Hermes Agent v0.12.0

"The Curator" Release

Released April 30, 2026

1,096 commits · 550 PRs · 217 contributors

🎯 Executive Summary

  • Autonomous Curator — self-maintaining skill library
  • Self-improvement loop — rubric-based, runtime-aware
  • 4 new inference providers — GMI Cloud, Azure AI Foundry, MiniMax OAuth, Tencent Tokenhub
  • 18th messaging platform — Tencent Yuanbao
  • Microsoft Teams — first pluggable gateway platform
  • ~57% TUI cold-start reduction

🧠 Autonomous Curator

  • Background agent on gateway cron ticker (7-day cycle default)
  • Grades, consolidates, and prunes skills automatically
  • Writes per-run reports to logs/curator/run.json + REPORT.md
  • hermes curator status ranks skills by usage
  • Defense-in-depth gates protect bundled/hub skills
  • Pinned skills block curator writes

🔄 Self-Improvement Loop

  • Class-first review — rubric-based grading, not free-form
  • Active-update bias — prefers skills the agent just loaded
  • Runtime inheritance — provider, model, credentials propagate
  • Scoped toolsets — memory + skills only, no shell/web sprawl
  • Clean shutdown — memory providers exit properly
  • Handles references/ and templates/ sub-files

📡 New Providers

  • GMI Cloud — first-class API-key provider
  • Azure AI Foundry — auto-detection, full wiring
  • MiniMax OAuth — PKCE browser flow
  • Tencent Tokenhub — new provider salvage
  • LM Studio upgraded to first-class — auth, doctor, reasoning, live /models

💬 Platforms & Integrations

  • Tencent Yuanbao — 18th native messaging platform
  • Microsoft Teams — first pluggable gateway adapter
  • Spotify — 7 tools (play, search, queue, playlists, devices), PKCE OAuth
  • Google Meet — join, transcribe, speak, follow up via realtime OpenAI transport
  • Gateway media parity — multi-image Telegram/Discord/Slack/Mattermost/Email/Signal
  • FLAC audio + Telegram document fallback

🎨 Skills Ecosystem

  • ComfyUI v5 — official CLI + REST, moved from optional to built-in default
  • TouchDesigner-MCP — bundled by default, GLSL/post-FX/audio/geometry refs
  • Humanizer — strips AI-isms from text
  • claude-design — HTML artifact skill
  • design-md — Google DESIGN.md spec
  • Direct URL install — hermes skills install <url>
  • /reload-skills slash command

⚡ Performance & UX

  • ~57% visible TUI cold-start cut
  • Lazy agent initialization
  • Lazy imports of OpenAI / Anthropic / Firecrawl
  • mtime-cached load_config()
  • Memoized get_tool_definitions() with TTL
  • Precompiled dangerous-command patterns
  • Prompt cache TTL configurable (5m default, 1h opt-in)

🎧 Voice & TTS

  • TTS provider registry — tts.providers.<name>
  • Piper local TTS ships as native provider
  • Google Gemini TTS provider
  • xAI Grok STT provider
  • KittenTTS local provider

🖼️ Dashboard & CLI

  • Models dashboard tab — rich per-model analytics, switch main + auxiliary
  • hermes -z one-shot non-interactive mode
  • hermes update --check preflight
  • Opt-in pre-update HERMES_HOME backup
  • LaTeX rendering in TUI
  • /reload .env hot-reload
  • Pluggable busy-indicator styles

🔒 Security & Reliability

  • Secret redaction off by default — prevents patch-corruption incidents
  • Global toggle for private/internal URL resolution
  • Block agent from self-destructing gateway
  • Telegram callback authorization on updates
  • SECURITY.md added

📊 Community Response

"The Curator release — Hermes Agent now maintains itself. An autonomous background Curator grades, prunes, and consolidates your skill library on its own schedule."
  • 137 GitHub reactions (80 +1, 26 heart, 16 rocket, 15 hooray)
  • 217 community contributors in this window
  • @kshitijk4poor led with 49 merged PRs

⚠️ Pain Points & Risks

  • Secret redaction disabled by default — users must opt-in via redaction.enabled: true
  • Large changelog — 1,096 commits may hide breaking changes in peripheral tooling
  • ComfyUI built-in — increases base install size

🔄 Migration Guide

  • Run hermes update --check before upgrading
  • Opt-in backup: set pre-update backup preference
  • Review redaction.enabled if you need secret filtering
  • Check pinned skills before curator runs
  • Configure new providers (GMI Cloud, Azure AI Foundry) in hermes model

📈 Roadmap Implications

  • Hermes is evolving into a self-maintaining agent platform
  • Curator + self-improvement loop = compounding skill quality
  • Plugin gateway architecture enables rapid platform expansion
  • Remote model catalogs decouple model availability from releases
  • Vercel Sandbox signals push toward hosted execution backends

📖 Sources

Generated by Hermes Agent — May 1, 2026