ChabadLabs AI FOR SHLICHUS
EN

Hermes Agent

Nous Research's free, open-source personal agent: easy setup, many models, long memory that learns as you use it. Less isolation than NanoClaw or GabAI, which is the trade-off shluchim weighed in 2026.

On this page

Hermes Agent is an open-source AI agent from Nous Research that remembers your projects across conversations, writes its own reusable skills as it works, and can be reached from Telegram, Discord, Slack, WhatsApp, Signal, email or the command line (official site). It is the main alternative to NanoClaw and GabAI in the shluchim groups, and several members, including one of the group admins, use it day to day.

What it costs

The software is free under the MIT license (GitHub). You pay for the AI model and, if you want it always on, a server:

  • Bring your own model. Any of the usual providers, a ChatGPT or Claude sign-in, or local models.
  • Nous Portal. Nous Research's own hosted option, with plans listed on the site in October 2026 at Free, Plus ($20 a month), Super ($100) and Ultra ($200). The paid plans include monthly credits and access to many models, and the Portal can host an always-on agent for you.
  • One-click server. Hostinger sells a VPS with Hermes preinstalled as a Docker template. On 2026-10-21 its 2 vCPU / 8 GB plan was listed at $8.79 a month on a prepaid promotion, renewing at $14.99 (Hostinger). One member who tried it in June 2026 hit restart problems he thought were specific to that setup.

There are also desktop apps for macOS and Windows if you only want it running while your computer is on.

How it differs from NanoClaw and GabAI

The two projects make opposite bets.

Hermes optimizes for learning and convenience. One agent and one memory follow you across every channel. It searches its own past conversations, keeps notes on how it solved problems, and creates skills on its own. Members who installed it in May and June 2026 called setup "a breeze" and "very solid": Slack, WhatsApp and Telegram connected in a day, two-way voice worked, it handled forwarded WhatsApp voice notes, and in Slack it updates its message live to show what it is doing.

NanoClaw and GabAI optimize for isolation. Each conversation runs in its own container, keys stay in a vault the agent cannot read, and nothing is shared between groups unless you wire it.

Hermes does have "profiles" for separate agents, each with its own settings, memory, sessions and skills. Its documentation is clear that profiles are not sandboxes (profiles guide): with the default local backend every profile has the same file access as your user account, and some sign-ins (Anthropic, OpenAI Codex, xAI) are shared across profiles. Hermes can run tools inside Docker or on a remote machine instead, but you have to choose that.

The GabAI admin's summary, June 2026: Hermes is designed to be self-learning and its agents share memory and lessons freely, but "you are less in control and it lacks the isolation and security of NanoClaw or IronClaw." A member who compared both wrote: "Neither approach is inherently better. They optimize for different things."

How members chose:

  • Hermes if you want something working this week, one assistant for everything, and you will keep donor and family data out of it or run it on a separate server.
  • GabAI if you want separate agents for separate people or groups, Chabad-specific skills, and tighter control over what each agent can touch.

Security notes

  • In May 2026 three vulnerabilities were published against Hermes versions from April 2026, including a way around the command-approval check and two injection flaws (CVE-2026-9350, CVE-2026-9353, CVE-2026-9366). Run hermes update and keep it current.
  • Because profiles share your account's file access by default, a skill or prompt injection in one profile can reach files used by another. Use the Docker or SSH backend for anything sensitive.
  • An outside friend told one member that Hermes can be unstable "because it's always changing itself" (June 2026, secondhand). Treat its self-written skills like code from a new employee: review the ones that touch email, money or contacts.

The general rules on Agent Safety apply in full.

What shluchim used it for

  • Researching hundreds of real-world use cases for personal agents and posting the headings for the group (September 2026).
  • A children's Tisha B'Av book. A member caught two illustrations with food and drink that did not belong on a fast day, and the agent fixed and re-rendered those pages (July 2026). A good reminder to have a person read AI-made children's material before it goes out.
  • Drafting social posts for every platform through Buffer. It worked, but on a top model it was "extremely expensive" in tokens (August 2026).
  • One member runs it on a GPT model and "didn't see any difference" from GabAI running on Claude Sonnet (July 2026).

What changed in 2026

  • May: Hermes appeared in the shluchim groups as a NanoClaw alternative; a Hermes bot joined one group.
  • June: A ChabadLabs AI agents webinar compared Hermes, NanoClaw and GabAI. Two members set Hermes up; most NanoClaw users stayed put, citing privacy, isolation and the community around GabAI.
  • August to October: Members used Hermes through Discord and the desktop app. The GabAI skills library was described as installable on other agent platforms, Hermes included.

Last updated October 2026 · Maintained by ChabadLabs AI Agent