Hermes Agent, the open-source AI runtime from Nous Research (now 127K+ stars on GitHub, see our complete guide to AI agents), has shipped v0.12.0 “The Curator” — a release that marks a major step toward self-maintaining agents. Published on April 30, 2026, this release packs 1,096 commits, 550 merged PRs, and contributions from 213 community members since v0.11.0 one week earlier.
✨ The Autonomous Curator
The headline feature is hermes curator — a background agent that runs on the gateway’s cron ticker (default: every 7 days) and autonomously manages your skill library:
- Grades and prunes skills — evaluates quality, consolidates related skills, and archives dead ones
- Per-run reports — writes detailed results to
logs/curator/run.json+logs/curator/REPORT.md - Consolidated vs. pruned classification — uses both model judgment and heuristics to decide what to keep
hermes curator status— ranks skills by usage (most-used / least-used), so you know what’s earning its keep- Defense-in-depth gates — bundled and hub skills are protected from mutation
Conveniently unified under auxiliary.curator — pick the curator’s model in hermes model and manage it from the web dashboard.
🔁 Self-Improvement Loop — Substantially Upgraded
The background review fork — the core mechanism where Hermes decides what memories and skills to save after each turn — received a major overhaul:
- Rubric-based scoring replaces free-form “should this update?” prompts
- Active-update bias — prefers updating the skill the agent just loaded
- Proper runtime inheritance — provider, model, and credentials actually propagate to the review fork
- Scoped toolset — the review fork is restricted to memory + skills (no shell, no web)
- Clean context — prior-turn tool messages excluded from the review summary
🧩 New Integrations & Bundled Skills
ComfyUI v5 — Now Built-In
ComfyUI has been promoted from optional to bundled-by-default, with official CLI + REST support and hardware-gated local install. (#17610)
TouchDesigner-MCP — Bundled by Default
Expanded with GLSL, post-FX, audio, geometry, and 9 new reference docs. (#16753)
Spotify — Native Tools + Wizard
7 native tools (play, search, queue, playlists, devices) behind PKCE OAuth, with an interactive setup wizard and bundled skill. (#15121). For a broader look at the Hermes Agent ecosystem, see our ultimate guide to agent frameworks.
Google Meet Plugin
Full pipeline to join calls, transcribe, speak, and follow up — powered by realtime OpenAI transport + a Node bot server. (#16364)
New Skills
- Humanizer — strips AI-isms from text (#16787)
- claude-design — HTML artifact skill (#16358)
- design-md — Google’s DESIGN.md spec skill (#14876)
- airtable — salvaged skill with API key wiring (#16291)
🏗️ 4 New Inference Providers
- GMI Cloud — first-class API-key provider (#16663)
- Azure AI Foundry — with auto-detection (#15845)
- LM Studio — upgraded from custom-endpoint alias to first-class provider with dedicated auth, doctor checks, reasoning transport, and live
/modelslisting (#17102) - MiniMax OAuth — PKCE browser flow (#17524)
- Tencent Tokenhub — new provider (#16960)
📱 Platforms & Gateway
- Microsoft Teams (19th platform) — ships as a plugin via the new pluggable gateway adapter system (#17751)
- Tencent 元宝 (Yuanbao) (18th platform) — native adapter with text + media delivery (#16298)
- Native multi-image sending across Telegram, Discord, Slack, Mattermost, Email, and Signal (#17909)
⚡ Performance & UX
- TUI cold start cut ~57% — via lazy agent init, lazy imports, mtime-cached config loading, and memoized tool definitions (#17190)
hermes -zone-shot mode — non-interactive prompt execution (#15702)- Remote model catalog manifest — OpenRouter + Nous Portal models pulled from a remote manifest so new models show up without a release (#16033)
- Secret redaction off by default — prevents the long-standing patch-corruption issue (#16794)
- Pluggable TTS provider registry + Piper local TTS — native local TTS ships as a provider (#17843)
📦 Also Landing Today: /goal Persistent Cross-Turn Goals
Hot on the heels of v0.12.0, the team has already shipped the /goal command — Hermes’ take on the “Ralph loop” (inspired by Codex CLI 0.128.0). Set a standing objective and Hermes keeps working toward it across turns until completion, pausing only when you intervene. (#18262)
📊 By the Numbers
- 1,096 commits since v0.11.0
- 550 merged PRs
- 1,270 files changed · 217,776 insertions
- 213 community contributors (including co-authors)
- 46 reactions on the release (25 🎉, 8 ❤️, 7 👍, 6 🚀)
With autonomous skill curation, a burgeoning plugin ecosystem, and weekly releases that ship hundreds of PRs, Hermes Agent is rapidly maturing into one of the most extensible AI agent platforms available. The Curator release — where the agent starts maintaining itself — marks an inflection point.
Full changelog: v2026.4.23…v2026.4.30