Features

Persistent Memory & User Modeling

Hermes Agent features industry-leading persistent memory. Unlike traditional "stateless" chatbots, Hermes remembers everything it learns, and as you use it over time, it develops an increasingly deep understanding of you.

Core Features

• Agent-curated memory — The agent autonomously decides which information is worth remembering and periodically reminds itself to persist important knowledge
• Automatic skill generation — After completing complex tasks, the agent automatically distills the solution process into reusable skills
• Cross-session search — Uses FTS5 full-text search to search all conversation history, with LLM-powered summarization
• User modeling — Builds an ever-deepening user profile using Honcho's dialect-based user modeling technology
• agentskills.io standard — Compatible with the open skills standard, allowing skills to be shared across agents

How It Works

When you chat with Hermes Agent, it stores information at three levels:

1. Working memory — context within the current conversation
2. Persistent memory — important facts, preferences, and knowledge retained across conversations
3. Skill library — standardized workflows distilled from experience

The agent autonomously determines which information is worth persisting during conversations and reminds itself to update memories at appropriate times.

Subagents & Parallel Tasks

Hermes Agent can spawn isolated subagents to process tasks in parallel. Each subagent has its own conversation context, terminal environment, and tool set, enabling true parallel execution.

Core Features

• Isolated execution — Each subagent has its own conversation history and terminal, fully independent from others
• Parallel processing — Handle multiple tasks simultaneously for significantly improved efficiency
• Python RPC — Write Python scripts that call tools via RPC, compressing multi-step pipelines into single interactions with zero context cost
• Flexible orchestration — The main agent coordinates multiple subagents' workflows and aggregates results

Use Cases

• Simultaneously review code across multiple repositories
• Research multiple topics in parallel and compile a combined report
• Batch process data cleaning tasks
• Break complex multi-step workflows into parallel subtasks

Multi-Platform Integration

Hermes Agent connects to multiple messaging platforms through a unified gateway process. You can start a conversation on Telegram and continue it on Discord with full context continuity.

Supported Platforms

Quick Setup

hermes gateway setup    # configure messaging platforms
hermes gateway start    # start the gateway

Once the gateway is running, send a message to your bot on any platform to start chatting.

Scheduled Automations

Hermes Agent includes a built-in cron scheduler that supports natural language task scheduling. You can have it send daily news digests, generate weekly reports, or schedule data backups — all running unattended.

Core Features

• Natural language scheduling — Describe your task schedule in everyday language, no need to memorize cron syntax
• Multi-platform delivery — Scheduled task results can be delivered to any connected platform
• Persistence — Scheduled tasks survive agent restarts
• Flexible configuration — Supports one-time and recurring tasks

Usage Examples

# Set scheduled tasks directly in conversation using natural language
"Send me a daily news digest every morning at 9 AM"
"Back up the /data directory every Monday"
"Check server status every hour and send a report"

Skills System & Learning Loop

Hermes Agent features a complete skills learning loop — this is the core differentiator from other agents. It doesn't just use predefined skills; it creates new ones from experience and continuously improves them through use.

Skills Ecosystem

• 77 built-in skills — covering programming, search, document processing, multimedia, and more
• 45 optional skills — advanced features requiring additional dependencies
• 521 community skills — contributed by the community, continuously growing
• 16 categories — AI programming, system administration, research analysis, multimedia creation, and more

Learning Loop

1. Task execution — The agent completes a complex task
2. Skill extraction — Automatically analyzes the solution process and distills it into a reusable skill
3. Skill storage — Saved to the local skill library
4. Self-improvement — When using the same skill later, the agent optimizes the workflow based on feedback
5. Knowledge persistence — Through self-reminder mechanisms, ensuring experience is never forgotten

Browsing Skills

/skills              # browse all available skills
/<skill-name>        # directly invoke a specific skill

You can also visit the Skills Hub to browse the complete skills catalog.