FlyHermes logoFLYHERMESDeploy Now

FlyHermes Documentation

Deploy, connect, and operate your hosted Hermes agent.

A practical field guide for getting from blank workspace to useful AI operator: channels, Memory, Skills, scheduled work, tools, and production-safe defaults.

1. Create your workspace

Start from the hosted checkout, confirm your workspace, and choose the first channel where your agent should live.

2. Connect Telegram

Add your bot token or invite the managed bot into the right chat. FlyHermes keeps topic and group context intact for long-running work.

3. Add tools and keys

Connect only the services you need first: GitHub, Gmail, Google Workspace, Stripe, Vercel, browser automation, or custom APIs.

4. Give it durable memory

Save stable preferences, project conventions, and workflow rules so the agent improves instead of resetting every session.

Core concepts

The pieces that make FlyHermes different from a chatbot.

Hermes is useful because it keeps context, calls real tools, and can keep working after the first message. FlyHermes packages those pieces into a hosted runtime so setup friction does not block usage.

  • Memory stores durable user preferences, environment notes, and stable project conventions — not disposable task logs.
  • Skills are reusable workflow playbooks: exact commands, pitfalls, verification steps, and project-specific operating rules.
  • Tools let the agent act: browser, terminal, GitHub, email, calendars, files, web research, images, and custom MCP servers.
  • Scheduled work lets the agent run recurring checks, briefs, monitors, and reminders without waiting for a manual prompt.

Channels

Start in one surface, expand later

Telegram is the fastest first channel. Add Discord, Slack, email, webhooks, or the CLI when the workflow needs them.

Safety

Keep secrets out of chat

Store API keys in the workspace environment, request credentials with direct links, and never paste private tokens into public threads.

Operations

Verify real outcomes

For code, run tests and builds. For browser work, inspect live pages. For automations, check logs and delivery targets before calling it done.