AI Coding Tool Setup and Reference (6/7) — Hermes Agent Install: One-Liner + SOUL.md

Nous Research official install.sh, auto-provisioned Python 3.11, ~/.hermes/ layout, persona file in practice

ํ•ต์‹ฌ ์š”์•ฝ

  • Audience: First-time installer of Hermes Agent (NousResearch/hermes-agent, Feb 2026 release).
  • What you'll get: Official install.sh one-liner, auto-provisioned dependencies (uv / Python 3.11 / Node 22 / ripgrep / ffmpeg), platform limits (native Windows unsupported), first hermes run, 9 core subcommands, and SOUL.md schema, location, and example — verbatim.
  • Prerequisite: Git pre-installed. Linux / macOS / WSL2 / Android (Termux).
  • License: MIT. Built by Nous Research.

1. What Hermes Agent actually is

Per the official docs:

  • Released February 2026 — open-source AI agent. Tagline: "The agent that grows with you."
  • Self-improving skills loop, persistent memory (SQLite FTS5), 40+ built-in skills.
  • 5 messenger channels: Telegram, Discord, Slack, WhatsApp, Signal (+Email +CLI).
  • LLM-agnostic: Nous Portal, OpenRouter (200+), NVIDIA NIM, Xiaomi MiMo, z.ai/GLM, Kimi, MiniMax, Hugging Face, OpenAI, Anthropic, custom endpoints, local vLLM.
  • Official OpenClaw migration path (hermes claw migrate).

Claude Code and Codex are IDE/terminal-centric. OpenClaw is a channel orchestrator. Hermes bundles long-term memory + self-learning + channels into one product.

2. Install — the official one-liner

2.1 Primary command

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

Then:

source ~/.bashrc    # or ~/.zshrc on macOS
hermes              # first run

No sudo needed. Installs entirely under ~/.hermes/, not system-wide.

2.2 Auto-provisioned dependencies

install.sh handles these automatically:

  • uv (Rust-based Python package manager)
  • Python 3.11
  • Node.js v22 (browser automation + WhatsApp)
  • ripgrep (file search)
  • ffmpeg (audio conversion)

Required pre-install: just Git.

2.3 Nix users

Official docs mention a Nix flake declarative setup. If you already use Nix, this path offers reproducibility.

2.4 Developer build (from source)

git clone https://github.com/NousResearch/hermes-agent.git
cd hermes-agent
./setup-hermes.sh

3. Platform support — actual boundaries

Per the docs:

Platform Support
Linux ✅
macOS (Apple Silicon / Intel) ✅
WSL2 ✅
Android (Termux) ✅
Windows native ❌ Marked "experimental", WSL2 recommended

On Windows you need WSL2 (Ubuntu/Debian). The installer follows the Linux path inside WSL.

4. Directory layout after install

~/.hermes/
├── SOUL.md              # persona (auto-created)
├── MEMORY.md            # long-term memory (optional)
├── USER.md              # user profile (optional)
├── skills/              # self-generated + OpenClaw-migrated
│   └── openclaw-imports/
├── (SQLite DB, session store)
└── (logs)
  • SOUL.md is auto-created on first hermes — if empty, falls back to a built-in default.
  • $HERMES_HOME env var can override the base location. $HERMES_HOME/SOUL.md wins.

5. First session

hermes

First run launches the interactive setup — internally it calls hermes setup.

5.1 Setup wizard

  1. LLM provider — Nous Portal / OpenRouter / OpenAI / Anthropic / custom
  2. API key (per provider)
  3. Default model
  4. Default tone / persona
  5. Add messenger channels (skippable; later via hermes gateway setup)

You can start chatting immediately after.

5.2 Core hermes subcommands

Command Effect
hermes Interactive CLI
hermes setup Full setup wizard
hermes model Switch LLM provider / model
hermes tools Toggle built-in tools
hermes config set Edit individual settings
hermes gateway setup Configure messenger creds
hermes gateway Start the gateway process
hermes gateway install Register systemd service (Linux)
hermes claw migrate Migrate OpenClaw settings / skills
hermes update Update to latest
hermes doctor Diagnose environment

5.3 Shared slash commands (CLI + messaging)

Command Effect
/new Start a new conversation
/reset Reset context
/model Switch model
/personality [name] Session-only persona overlay
/retry Regenerate last response
/undo Rewind
/compress Summarize / compress conversation
/usage Tokens / cost
/insights Session analytics
/skills Loaded skills list
/stop Stop current task

6. SOUL.md — the persona file

Hermes's distinctive feature: SOUL.md is slot #1 in the system prompt. Documented explicitly.

