NousResearch/hermes-agent
Open-source, self-improving AI agent by NousResearch that builds persistent memory and auto-generates skills from experience. Unlike IDE copilots or chatbot wrappers, Hermes implements a closed learning loop — creating procedural skills across sessions, building a model of your preferences and projects, and operating simultaneously across Telegram, Discord, Slack, WhatsApp, Signal, and CLI from a single gateway. Supports 200+ LLM models via OpenRouter with no vendor lock-in.
Why It Matters
Hermes is the first mainstream open-source agent to treat persistent memory and self-improvement as first-class primitives, not afterthoughts. Developers get an agent that genuinely learns their codebase and workflows over time — automating recurring tasks via natural-language cron scheduling, delegating work to parallel subagents, and sharing generated skills publicly via the Skills Hub at agentskills.io. It's model-agnostic, MIT-licensed, and installs in under two minutes with a single curl command.