 | v0.11.0 |
|---|---|---|---|
| Focus | Base CLI agent | Tool Gateway only | Full-stack overhaul |
| TUI | v1 | v1 | v2 (React/Ink) |
| LLM providers | ~195 models | same | 200+, 5 new native |
| Messaging platforms | 16 | 16 | 17 (+QQBot) |
| Transport | Hardcoded | Hardcoded | Pluggable |
Use cases
- Solo devs & indie hackers — run Hermes on a $5 VPS as a 24/7 personal agent reachable via Telegram, Discord, Slack, WhatsApp, Signal, or email.
- Teams in the QQ/WeChat ecosystem — with QQBot landing next to WeCom/Weixin, Hermes is now the first major open agent natively embedded in the Chinese messaging stack.
- Model stackers — route reasoning to GPT-5.5, coding to Kimi K2.6, image to Grok-Imagine, all from one agent via the new transport layer.
- Plugin authors — the expanded plugin surface turns Hermes into a publishable platform, not just a chat client.
- Researchers — batch trajectory generation and RL environment support for training custom agents.
Limitations & pricing
Hermes Agent is MIT-licensed and free. You pay only for the underlying model API (OpenAI / Anthropic / etc.) and optional hosting — local Docker, SSH, Singularity, Modal, or a $5 VPS all work. Caveats: it's self-hosted, not managed SaaS; the learning loop quality depends on your own usage data, so cold-start carries less context than a matured deployment; TUI v2 is fresh and some edge-cases are still being polished.
What's next
Nous Research is shipping on a weekly cadence (calendar-versioned as v2026.4.X), so expect deeper learning-loop tooling, more native transports, additional messaging platforms, and tighter integration with the Skills Hub (agentskills.io) for community-shared skills. The top contributors for this release were @kshitijk4poor (49 PRs) on the transport refactor and Step Plan provider, @OutThisLife (31 PRs) on TUI polish, stopwatch, and the observability overlay, and @helix4u (11 PRs) on voice features and MCP stability. With Hermes now running on everything from a $5 VPS to a GPU cluster, and reachable from 17 messaging platforms, Nous is building an agent that is less a tool you open and more a presence you talk to.
Nguồn: GitHub release notes, hermes-agent.nousresearch.com, Wes Roth on X.