6.1 Location and loading rules

  • Default: ~/.hermes/SOUL.md
  • Custom: $HERMES_HOME/SOUL.md
  • Caveat: SOUL.md in the current working directory is not loaded. Only HERMES_HOME is consulted.
  • Auto-created if missing. Never overwrites an existing file.
  • If empty or unreadable → falls back to a built-in default.
  • Injected into the system prompt once (slot #1). No duplication.
  • Injected verbatim after security scan + length truncation.

6.2 Put in SOUL.md

Per the official recommendation: durable voice elements.

  • Tone and communication style
  • How direct to be
  • How to handle uncertainty
  • What to avoid stylistically

6.3 Don't put in SOUL.md

Project-specific details belong in AGENTS.md instead.

6.4 Official example structure

You are a pragmatic senior engineer with strong taste...

## Style
- Be direct without being cold
- Prefer substance over filler
- Push back when something is a bad idea

6.5 Extended example

You are a pragmatic senior engineer with a clear sense of quality.

## Style
- Direct, not cold
- Substance over filler
- Push back on bad decisions with reasons

## Uncertainty
- Tag things you don't know as "needs verification"
- When speculating, say so explicitly

## Avoid
- Over-apologizing, performative hedging
- Repeating things the reader already knows

7. Three related systems — SOUL.md / AGENTS.md / /personality

Official comparison:

System Scope Use
SOUL.md Persistent, across all contexts Fundamental persona
AGENTS.md Per-project Paths, conventions, instructions (same concept as Codex / OpenClaw)
/personality [name] Session only Temporary overlay

Built-in /personality presets: helpful, concise, technical, creative, teacher, kawaii, catgirl, pirate, shakespeare, surfer, noir, uwu, philosopher, hype.

Example:

/personality technical

Only this session's tone shifts. On next session, it reverts to SOUL.md.

8. Supported model providers — April 2026

Per the official README/homepage:

  • Nous Portal (portal.nousresearch.com)
  • OpenRouter (200+ models)
  • NVIDIA NIM (Nemotron family)
  • Xiaomi MiMo
  • z.ai / GLM
  • Kimi (Moonshot)
  • MiniMax
  • Hugging Face
  • OpenAI
  • Anthropic
  • Custom endpoints (OpenAI-compatible)
  • Local vLLM

Switch interactively:

hermes model

Or edit config directly.

9. Memory system — SQLite FTS5

Officially documented:

  • Storage: SQLite DB under ~/.hermes/ with an FTS5 full-text index.
  • Session search + LLM summarization for cross-session recall.
  • MEMORY.md / USER.md: traditional markdown memory, especially populated by OpenClaw migrations.
  • Honcho dialectic user modeling (observation → hypothesis → verified).

Details covered in a dedicated post — [#14 Persistent memory FTS5 in practice] (coming soon).

10. Gateway — channel connection

hermes gateway setup     # configure credentials
hermes gateway           # start process
hermes gateway install   # register systemd service (Linux, 24/7)

Supported channels: Telegram, Discord, Slack, WhatsApp, Signal, Email, CLI. Supports voice-memo transcription and cross-platform conversation continuity.

Detailed multi-channel ops: separate post.

11. Diagnostics — hermes doctor

When things break:

hermes doctor

Checks (official): - Python 3.11 / uv installed - Node 22 / ripgrep / ffmpeg availability - $HERMES_HOME accessibility - Configured provider API key validity - Whether a post-update migration is needed

Fix pattern: resolve items → hermes update → hermes doctor again.

12. Updating

hermes update

Official advice: always run hermes doctor after updating to catch migration requirements.

  • Breaking changes possible: this is an early-stage project (first released Feb 2026) still iterating quickly. Team environments should stage updates first.

13. Counter-scenarios — when Hermes isn't the right fit

  • Native Windows is a hard requirement → WSL2 is mandatory for Hermes. Consider OpenClaw (Windows-supported) or Claude Code instead.
  • You don't want Python around → Hermes is Python 3.11-based. Codex (Rust) / Claude Code (native binary) are lighter.
  • IDE-only coding workflow → Claude Code / Codex fit better. Hermes's value is multi-channel + long-term memory.
  • Strict enterprise audit needs → Early-stage project, no formal SOC2/etc. guarantees yet. Safer for individuals and startups.
  • Offline environment → local vLLM is an option for the model, but install still needs internet.

