Core Concepts

Summary

LightVela has four core concepts. A Hermes Agent is the cloud-run conversational entity, with its own isolated runtime, memory that persists across conversations, and 24/7 availability. The model is the large language model it calls — either a Model Coding Plan (monthly), Model API (pay as you go), or a custom OpenAI-compatible endpoint. Channels are the messaging platforms you talk to it through; multiple channels can connect to the same agent and share one persona and memory. Skills are installable capability packs that extend what it can do. A full conversation flows like this: a channel receives your message, the agent decides whether to invoke a skill, it calls the model to generate a response, and the reply comes back through the same channel.

LightVela involves four core concepts: Hermes Agent, models, channels, and skills. This page explains their roles. For detailed configuration steps, see the corresponding configuration pages.

This page covers

• The relationship between the four concepts

• Characteristics of a Hermes Agent instance

• The positioning and configuration entry points of models, channels, and skills

• The internal flow of a complete conversation

Concept relationship

┌──────────────────────────────────────────────────────────────┐
│                         Hermes Agent                         │
│              (Agent running in the LightVela cloud)           │
└──────────────────────────────────────────────────────────────┘
      │                          │                        │
┌─────┘                          │                        └──────┐
▼                                ▼                               ▼
┌───────┐                   ┌────────┐                   ┌────────┐
│ Model │                   │Channel │                   │ Skill  │
└───────┘                   └────────┘                   └────────┘
 Model provider    Telegram/WhatsApp/Discord/Slack/WeChat      SkillHub

Hermes Agent

Hermes Agent is the AI Agent instance in LightVela. Creating a Hermes Agent is equivalent to starting an independent Agent instance in the cloud.

Hermes Agent has the following characteristics:

• Isolated runtime environment: Each Hermes Agent runs in a dedicated environment isolated from other instances

• Persistent memory: Conversation content is saved across sessions, and the Agent uses historical context when responding

• Always online: Runs 24/7 and does not depend on the user's local device

The current Launch plan supports creating up to 1 Hermes Agent instance.

Models

A model is the large language model (LLM) that drives Hermes Agent thinking and responses. LightVela does not train models itself. Instead, users select a calling method in the "Model" card of the console:

In addition, the Launch plan includes 100 trial credits, so you can chat out of the box without any configuration; after the trial credits run out, you can configure your own API Key.

For supported models, API Key acquisition, and model selection guidance, see Configure Models.

Channels

Channels are the messaging platform entry points where users chat with Hermes Agent. LightVela currently supports the following channels:

For WeChat, choose the channel and click "Authorize Access" to generate a QR code. Scan it with your phone to complete authorization.

Detailed access steps for each channel: Configure Channel - WeChat / Telegram / WhatsApp / Discord / Slack.

Skills

Skills are additional capability packages callable by Hermes Agent. They extend Hermes Agent beyond basic conversation for specific tasks such as organizing to-dos, generating daily reports, browsing the web, generating images, and more.

All skills come from SkillHub, an independent skill marketplace launched by Tencent. In SkillHub, you can browse listed skills and view their feature descriptions and examples.

Enter the skill name recorded on SkillHub in the "3. Skills" card of the console and click "Install Skill" to install it. You can also ask Hermes Agent directly in conversation to install or delete skills. See Configure Skills.

Interaction flow

A complete conversation flow works as follows:

1. The user sends a message to Hermes Agent through a channel (such as WeChat)

2. Hermes Agent determines whether a skill needs to be called based on the message

3. The model processes the context and generates a response

4. The response returns to the user through the original channel

Next steps

• Configuration Overview — Full console feature guide

• Quickstart — Getting started when no Hermes Agent has been created