NousResearch/hermes-agent

Hermes Agent is a self-improving AI assistant from Nous Research that learns from every conversation. It creates skills from experience, improves them during use, and builds a persistent model of who you are across sessions. Version 0.6.0 dropped March 30, 2026. The core differentiator is the closed learning loop. Most AI agents start fresh every session. Hermes Agent curates its own memory, periodically reinforces knowledge, and searches past conversations using FTS5 session search with LLM summarization. It does not just remember — it actively improves its understanding of you over time through Honcho dialectic user modeling. Multi-platform access is built in: Telegram, Discord, Slack, WhatsApp, and Signal through a unified gateway. Or run it locally via a full TUI terminal with multiline editing, slash-command autocomplete, and streaming tool output. It works with 200+ models through Nous Portal, OpenRouter, OpenAI, Anthropic, z.ai/GLM, Kimi/Moonshot, and MiniMax. Hardware requirements are minimal — it runs on a $5 VPS. Serverless options (Daytona, Modal) offer near-zero idle costs with automatic hibernation. For compute-heavy work, it supports Docker, SSH, and Singularity execution environments. You can spawn isolated subagents for parallel workstreams with RPC tool access. 20.4K stars on GitHub. MIT licensed. Python-based (92.8%). The built-in cron scheduler lets you automate unattended tasks with delivery to any connected platform. Research capabilities include batch trajectory generation and RL environment integration via Tinker-Atropos. Explore more open-source AI agent frameworks or check out related AI agent tools in our directory.