Hermes Agent: A Self-Improving AI Agent
/hermes-agent
Hermes Agent is a self-improving AI agent from Nous Research that builds and refines its own skills while you use it. It runs on a terminal, a $5 VPS, or serverless infrastructure, and reaches you over Telegram or Discord instead of staying tied to one laptop. Reach for it if you want an agent that keeps learning and remembers your workflow across sessions; skip it if you just need a stateless assistant with no setup.
Understanding Hermes Agent
Hermes Agent is a self-improving AI agent built by Nous Research that runs as a terminal UI or a messaging gateway reachable from Telegram, Discord, Slack, WhatsApp, and Signal. It runs a closed learning loop: creating skills from what it does, improving them with use, and searching its own past conversations, and you can point it at Nous Portal, OpenRouter, OpenAI, or your own model endpoint.
Core Capabilities and Design
- ✓Skill learning that creates skills from complex tasks and refines them automatically the more you use them, per the README.
- ✓Memory management via agent-curated notes, periodic self-reminders to persist what it learns, and FTS5 session search with LLM summarization for pulling up old conversations.
- ✓A full terminal UI, which incorporates multiline editing, slash-command autocomplete, conversation history, and interrupt-and-redirect, also enables the use of 40+ built-in tools that are categorized into toolsets, as shown in the documentation table in the README.
- ✓A single messaging gateway process reaching Telegram, Discord, Slack, WhatsApp, and Signal, with voice memo transcription and conversation continuity across platforms.
- ✓A built-in cron scheduler that runs jobs like nightly backups or weekly audits in plain language, unattended, per the README.
- ✓Subagent delegation: spawn isolated subagents for parallel workstreams, or write Python scripts that call tools via RPC to collapse multi-step pipelines into one turn.
- ✓Seven terminal backends (local, Docker, SSH, Singularity, Modal, Daytona, and Vercel Sandbox), with Daytona and Modal hibernating the environment when idle.
- ✓Provider flexibility through Nous Portal (which the README says gives access to 300+ models plus a bundled Tool Gateway for web search, image generation, TTS, and a cloud browser), OpenRouter, OpenAI, or your own endpoint, switched with a single `hermes model` command.
Installing Hermes Agent
On Linux, macOS, WSL2, or Termux, install with curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash. On native Windows, run iex (irm https://hermes-agent.nousresearch.com/install.ps1) in PowerShell. No WSL required, though the Linux one-liner still works inside WSL2 if you'd rather use that. The installer sets up uv, Python 3.11, Node.js, ripgrep, ffmpeg, and on Windows a portable Git Bash (MinGit, about 45MB) so it doesn't touch any system Git install; it reuses your existing Git if you already have one. After install, run `source ~/.bashrc` (or `~/.zshrc`) and then `hermes` to start chatting. Termux users get a curated `.[termux]` extra because the full `.[all]` extra pulls in Android-incompatible voice dependencies, per the README. If your antivirus flags `uv.exe` as malware, the README calls that a false positive: it's Astral's uv package manager, and the README documents a GitHub attestation check to verify the binary.
Automations and Parallel Workflows
- •Setting up agentic AI automation with the built-in cron scheduler, running nightly backups or weekly audits in plain language, unattended.
- •Spawning subagents for parallel workstreams so a multi-step pipeline runs as one turn instead of a long back-and-forth.
- •Working from a phone over Telegram while a multi-platform AI setup keeps executing on a cloud VM, GPU cluster, or serverless backend that hibernates between sessions.
- •Migrating an existing OpenClaw setup with `hermes claw migrate`, importing skills, memory files, and allowlisted API keys in one pass.
- •Generating and compressing batch trajectories for training your own tool-calling models, per the README's research-ready note.
Strengths
- ✓Skills accumulate instead of resetting each session, so the agent gets more useful for your specific workflow the longer you run it.
- ✓Not tied to your laptop: run it on a $5 VPS or serverless backend and reach it from Telegram or Discord while it keeps working.
- ✓Nous Portal is optional: you can still bring your own per-tool API keys instead of using the bundle, per the README.
- ✓A documented migration path from OpenClaw, with a `--dry-run` flag to preview what `hermes claw migrate` would change before committing.
- ✓MIT license, so there's no proprietary lock-in on the code itself.
Known Installation Considerations
- △Native Windows setup either detects an existing Git install or downloads a roughly 45MB portable MinGit, and the README documents that antivirus tools like Windows Defender or Bitdefender have flagged the bundled `uv.exe`, even though it calls that a false positive.
- △Termux on Android doesn't get the full `.[all]` extra; per the README it installs a separate curated `.[termux]` extra because the full set pulls in voice dependencies that don't work on Android.
- △The README warns that a Python virtual environment (venv) located within the directory where the agent operates could be wiped by a relative-path command executed by the agent itself. Therefore, for a manual clone install (bypassing the managed installer), its Python virtual environment must be built outside the source tree.
- △The README doesn't list a version number, release date, or changelog, so there's no documented way to check what release you're running beyond the `hermes update` command.
- △Pricing for the LLM providers themselves, Nous Portal, OpenRouter, OpenAI, and others, isn't covered in the README beyond naming them as options, so budgeting for usage is left to you.
Migrating from OpenClaw
Common Questions
Hermes Agent is a self-improving AI agent built by Nous Research. It runs as a terminal UI or a messaging gateway, creating and refining its own skills as you use it instead of shipping with a fixed toolset.
Installing Hermes Agent takes one command: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash on Linux, macOS, WSL2, or Termux, or the PowerShell one-liner on native Windows. The installer sets up Python, uv, and the other dependencies for you.
Hermes Agent works with Nous Portal, OpenRouter, OpenAI, your own endpoint, and other large language model providers listed in its docs. Switch between them anytime with the `hermes model` command, without changing any code.
Hermes Agent runs on Telegram, Discord, Slack, WhatsApp, Signal, and Email through a single gateway process, alongside its CLI and terminal UI. You keep one conversation history across whichever platform you use.
Hermes Agent is released under the MIT license, as shown in its GitHub repository and LICENSE file.
Hermes Agent's learning loop creates skills from what it does, improves them the more you use it, nudges itself to save what it learns, searches past conversations for context, and builds a model of you across sessions using Honcho dialectic modeling.
The problem it solves
Most AI agents start every conversation from zero: no memory of what worked before, no way to get better at a recurring task without you re-explaining it, and no path to running unattended once you close the laptop. That's the gap agentic AI tooling is supposed to close, and Hermes Agent's README frames its own reason for existing around a loop that curates its own memory, builds skills from what it just did, and keeps running from a $5 VPS or a serverless backend that costs next to nothing while idle.
Who should try it — and who should skip
Try Hermes Agent if you're comfortable living in a terminal but want an assistant that keeps improving and follows you across devices, from a phone via Telegram to a cloud VM you rarely touch directly. It's also a reasonable pick if you're already running OpenClaw and want an upgrade path that keeps your skills and memory intact. Skip it if you want a simple, stateless chatbot with no setup step, or if you're not willing to trust an agent with shell access and persistent memory of your conversations.
Related repositories
Still deciding about hermes-agent?
One click hands the question to an AI along with this page — see what it says about hermes-agent.