14. 30-minute checklist

  • [ ] Git present (git --version)
  • [ ] Run curl -fsSL ... install.sh | bash
  • [ ] No errors; confirm uv / Python 3.11 / Node 22 / ripgrep / ffmpeg auto-installed
  • [ ] Reload shell (source ~/.bashrc or ~/.zshrc)
  • [ ] Run hermes; complete the wizard (provider + API key + default model)
  • [ ] Send a "hello" test message, confirm reply
  • [ ] Open ~/.hermes/SOUL.md, verify auto-created content
  • [ ] Add 3–5 personal style rules to SOUL.md (see 6.5)
  • [ ] Try /personality technical to verify session overlay works
  • [ ] Run hermes doctor for final environment check
  • [ ] (Optional) hermes gateway setup to connect Discord / Telegram / etc.

15. What's next

With Hermes basics in place:

  1. hermes command complete cheatsheet — subcommands + slash commands (coming soon)
  2. Persistent memory FTS5 in practice — queries, Korean tokenization (coming soon)
  3. OpenClaw → Hermes migration checklist (coming soon)

References

This is post 12/15 in the "AI Coding CLI Entry Guide" series. Hermes Agent part begins here.
last verified: 2026-04-25 (per NousResearch/hermes-agent GitHub + hermes-agent.nousresearch.com/docs).

Series overview: Series index

๋Œ“๊ธ€

๋Œ“๊ธ€ ์“ฐ๊ธฐ

์ด ๋ธ”๋กœ๊ทธ์˜ ์ธ๊ธฐ ๊ฒŒ์‹œ๋ฌผ

"ML Foundations (9/9) — PyTorch vs TensorFlow, and the Road to Local LLMs"

์ด๋ฏธ์ง€
← 8/9 Deep Learning Architec… ๐Ÿ“š Series Index (series end) The final part. So far we've focused on models. Now we focus on the tools that actually run them . We'll lay out the philosophical difference between PyTorch and TensorFlow, and trace how Hugging Face transformers , llama.cpp , MLX , and Ollama built the bridge to running large language models on your own machine. By the end you should have the full mental model of "download a pretrained LLM and serve it locally." 0. Learning Objectives Explain the eager-vs-graph difference between PyTorch and TensorFlow. Explain, in graph terms, how autograd automates the backward pass. Use Hugging Face transformers ' AutoModel , AutoTokenizer , and pipeline abstractions. Describe llama.cpp's GGUF quantization, INT4 inference, and CPU-first flow. Describe Apple MLX's use of unified memory and how it differs from PyTorch. Run a local LLM with Ollama and call it through an OpenAI-compatible API. ...
Read more »

Agent Memory Engine (2/10) — Building an AI Agent Memory System with SQLite Alone

← 1/10 SQLite Memory Engine f… ๐Ÿ“š Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere ํ•ต์‹ฌ ์š”์•ฝ This post covers how to design an AI agent memory engine (memcore) on a single-file SQLite backend. - Structure: 25 modules, ~6,464 lines, 22 DB tables - Principles: no daemon, single-file portability, no required external dependencies, host-agnostic - Core techniques: 4-layer hybrid search (topic · vector · FTS5 · LIKE), U-tag dialectic-based confidence evolution, ingestion-layer bias rejection rules What This Post Covers One approach to building AI agent memory without a server or cloud vector DB. Specifically: (1) the limitations of text-file-based memory, (2) vector DB alternatives compared and selection criteria, (3) how to layer search tiers, (4) how to evolve confidence scores over time, and (5) how to block biased memory accumulation at the ingestion stage. Design Principles: Why Single-File SQLite Text-file-based ...
Read more »

"ML Foundations (8/9) — Deep Learning Architectures: CNN, RNN, Attention"

์ด๋ฏธ์ง€
← 7/9 Deep Learning Training ๐Ÿ“š Series Index 9/9 PyTorch vs TensorFlow → What did we keep adding on top of an MLP? Vision went CNN, sequences went RNN/LSTM, and eventually both converged on Attention and the Transformer. Each architecture is easier to remember if you read it as what it gave up to gain something else . This part is the one-line summary of the decisive inflection points: LeNet → AlexNet → ResNet → LSTM → Attention → Transformer. 0. Learning Objectives Explain why CNN's convolution, pooling, and weight sharing match images so well. Trace what got unblocked from LeNet → AlexNet → VGG → ResNet as depth grew. State BPTT for RNNs and the vanishing/exploding gradient problem. Write the LSTM gate equations (forget, input, output) with intuition. Write the attention formula in query/key/value form and the scaled dot-product variant. State the precise reasons Transformers replaced RNNs (parallelism + long-range dependencies). 1. ํ•ต์‹ฌ ์š”์•ฝ CNN : weight sharin...
Read more »

