Personal AI assistant agent from Nous Research. Install site: hermes-agent.nousresearch.com — pitched as a one-command install, then guided step-by-step setup. Positioned as a self-hosted personal assistant that can write and send emails, run research, and pair with a coding tool.
Spotted via a Russian-language AI community channel (Клуб ИИнутых, ~629 subs). The author’s setup notes:
- Ran it on Codex with a GPT-5.5 subscription, framing the cost choice as binary: “either pay Anthropic for tokens, or use Codex.” (i.e. the agent needs a capable backing model and that’s the spend.)
- Bootstrapped the agent’s context by dumping a single “build a profile of me” prompt into ChatGPT, Claude, and Manus, then feeding the result to the agent as its base knowledge.
Why This Matters
Fits the agent-stack / personal-assistant-orchestration interest. Hermes Agent is another entrant in the “always-on personal agent that acts on email + research + code” category — adjacent to how I already run Claude Code + memory. Worth a look at how Nous structures the install, the tool permissions (email send is the interesting/risky one), and whether the agent loop is meaningfully different from a Claude Code + skills setup.
The model-cost framing is the real strategic note: these agents are only as good as the backing model, and the spend is the token bill either way.
The Bootstrap Prompt (reusable)
The author’s onboarding trick is genuinely reusable for any new assistant (including refining my own CLAUDE.md / memory). Translated from the original Russian:
Build a brief profile about me for a new personal AI assistant, based on our past conversations.
Include:
- Who I am, what I do, my roles and projects.
- My long-term goals.
- My current active projects and interests.
- What communication style I prefer.
- How I like to receive answers: brief/detailed, lists/tables, with/without recommendations.
- Which languages to use.
- My typical tasks for the assistant.
- Which decisions I usually make myself, and what can be delegated.
- My limits, taboos, sensitive topics.
- Any stable preferences: tools, services, workflows.
- What’s important for the new assistant NOT to forget about me.
Don’t include passwords, API keys, private tokens, banking data, or other secrets. Do it in a structured form.
Run this against an existing ChatGPT/Claude history to generate a portable “profile of me” any new agent can ingest — a fast way to seed a fresh assistant with accumulated context.
Questions
- Open source / self-hosted, or hosted service with a key? What’s the actual trust/permission model for email send?
- How does the agent loop compare to Claude Code + skills — does it add orchestration, or is it a thinner wrapper?
- What backing models are supported beyond GPT-5.5 / Codex? Can it run on Anthropic models directly?
Source
- Platform: Telegram — Клуб ИИнутых (~629 subscribers), post “Поставил Hermes Агента”
- Product: hermes-agent.nousresearch.com (Nous Research)
- Date spotted: 2026-06-07
part of tooling