Hermes Agent is an open-source self-improving AI agent created by Nous Research, released under the MIT License (github.com/NousResearch/hermes-agent). It stands out with a built-in closed learning loop — the agent creates new skills from experience, improves them during use, nudges itself to persist knowledge, and builds a deepening model of the user across sessions. It is not an IDE plugin nor a thin wrapper around a single API — it is an autonomous agent that gets more capable the longer it runs. Current public version: Hermes Agent v0.18.2.
Key product characteristics: (1) Closed learning loop — agent-curated memory with periodic nudges, autonomous skill creation, skill self-improvement during use, FTS5 cross-session recall with LLM summarization, [Honcho](https://github.com/plastic-labs/honcho) dialectic user modeling; (2) Runs anywhere — 6 terminal backends: local, Docker, SSH, Daytona, Singularity, Modal. Daytona and Modal offer serverless persistence (the environment hibernates when idle, cost near zero); (3) Multi-platform surfaces — CLI, Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Mattermost, Email, SMS, Microsoft Teams, Google Chat, Home Assistant, DingTalk, Feishu, WeCom, Weixin, QQ Bot, Yuanbao, BlueBubbles — 20+ platforms from one gateway.
Product characteristics continued: (4) Built by model trainers — creators are Nous Research (the lab behind Hermes, Nomos, Psyche models). Compatible with Nous Portal, OpenRouter, OpenAI, or any endpoint; (5) Scheduled automations — built-in cron with delivery to any platform; (6) Delegation and parallelism — spawn isolated subagents for parallel workstreams; Programmatic Tool Calling via `execute_code` collapses multi-step pipelines into a single inference call; (7) Open standard skills — compatible with agentskills.io. Skills are portable, shareable, and community-contributed (Skills Hub); (8) Full web control — search, extract, browse, vision, image generation, TTS — one Nous Portal subscription bundles them; (9) MCP support — connect to any MCP server for extended tool capabilities; (10) Research-ready — batch processing, trajectory export, RL training with Atropos.
Distribution model: Hermes Agent is free as software (MIT License). One-line shell install on Linux/macOS/WSL2/Android (Termux) or PowerShell on Windows; native Desktop installers are also available (.dmg for macOS, .exe for Windows). The paid tier is Nous Portal — a subscription portal with Free/Plus/Super/Ultra plans providing monthly credits for 300+ models and four Tool Gateway tools (web search, image generation, TTS, browser) via a single OAuth. Alternatively, connect the agent to any backend: OpenAI, Anthropic, OpenRouter, or local models via Ollama/LM Studio.
Architecture and positioning: Hermes Agent runs on any model foundation, but is closest to Nous Research's own Hermes family. It competes with: Claude Desktop (Anthropic, closed source, macOS/Windows only), ChatGPT Desktop (OpenAI, closed source, mainly for prompt entry), Cline (open source, but a VS Code extension), Continue (VS Code plugin), Cody (Sourcegraph, code-focused), Codename Goose (Block, open source). Main differentiators: open-source philosophy, closed learning loop as a first-class citizen, no lock-in to a single communication channel, serverless support as a native deployment option. Documentation available at hermes-agent.nousresearch.com/docs with dedicated llms.txt (~17 KB) and llms-full.txt (~1.8 MB) for LLM ingestion.
