About Herm
Herm is an open source coding agent that runs inside containers. It is written in Go, ships as a single binary, and starts one container per session that extends itself with development tools as you prompt. The central idea is that because the agent works in a sealed container, your host machine stays untouched, so the agent can act without constant approval prompts.
It is aimed at developers who want an autonomous terminal coding assistant without giving it direct access to their system. Herm supports several model providers: Anthropic, OpenAI, Google Gemini, xAI Grok and local models through Ollama. You add your API key through the command line on first run and can switch models at any time. The project is MIT licensed and publishes its system prompts openly. The comparison table on the site is candid about limits: it lists MCP support and IDE integration as not yet available, while showing them as present in Claude Code and OpenCode.
There is no pricing page and the software is free to install, so the only cost is whatever the chosen model provider charges for API use. Installation options are a Homebrew tap, building from source on GitHub, or a shell installer script piped from the project's own site.
The name refers to a hermit crab called Herman, a nod to the sealed, hermetic design.