HERMES
self-improving agent — nous research

The Self- Improving Agent.

gets
smarter
every run

Not a copilot. Not a wrapper. An autonomous agent with a closed learning loop — it compounds.

$ curl -fsSL hermes-agent.nousresearch.com/install.sh | bash
open source — github · docs · desktop for mac/win/linux
01 / the loop

It learns like you do.

The only agent that ships with a closed learning loop. Experience becomes memory, memory becomes skill, skill becomes instinct — every single session.

  • Curates its own memory, nudges itself on a schedule
  • Writes new skills autonomously after solving something hard
  • Patches skills mid-use the moment they misfire
  • Full-text recall across every past session
  • Dialectic user modeling via Honcho
02 / infrastructure

Lives on your terms. Not your laptop's.

Six terminal backends. A five-dollar VPS, a GPU cluster, or serverless cloud that hibernates when idle and costs nearly nothing.

  • Local · Docker · SSH · Daytona · Singularity · Modal
  • Serverless persistence — wakes on demand, sleeps for free
  • Chat on Telegram while it works on a VM you never SSH into
03 / presence

One gateway. Every app you already use.

21+ messaging platforms, one agent. It follows you — not the other way around.

  • Telegram · Discord · Slack · WhatsApp · Signal · Matrix
  • Email · SMS · Teams · Google Chat · Mattermost · IRC
  • WeChat ecosystem — DingTalk · Feishu · WeCom · Weixin · QQ · Yuanbao
  • iMessage via BlueBubbles · Home Assistant + plugin extras
04 / toolbelt

Sixty-plus tools. Zero ceiling.

Search, browse, see, speak, build, ship — out of the box. Then plug in any MCP server and the belt keeps growing.

  • Web search & extraction · browser automation with vision
  • Image generation · text-to-speech · speech-to-text
  • Cron · sandboxed code execution · computer use
  • Programmatic Tool Calling — whole pipelines in one inference call
05 / procedural memory

Skills that write themselves.

Hard-won experience gets distilled into reusable playbooks — then improved every time they're used. A background Curator prunes what goes stale.

  • Autonomous creation after novel solves
  • Self-patching when steps fail in the wild
  • Curator: usage tracking, staleness review, archival
  • Open standard at agentskills.io — portable, shareable, community hub
06 / identity

Memory that compounds.

Not a context window. A model of you that deepens over months — and shapes every future conversation.

  • MEMORY.md + USER.md, curated and always injected
  • Full-text search over every session ever
  • Provider plugins: Honcho · Mem0 · Supermemory + more
  • Context files — SOUL.md, AGENTS.md — shape behavior globally
07 / multi-agent

Hire a team, not a chatbot.

Bot Mode spins up named specialists — each with its own model, memory, skills, avatar. They coordinate without you.

  • Per-bot chat, role, model, memory & skill stack
  • Routines, shared group chats, @mention handoffs
  • Parallel subagents for isolated workstreams
08 / senses

It talks. It sees. It ships files.

Answers are the floor. Deliverables are the point — finished artifacts land straight in your chat.

  • Real-time voice mode in CLI, Telegram, Discord & VC
  • Vision: screenshots, photos, documents
  • Deliverable Mode: charts, PDFs, spreadsheets as native attachments
  • Built-in image generation + video-gen providers
09 / automation

Set it once. It runs forever.

Built-in cron delivers anywhere you live. No babysitting required.

  • Cron to any platform · recurring loops · session heartbeats
  • Webhook subscriptions for event-driven runs
  • Schedule from dashboard, CLI, TUI or messenger
10 / control

Autonomous, never unaccountable.

You hold the keys — always.

  • Approval gates before dangerous commands
  • Checkpoints & rollback for destructive ops
  • Container isolation · scoped secrets · egress proxy
11 / begin

Stop reading.
Start running.

60+
built-in tools
21+
platforms
6
backends
60s
install
$ curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash $ hermes setup --portal
scroll