TAI Labs
TAI Lightning Lesson · Free

Hermes
Agent,
hands-on

The agent that grows with you. Remembers. Schedules. Writes its own skills.
Session
Live · one sitting
You bring
A laptop · one recurring workflow you want to automate
Led by
Aki WijesundaraAki Wijesundara
Manu JayawardanaManu Jayawardana
TAI Labs
Overview

What we will cover. Setup at the end.

This session is about what Hermes is and how it learns. Setup stays at the end as a short starting path.

01

What Hermes is

An autonomous agent that compounds over time.

02

Learning loop

Memory, skills, recall, and user modelling.

03

Where it lives

CLI, messaging, cron, and voice from one agent.

04

Skills engine

SKILL.md playbooks that write and improve themselves.

05

Capabilities

Tools, MCP, delegation, and scheduled work.

06

OpenClaw

How the two projects make different bets.

07

Use cases

Recurring work that deserves muscle memory.

08

Where to start

One install path and one first skill.

TAI Labs
Definition

Not a chatbot wrapper. Not an IDE copilot.

Hermes Agent is an autonomous agent that gets more capable the longer it runs. Built by Nous Research. Works with any OpenAI-compatible endpoint.

01

It remembers

Persistent memory carries preferences, environment facts, and lessons across sessions.

02

It learns procedures

Successful workflows become SKILL.md playbooks that Hermes can reuse later.

03

It lives where you work

Talk to it from the CLI or messaging apps while it runs on a VPS, laptop, or serverless host.

The pattern: Hermes creates skills from experience, improves them during use, persists knowledge, and builds a deepening model of who you are across sessions.
TAI Labs
Core idea

The closed learning loop.

Hermes is organised around a learning loop, not around a single chat window. Read clockwise from 01.

01 · Do the work Hermes solves a task with tools, the same way any agent would in a single session. 02 · Persist knowledge Useful facts land in MEMORY.md and USER.md, ready for next session. 03 · Capture procedures A background review can write or patch a SKILL.md so the workflow is reusable. 04 · Reuse next time The next similar task loads skills and memory instead of rediscovering the approach. loops back to 01
The centre of the diagram: do, capture, reuse. The value compounds every time this loop closes, not every time you start a new chat.
TAI Labs
Architecture

One agent. Many surfaces.

The same Hermes runtime powers the CLI, messaging platforms, scheduled jobs, and voice. Front ends change. The agent and its memory stay.

CLI

Terminal-first

Core conversation loop, tools, and prompt assembly. This is the primary way to build and improve skills.

Messaging

Chat platforms

Telegram, Slack, Discord, WhatsApp, and more. Same agent, reached from wherever you already are.

Cron

Scheduled work

Scheduled agent jobs delivered to any connected platform. Recurring work runs itself.

Voice

Real-time

Real-time voice in the CLI and supported chat apps. Same memory. Same skills.

The hub-and-spoke: the memory + skills library sits at the centre. Every surface calls into the same brain.
TAI Labs
Memory

Facts, procedures, and recall. Priced by cost.

Hermes splits knowledge by cost. Small facts stay in every prompt. Longer procedures load only when needed. Full history is searchable on demand.

Always on · MEMORY.md and USER.md

Small durable facts stay in every session prompt. They live in ~/.hermes/memories/ and are injected at session start.

↓ next layer is larger, so it loads only when relevant

On demand · Skills (SKILL.md)

Longer playbooks live under ~/.hermes/skills/. Hermes loads a skill only when the task matches its description.

↓ if neither layer has the detail, search history

Searchable history · Session search (FTS5)

Past CLI and messaging sessions stay in SQLite. Hermes can retrieve real messages without putting the full history into every prompt.

↔ optional add-on, not a replacement

Optional external providers

Honcho, Mem0, and similar plugins can deepen user modelling beside this stack. They do not replace the built-in files.

The rule of the session

The value is not in the first chat. It's in what compounds after.

Every stateless agent restarts from zero. Hermes doesn't. Memory keeps facts across sessions. Skills keep procedures across sessions. That's the whole reason to run an agent that has a home directory instead of a chat tab.

TAI Labs
Skills

Skills are procedural memory.

A skill is an on-demand instruction document. Hermes reads short descriptions for free, then loads the full SKILL.md body only when the task needs it.

01

What a skill contains

Each skill has a name, a short description, a step-by-step procedure, and optional scripts or reference files.