"RAG Core Study (14/26) — Evaluation Sets with RAGAS & DeepEval"

์ด๋ฏธ์ง€
Series overview: Series index ๐Ÿ“š Series Index Without an evaluation set, every RAG improvement is a story, not evidence. RAG systems fail in more than one place: retrieval, context selection, answer generation, and grounding. That means "looks better" is not a metric. Part 14 explains how to build a golden dataset , how RAGAS and DeepEval fit into the workflow, and why teams must separate retrieval quality from answer quality if they want tuning decisions to hold up. 0. Prerequisites Part 13 reranking — system quality now depends on multi-stage ranking. Part 1 grounding — a correct-looking answer is not enough. Part 12 Hybrid — multiple retrieval settings need comparison on fixed data. 1. Learning Objectives Build a small but useful golden dataset for RAG. Distinguish retrieval metrics from answer metrics . Use RAGAS and DeepEval in the right roles. Avoid the common traps in judge-model-based evaluation. 2. ํ•ต์‹ฌ ์š”์•ฝ An evaluation set for RAG is ...
Read more »

"ML Foundations (7/9) — Deep Learning Training: Optimizers, Regularization, Initialization"

← 6/9 Neural Networks ๐Ÿ“š Series Index 8/9 Deep Learning Architec… → Building an MLP, as we did in Part 6, is not the same as getting it to train well . The right optimizer, regularization, initialization, and learning rate — when these line up, deep networks converge. When they don't, the network refuses to learn at all. This part is the catalog of those crafts, with formulas, paper citations, and code in one place. 0. Learning Objectives Compare and write the update rules for SGD, Momentum, Nesterov, and Adam. Explain how Dropout, BatchNorm, and LayerNorm work and where they belong in a model. Derive the variance formulas for Xavier and He initialization and match them to activations. Implement step, cosine, and warmup learning-rate schedules in PyTorch. Explain why gradient clipping is effectively required for RNNs and Transformers. Diagnose the most common training failures (NaN, plateau, overfitting) and apply first-line fixes. 1. ํ•ต์‹ฌ ์š”์•ฝ SGD : \(w \leftarro...
Read more »

AI Agents I Built (5/7) — Building an Automated Blogger API Publishing System

← 4/7 My Life Organizer Agent ๐Ÿ“š Series Index 6/7 Dual → Designing a Markdown-Based Auto-Publish Pipeline with Blogger API v3 + OAuth 2.0 Key Summary Blogger API v3 + OAuth 2.0 enables a pipeline that converts Markdown files into publishable HTML posts blogger_publish.py handles the full flow — frontmatter parsing → body cleanup → HTML conversion → CSS injection → API post — in a single script For bulk publishing, quota management (delay, retry, TOC suppression) is the critical design concern Platform Selection Criteria API access is mandatory for AI agent-driven content publishing. A comparison of platforms: Platform API Cost AdSense Decision WordPress.com REST API Paid plan required Paid only Cost overhead Ghost REST API Self-hosted or paid Manual setup Ops overhead Blogger v3 API Free Native integration Adopted Blogger is free, exposes a REST API, and integrates natively with AdSense. For personal blog automation, this combination i...
Read more »

"ML Foundations (6/9) — Neural Networks: From Perceptron to MLP"

์ด๋ฏธ์ง€
← 5/9 Regularization & Model… ๐Ÿ“š Series Index 7/9 Deep Learning Training → A neural network is logistic regression from Part 4 stacked into layers. Each layer passes through a new nonlinear activation, expanding the family of functions the network can represent. In this part we put how a neural network learns onto one page. Perceptrons, XOR, MLPs, backpropagation, activation functions, and the universal approximation theorem all meet here. 0. Learning Objectives Write the perceptron's learning rule and explain why it cannot solve XOR. Express an MLP's forward pass as one line of matrix products. Derive backpropagation by hand on a two-layer MLP. Compare the definitions, derivatives, and trade-offs of sigmoid, tanh, ReLU, and GELU. State the universal approximation theorem (Cybenko 1989 / Hornik 1991) precisely. Train an MLP end-to-end in PyTorch. 1. ํ•ต์‹ฌ ์š”์•ฝ Perceptron (1958): \(\hat{y} = \mathrm{step}(w^\top x + b)\). Converges only on linearly separabl...
Read more »