Hermes Atlas

The ascii-video skill is a bundled capability in the Hermes ecosystem used to convert video, audio, images, or generative input into colored ASCII character video outputs such as MP4 or GIF. It is designed for creating retro, terminal-style, or matrix-style visual effects and audio visualizers.

Skill Metadata:

• Path: skills/creative/ascii-video
• Platforms: linux, macos, windows
• Installation: Bundled (installed by default)

Hermes Agent is an autonomous AI agent designed with a built-in learning loop. Unlike standard coding copilots or chatbot wrappers, it is designed to improve through experience by creating and refining skills, persisting knowledge, and building a model of user preferences across sessions.

Key architectural characteristics include:

• Portability: It can run on local machines, Docker, SSH, or serverless infrastructure like Daytona and Modal.
• Persistence: It can operate on remote cloud VMs, allowing interaction via messaging platforms (like Telegram) without needing a direct SSH connection.
• Learning Loop: Features agent-curated memory, autonomous skill creation, and FTS5 cross-session recall.

The v2026.4.30 release (v0.12.0) introduces significant autonomy and integration capabilities:

• Autonomy: The new Autonomous Curator manages the skill library automatically.
• Self-Improvement: An upgraded background review fork uses rubric-based logic to decide which memories and skills to save/update.
• Expanded Integrations:
  • Built-in Skills: ComfyUI v5 and TouchDesigner-MCP are now bundled by default.
  • Messaging: Support for Microsoft Teams (via plugin) and Tencent Yuanbao.
  • Media: Native Spotify tools (via PKCE OAuth) and Google Meet plugin.
• Inference Providers: Native support for LM Studio, GMI Cloud, Azure AI Foundry, MiniMax (via OAuth/PKCE), and Tencent Tokenhub.
• Performance: ~57% reduction in TUI cold start via lazy initialization and caching.

The Hermes ecosystem includes several built-in and optional skill integrations designed for specific workflows:

Built-in & Promoted Skills:

• ComfyUI v5: Official CLI + REST integration (moved from optional to built-in).
• TouchDesigner-MCP: Bundled by default; supports GLSL, post-FX, audio, and geometry references.
• Humanizer: Removes AI-typical phrasing from text.
• claude-design: Specialized HTML artifact skill.
• design-md: Implements the Google DESIGN.md specification.
• airtable: Integrated via API keys configured in .env.
• pretext: Enables creative browser demonstrations.
• spike & sketch: Tools for throwaway experiments and HTML mockups.

Integration Note: For airtable skills, ensure API keys are correctly set in your environment variables.

Plugins extend the Hermes framework by adding new tools, data sources, or behaviors. Notable plugins include:

• hermes-plugins: Provides goal management, inter-agent bridges, model selection, and cost control.
• evey-bridge-plugin: Enables context sharing between Claude Code and Hermes.
• hermes-web-search-plus: Provides multi-provider web search routing with fallbacks.
• hermes-weather-plugin: Provides professional weather data and NWS imagery.
• hermes-payguard: Handles safe USDC and x402 payments for agents.

To run multiple Hermes agents or coordinate agent swarms, use orchestration frameworks:

• mission-control: A self-hosted orchestration tool to dispatch tasks, run multi-agent workflows, and monitor spend.
• swarmclaw: A framework to build autonomous swarms with orchestration and multiple model providers.
• opencode-hermes-multiagent: Provides 17 specialized agents with structured communication protocols.

The Axolotl skill provides expert guidance for fine-tuning Large Language Models (LLMs) using YAML configurations. It supports various techniques including LoRA, QLoRA, DPO, KTO, ORPO, and GRPO, as well as multimodal support.

When to use this skill:

• Working with Axolotl configurations or features.
• Implementing Axolotl-based fine-tuning solutions.
• Debugging Axolotl code or YAML configs.
• Learning Axolotl best practices.

Skill Metadata:

• Path: skills/mlops/training/axolotl
• Version: 1.0.0
• Dependencies: axolotl, torch, transformers, datasets, peft, accelerate, deepspeed

The following foundational repositories are maintained by Nous Research and form the core of the Hermes ecosystem:

• hermes-agent: The core autonomous agent featuring persistent memory, 47 tools, 14 messaging platforms, and 6 execution backends. (Maturity: Production)
• hermes-agent-self-evolution: Optimizes skills, prompts, and code using DSPy + GEPA. (Maturity: Beta)
• hermes-paperclip-adapter: Allows running Hermes as a managed employee in Paperclip company systems. (Maturity: Beta)
• autonovel: An autonomous novel-writing pipeline. (Maturity: Beta)
• Hermes-Function-Calling: Provides function calling examples and training data for Hermes LLM models. (Maturity: Production)
• atropos: A framework for RL training environments for tool-calling models. (Maturity: Beta)

The Hermes Mattermost adapter is designed for flexibility and ease of deployment:

• Self-hosted friendly: Compatible with any self-hosted Mattermost instance; no Mattermost Cloud account or subscription is required.
• Edition compatibility: Works with both Mattermost Team Edition (free) and Enterprise Edition.
• Dependencies: Uses aiohttp for HTTP and WebSocket communication, which is already included with the Hermes Agent.