02

Progressive disclosure

Adding skills does not bloat every request, because the full playbook is loaded only when relevant.

03

Open standard

Skills follow agentskills.io, so they are portable across tools and community hubs.

04

Familiar pattern

If you already write SKILL.md files for Claude Code, you already know the muscle Hermes uses.

TAI Labs
Self-improvement

Skills that write themselves.

Hermes can solve a problem once, save the working approach as a skill, and reuse that playbook the next time a similar task appears.

SolveTrial and error with tools
→
ReviewBackground pass extracts the pattern
→
Write skillCreate or patch SKILL.md
→
ReuseNext similar task loads it
Controls

You stay in the loop

Skills become slash commands. You can author with /learn. Setting skills.write_approval to true stages writes until you approve them.

Curator

The library stays clean

A background maintenance pass archives unused agent-created skills so the library does not fill with near-duplicates. Hub skills stay untouched.

TAI Labs
Capabilities

What Hermes can do day to day.

Beyond memory and skills, Hermes ships as a full agent runtime with tools, scheduling, and parallel work.

Ask
→
Use tools
→
Act
→
Schedule / delegate
→
Deliver
01

Tools and MCP

Built-in terminal, browser, and file tools. MCP servers extend the set safely.

02

Scheduled work

Cron runs agent jobs on a schedule and can deliver results to any connected platform.

03

Parallel subagents

delegate_task spawns isolated children, up to three in parallel by default, then returns summaries.

04

Runs anywhere

Local machine, Docker, SSH, or serverless hosts such as Daytona and Modal when idle cost matters.

05

Model choice

Nous Portal, OpenRouter, OpenAI, Ollama, or any OpenAI-compatible endpoint.

06

Research path

Batch runs and trajectory export support training workflows from the same agent runtime.

TAI Labs
Comparison

Hermes vs OpenClaw.

Both are self-hosted agent systems. They make different bets about what should improve over time.

Dimension
Hermes
OpenClaw
Core bet
One agent that improves through memory and skills
An organization of agents under a control plane
Shape
Python CLI first, with an optional messaging gateway
A central gateway daemon with hub-and-spoke routing
Multi-agent
Temporary subagents through delegate_task
Persistent named agents bound to channels
Skills
Agent-written and hub skills via agentskills.io
Large built-in set plus marketplace plugins
Which one, when: choose Hermes when automation should get better over time. Choose OpenClaw when the main problem is orchestrating many agents across channels.
TAI Labs
Use cases

Work that deserves muscle memory.

Hermes fits recurring workflows that happen often enough to become skills and improve with use.

Recurring loops
  • Daily reports and status digests
  • Content and research pipelines
  • Data collection and monitoring jobs
  • Inbox-to-summary routing
Consultancy examples
  • A nightly client status digest
  • A repeatable data-cleaning skill
  • A scheduled compliance check
  • Client delivery work that should become reusable
The tell: if a workflow happens more than a few times a month and you keep re-explaining it to the model, it's a candidate skill. That's the shape.
TAI Labs
Where to start

Stand one up. Then teach it one workflow.

Install is intentionally short. The compounding value starts after Hermes has a model and one real recurring task.

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
01

Choose a model

Run hermes model, or hermes setup --portal for Nous Portal OAuth and tool gateway access.

02

Talk to it

Start in the CLI, then optionally connect Telegram or another platform with hermes gateway setup.

03

One skill

Write or let Hermes write one SKILL.md for a workflow that already repeats every week.

Where things live: secrets belong in ~/.hermes/.env. Behaviour settings belong in config.yaml. Docs: hermes-agent.nousresearch.com/docs
TAI Labs
The habit to take with you

Remember. Schedule. Write skills. Compound.

Hermes is valuable because the loop keeps running after the first chat. Every session teaches the next one. Install, connect a model, capture one recurring workflow as a skill. That's the whole first move.

Next

Agentic AI Engineering Bootcamp

Nine weeks. Build the memory, skills, and evals that make self-hosted agents ship. Certificate for engineers and AI PMs.

Where this leads
Now

Questions in chat

Post one recurring workflow you'd hand to Hermes. Aki or Manu will name the first skill you'd write.

Open floor
Later

Install and one skill

Run the install line, pick a model, then teach it one workflow this week. Email us the skill. We reply with a short audio review.

The receipt
01 / 14