ClaudeMap

hermes-agent

Nous Research's self-improving agent: a closed learning loop that creates skills from experience, improves them during use, searches its own past sessions — run on a $5 VPS, a GPU cluster, or serverless that costs nearly nothing when idle

Plugins & Agentsagent harnessself-improvingmemoryskillsmulti-platformnous research

Hermes Agent is Nous Research's open-source agent built around what the README calls a closed learning loop: agent-curated memory with periodic nudges to persist knowledge, autonomous skill creation after complex tasks, skills that self-improve during use, FTS5 session search with LLM summarization for cross-session recall, Honcho dialectic user modeling, and compatibility with the agentskills.io open standard. It lives where you do — Telegram, Discord, Slack, WhatsApp, Signal and a full terminal TUI (multiline editing, slash-command autocomplete, interrupt-and-redirect, streaming tool output) from a single gateway process, with voice memo transcription and cross-platform conversation continuity. Models are swappable with hermes model — Nous Portal, OpenRouter, OpenAI, your own endpoint and more, no code changes. Scheduled automations ship built in (cron with delivery to any platform, described in natural language). It delegates: isolated subagents for parallel workstreams (provider, model and limits configurable in Desktop settings) and Python scripts that call tools via RPC to collapse multi-step pipelines into zero-context-cost turns. It runs anywhere via six terminal backends — local, Docker, SSH, Modal, Daytona and Vercel Sandbox — plus plugin backends such as Singularity/Apptainer; Daytona and Modal offer serverless persistence where the environment hibernates when idle. Research-ready: batch trajectory generation and trajectory compression for training tool-calling models.

What it does

  • Closed learning loop: agent-curated memory with periodic nudges, autonomous skill creation after complex tasks, skills that self-improve during use — compatible with the agentskills.io open standard
  • Lives where you do: Telegram, Discord, Slack, WhatsApp, Signal and a full TUI from a single gateway process, with voice memo transcription and cross-platform conversation continuity
  • Model-agnostic via hermes model: Nous Portal, OpenRouter, OpenAI, your own endpoint and more — no code changes, no lock-in
  • Six terminal backends (local, Docker, SSH, Modal, Daytona, Vercel Sandbox) plus plugin backends (Singularity/Apptainer); Daytona and Modal hibernate the environment when idle for near-zero idle cost
  • Delegation and research: isolated subagents for parallel workstreams, RPC-based Python tool scripting, scheduled automations delivered to any platform, batch trajectory generation

When to use it

Use when a personal or team agent should persist and compound: an assistant that remembers your projects across weeks (FTS5 search plus curated memory), a Telegram-first agent that works on a cloud VM while you message it, or a scheduled-automation runner that files daily reports from natural-language descriptions. The research features (trajectory generation/compression) also make it a substrate for tool-calling-model experiments.

Getting started

Linux, macOS and WSL2:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

After install, switch models with hermes model (Nous Portal, OpenRouter, OpenAI or your own endpoint). Docs live at hermes-agent.nousresearch.com (the /docs path has shown redirect flapping between /docs and /docs/ — use the deep links in the docs sidebar). The repo also ships README.zh-CN.md and other locale READMEs.

Things to know

The README claims it is "the only agent with a built-in learning loop" — that is the vendor's own superlative, not a measured fact. The learning-loop behaviors (skill self-improvement, memory nudges, Honcho user modeling) are the README's description; skill quality compounds or decays with use, so curate the skills folder. Serverless persistence depends on third-party platforms (Modal, Daytona) and their pricing; the "$5 VPS" figure is the README's example floor, not a quote. Multi-platform delivery runs through one gateway process — that process is the single point of failure for all channels. MIT. 264,479 stars as of writing (2026-11-19).

Repository

NousResearch/hermes-agent

Charted

Related on the map

  • clawd-conduit

    Plugin / Agent

    Wire all 15 Claude Code hook events into the Clawd on Desk desktop pet — and turn permission requests into a blocking desktop card.

    • hooks
    • desktop
    • notifications
    • claude-code
  • CometixCode

    Plugin / Agent

    An unofficial Rust 1:1 reimplementation of the Claude Code terminal UI, ported file-for-file onto an iocraft TUI framework.

    • terminal-ui
    • rust
    • reimplementation
    • unofficial
  • Abide

    Plugin / Agent

    Hook-level enforcement for the natural-language rules in AGENTS.md and CLAUDE.md — every edit is judged and violations are repaired in the same turn.

    • hooks
    • code-review
    • rules
    • agents-md