Skip to main content
Hermes is the primary conversational agent in ClawHQ. While OpenClaw agents (Mike, Scout, Codex, etc.) are specialized execution engines triggered by commands, Hermes is the always-on brain — it holds context across every conversation, remembers your team, your goals, and past interactions, and gets smarter the more you use it. Hermes is built on the Nous Research Hermes model series and runs as its own service alongside OpenClaw.

​
Hermes vs OpenClaw agents

HermesOpenClaw agents
PurposeConversational AI, memory, NL interfaceTask execution, automation
MemoryPersistent across all sessionsPer-conversation (configurable)
TriggersAlways listening in connected channelsKeyword or command triggers
ModelSingle configurable model (HERMES_MODEL)Model Router — right model per task
Best for”What’s going on?”, complex Q&A, long-running context”Research X”, “Code review Y”, scheduled tasks
Think of it this way: Hermes is your AI colleague you talk to every day. OpenClaw agents are specialists you bring in for specific jobs.

​
Setup

Hermes starts automatically with docker compose up. The only required config is an API key for your chosen model provider.

​
Environment variables

Add these to your .env file: 
# Required — model for Hermes to use
HERMES_MODEL=anthropic/claude-sonnet-4-6

# Required — at least one channel token
HERMES_DISCORD_BOT_TOKEN=your-discord-bot-token

# Optional — additional channels
HERMES_TELEGRAM_BOT_TOKEN=your-telegram-bot-token
HERMES_SLACK_BOT_TOKEN=your-slack-bot-token
HERMES_SLACK_APP_TOKEN=your-slack-app-token

# Optional — port (default: 4300)
HERMES_PORT=4300
Hermes needs its own bot tokens — separate from the OpenClaw bot. Create a second Discord application at discord.com/developers for Hermes. Having two bots lets you address them independently in Discord (@Hermes vs @Mike).

​
Supported models

Hermes uses a single model configured via HERMES_MODEL. Use any model string in provider/model format:
ProviderExample value
Anthropicanthropic/claude-sonnet-4-6
OpenAIopenai/gpt-4o
OpenRouteropenrouter/nousresearch/hermes-3-llama-3.1-70b
Groqgroq/llama-3.3-70b-versatile
Local (Ollama)ollama/llama3.2
The recommended default is anthropic/claude-sonnet-4-6 — fast, capable, and cost-effective for conversational use.

​
Talking to Hermes

Once configured and your bot is invited to your Discord server: 
@Hermes what's on my plate today?

@Hermes summarize what we discussed about the launch last week

@Hermes remember that the Q2 target is $50k MRR
Hermes automatically retrieves relevant context from memory before every response — you don’t need to re-explain things you’ve already told it.

​
Memory

Hermes stores conversation history in SQLite and uses semantic search to retrieve relevant past context before each reply. This happens automatically. You can interact with memory directly:
CommandWhat it does
@Hermes remember [fact]Stores something explicitly
@Hermes forget [topic]Removes memory entries about a topic
@Hermes what do you know about [topic]Shows what Hermes has stored
Memory is per-deployment — shared across all team members using the same ClawHQ instance.

​
Skills

Hermes can load skills from ~/.openclaw/skills/. Skills extend what Hermes can do — web search, calendar access, file operations, custom integrations. To see loaded skills: 
@Hermes what skills do you have?
Skills are installed via the OpenClaw CLI or by dropping YAML files into the skills directory. See the Skills reference for the schema.

​
Dashboard

Hermes activity is visible in the ClawHQ dashboard under Team → Hermes. From there you can:
  • View conversation history
  • See which skills are loaded
  • Adjust the model
  • Clear or export memory

​
Next steps

Connect Discord

Set up a dedicated Hermes bot in Discord

Model Router

How OpenClaw agents route tasks to the right model

First Agent

Configure a specialized OpenClaw agent

Environment Variables

Full list of Hermes config options