Choose a framework

A framework is the agent software that runs inside your agent's machine. Every ready-made template is a real agent framework, pre-configured and one image pull from live. Pick one, name it, deploy. Each guide covers the image, how it works, its environment variables, and how to talk to it.

FrameworkTemplate IDRuntimeBest forLLM key
Claude Codeclaude_codeNode.js (Claude Agent SDK)A coding agent with real bash and file tools in a persistent workspaceYour ANTHROPIC_API_KEY (required)
CodexcodexNode.js (Codex CLI)OpenAI's coding agent with shell and file tools in a persistent workspace, plus the interactive TUI in the terminalOptional (RunBerth LLM by default; your OpenAI key or ChatGPT sign-in)
OpenClawopenclawNode.jsA general autonomous assistant: 100+ skills, channels, web browsingOptional (RunBerth LLM by default)
HermeshermesPythonA self-improving agent with persistent memory and its own full dashboardOptional (RunBerth LLM by default)
ZeroClawzeroclawRust (single binary)Ultra-light agents: <5 MB RAM, sub-second start, good for running manyOptional (RunBerth LLM by default)
DeepSeek HarnessdshNode.js (plugin runtime)DeepSeek's agent harness with a full web workbench and session replayNone needed (RunBerth LLM by default)

Not sure? Start with OpenClaw for a general assistant, or Claude Code for a coding agent. You can always create another agent from a different template.

Deploying a template

In the dashboard: New agent, pick the template, name it. From the terminal:

runberth templates                       # list what's available
runberth create my-agent --template openclaw

Worth knowing

  • Claude Code and Codex are chat-only: neither has a web UI of its own, so you talk to it from the dashboard chat, the CLI, the API, or a connected channel.
  • DeepSeek Harness is version-pinned: upstream is a fast-moving developer preview, so RunBerth upgrades the image deliberately after testing.

None of these fit?

Any Docker image that speaks a two-endpoint contract becomes a full RunBerth agent: CrewAI, LangGraph, AutoGen, or your own code. See Deploy any container.