Hermes Agent Mattermost Setup for Self-Hosted Teams

Why run the assistant on Mattermost?

Because Mattermost is self-hosted, running the assistant there keeps conversations on infrastructure you control while still giving the team a familiar chat interface.

For teams that already rejected cloud chat for compliance reasons, Mattermost is the option that lets them adopt an assistant without reopening that decision.

How is the bot created?

A system administrator enables bot accounts on the server, then creates a bot account for the assistant. The bot gets its own identity and token, separate from any human login.

• Bot accounts must be enabled server-side first; there is no per-team override.
• Use a dedicated bot rather than a shared human account, so revocation is clean.
• Store the token once and never leave it in a channel or ticket.
• Give the bot a recognisable display name so colleagues know what they are talking to.

When does the Mattermost bot answer?

Direct messages are always answered, and each direct message keeps its own session. In public and private channels the assistant answers only when it is mentioned.

• Direct messages: always answered, one session per conversation.
• Channels: mention required, otherwise the assistant stays quiet.
• Threads: each thread keeps its own context namespace.
• Shared channels: session history is isolated per user by default.

Should the assistant reply in threads or flat?

Threads are usually the better default in active channels, because replies stay attached to the question instead of interleaving with unrelated conversation.

Flat replies keep a channel readable when traffic is low, but in a busy channel they make the assistant look noisy. Changing the reply mode later is a configuration change, so start with whichever your team already expects.

How are shared channels handled?

Each user in a shared channel gets their own session by default, so colleagues do not accidentally share one transcript or one growing context window.

Sharing one session across a channel is possible but changes the economics and the experience: everyone pays for the context growth, and one person's long task can interrupt another person's follow-up.