Integrations

Hermes connects to multiple services through gateways, API bridges, and direct tool access. Here’s what we’ve wired up and the gotchas we’ve hit.

Discord

Hermes communicates via a Discord bot through the messaging gateway. The gateway must be running for Discord messages to flow in or out.

Setup:

DISCORD_BOT_TOKEN in ~/.hermes/.env
Gateway process runs the Discord bridge alongside the API server
Messages from Discord channels trigger agent sessions; agent responses are delivered back to the originating channel

Key pitfall — Dev profile messaging: The dev profile was initially created without the messaging toolset, meaning it couldn’t use send_message to reach Discord. Fix: add messaging to the profile’s enabled_toolsets in config.yaml. Direct access is cleaner than relaying through another agent — eliminates one hop.

Delivery targets:

discord — Delivers to the home channel
discord:#channel-name — Delivers to a specific channel
discord:chat_id:thread_id — Delivers to a specific thread/topic

Firecrawl (Self-Hosted)

A self-hosted Firecrawl instance runs on a different server for web scraping, crawling, and content extraction.

Core features (Scrape, Crawl, Map, Search) works without any LLM — they handle HTTP fetching, HTML parsing, and content extraction deterministically.

Extract endpoint (/v1/extract) requires an LLM for structured data extraction using JSON schema. It reads OPENAI_API_KEY and OPENAI_BASE_URL environment variables. We wired LM Studio as the LLM backend:

Set OPENAI_BASE_URL to LM Studio’s OpenAI-compatible endpoint (http://host:port/v1)
Set OPENAI_API_KEY to any value (LM Studio ignores it)
This enables structured extraction without a paid cloud API key

Forgejo (Self-Hosted Git)

Internal Git server used for:

Hermes config backups — Cron job pushes configs, profiles, skills, and cron state to Forgejo weekly
AI wiki repository — Knowledge base repo in the Agents organization
Self-hosted alternative to GitHub for internal/sensitive repos

Email (Himalaya)

Terminal-based IMAP/SMTP email via the himalaya CLI. We built a skill covering:

Outlook.com/Hotmail.com authentication (3 methods: OAuth2, App Password, Basic Auth)
Server settings and token refresh patterns
Step-by-step setup guide in references/outlook-config.md

Hermes API Server

Hermes exposes an OpenAI-compatible API server on port 8642 (configurable). This allows:

Other agents (like Claude Code) to use Hermes as an LLM backend
Programmatic access to Hermes capabilities
API key authentication via API_SERVER_KEY in .env

NGINX Proxy (Playground)

The Playground profile manages a Hermes created site served via NGINX on an LXC container. The nginx-site-admin skill is the governing umbrella for all site operations, cron jobs, and backup workflows.