Nous Research Releases Hermes‑Agent, a Self‑Improving AI Assistant for Cloud and Desktop Use

Overview

Nous Research has opened source Hermes‑Agent, an AI assistant that claims to improve itself over time. The project is hosted on GitHub and is positioned as a flexible, model‑agnostic tool that can run on inexpensive cloud VMs, a GPU cluster, or even a server‑less environment.

Key Features (as described by the repository)

• Built‑in learning loop – the agent creates new “skills” from experience, refines them during use, and stores knowledge across sessions.
• Model‑agnostic – works with Nous Portal, OpenRouter, OpenAI, custom endpoints, and many others; switching models requires no code change.
• Multi‑platform support – native Windows CLI, PowerShell installer, WSL2, Linux/macOS one‑liners, and an Android/Termux APT repository for aarch64 devices.
• Messaging gateways – can be accessed via Telegram, Discord, Slack, WhatsApp, Signal, or email, alongside a terminal UI (TUI).
• Import tools – can migrate settings, memories, and skills from agents such as OpenClaw, Claude Code, Codex CLI, or Cursor.

Installation Highlights

• Windows: a PowerShell one‑liner installs Python 3.14, Node.js, npm, pnpm, ripgrep, FFmpeg, and other dependencies under `%LOCALAPPDATA%\hermes`. The same script works under WSL2.
• Linux/macOS: a single command line pulls the same stack into `~/.hermes`.
• Android/Termux: a signed APT repo provides a pre‑built package that includes the required runtimes and the TUI.
• Antivirus false positives – the bundled Rust‑based package manager `uv.exe` may be flagged; the README advises adding an exclusion path rather than a file hash.

Tool Gateway & API Keys

Hermes can route web search, image generation (FAL), text‑to‑speech (OpenAI), and a cloud browser through a Tool Gateway. Users may rely on a single subscription from Nous Portal to cover all required API keys, or they can supply their own per‑tool keys.

Practical Implications

• Cost‑effective deployment – the agent can run on a $5 VPS and stay idle at near‑zero cost, making it attractive for hobbyists and small teams.
• Cross‑device continuity – because knowledge is persisted across sessions, a user can start a conversation on a laptop and continue it later from a phone via Telegram.
• Open‑source transparency – all code, documentation, and migration scripts are publicly available, allowing independent verification of the self‑learning claims.

How to Verify the Announcement

1. Visit the official repository: https://github.com/nousresearch/hermes-agent**
2. Review the `README.md` and `docs/` folder for the latest feature list and installation commands.
3. Check the release tags for version numbers and the changelog for any updates since the initial launch.
4. Test the agent in a sandbox environment (e.g., a fresh VPS) following the documented steps.

Limitations & Unknowns

• The repository states the learning loop is self‑improving, but independent benchmarks of skill quality over time are not provided.
• Performance on large language models or GPU‑heavy workloads has not been publicly measured.
• No official roadmap or timeline for future releases is included in the source.

Keywords

• Hermes‑Agent
• self‑improving AI assistant
• model‑agnostic AI
• cross‑platform deployment
• Nous Research
• AI tool gateway
• open‑source AI agent

--- *All information above is taken directly from the GitHub repository and its accompanying documentation. No external claims or unpublished specifications have been added.*