1. What OpenHermit Is

OpenHermit is a platform for running AI agents. You give it a model (Claude, GPT, Gemini, …) and connect it to the places you already talk — Telegram, Discord, Slack, the web, the command line — and your agent then lives in the cloud as a long-running service. You send it a message, it replies. You ask it to run a command, it runs one. You tell it to file a report every weekday morning, it files one every weekday morning.

The main difference from local agents like Claude Code or Cursor is that OpenHermit is not tied to a single machine. It has a persistent identity in the cloud — memories, sessions, skills, configuration, scheduled jobs — and any entry point you use sees the same agent.

This manual is for people who already have an OpenHermit agent. Setting up a new OpenHermit instance is a different topic and lives in the operator docs.

What It Solves

If you have used Claude Code or similar local tools, you have probably hit some of these:

  • You taught the agent your preferences on your work laptop, then on your home laptop it remembered nothing.
  • You wanted the agent to run a job overnight, but your laptop sleeps.
  • You wanted a friend to use your agent, but its memory and history are tied to your machine.
  • You wanted to reach the agent from Telegram or Slack instead of a terminal, and configuring a bridge turned into a side project.

OpenHermit’s design starts from those problems. The agent has durable cloud-side state, multiple entry points reach the same agent, multiple people can share an agent with proper isolation, and the agent keeps working while you are offline.

What You Can Use It For

  • Personal assistant — remembers your preferences and tasks, reachable from phone or laptop, sends you scheduled reminders or digests.
  • Shared team assistant — a few people share one agent: a shared knowledge base, daily standup summaries, customer support triage.
  • Vertical chatbot — drop it into a Telegram or Discord group to serve a community around a specific topic.
  • Background worker — scheduled jobs to pull data, scrape, write daily reports, check on services.
  • MCP-driven automation — connect GitHub, Linear, Slack, or your own internal APIs as MCP tools, then operate them in natural language.

What It Is Not Good At

  • Fully offline work. OpenHermit assumes a gateway is reachable. For air-gapped environments, a local-only tool is a better fit.
  • Massive end-user SaaS. You can build on it, but rate limits, billing, abuse handling — those are still your job.
  • Replacing an IDE. It can write code, edit files, and run commands, but it is not VS Code. Code editing inside an IDE is still smoother.
  • Acting as an unrestricted root. By default it can touch its workspace and run sandboxed commands; reaching the outside world needs MCP servers or secrets that you provision. It is not free to do anything.

Two Words You Will See Throughout

Even if you are the only person using your agent, two terms are worth holding onto:

  • Agent — the “it” that replies to you. An OpenHermit instance can host multiple agents (say, one named work and one named home) that share nothing — memories, configuration, and access lists are all independent.
  • You as owner — you created the agent, so you are its owner. Owners can change settings, add members, see every session, and use every tool. Anyone else you invite gets a more restricted role: user or guest. The details are in Chapter 5 · Users and Identity.

Where to Go Next

  • Want to get something running first: Chapter 2 · Quickstart.
  • Want to read concepts before touching anything: Chapter 3 · Core Concepts.
  • Already using it, looking for a specific recipe: open the relevant chapter and scroll to its How-to recipes section.